Skip to main content
The Cochrane Database of Systematic Reviews logoLink to The Cochrane Database of Systematic Reviews
. 2012 Mar 14;2012(3):CD007176. doi: 10.1002/14651858.CD007176.pub2

Antioxidant supplements for prevention of mortality in healthy participants and patients with various diseases

Goran Bjelakovic 1,2,, Dimitrinka Nikolova 2, Lise Lotte Gluud 3, Rosa G Simonetti 4, Christian Gluud 2
Editor: Cochrane Hepato‐Biliary Group
PMCID: PMC8407395  PMID: 22419320

Abstract

Background

Our systematic review has demonstrated that antioxidant supplements may increase mortality. We have now updated this review.

Objectives

To assess the beneficial and harmful effects of antioxidant supplements for prevention of mortality in adults.

Search methods

We searched The Cochrane Library, MEDLINE, EMBASE, LILACS, the Science Citation Index Expanded, and Conference Proceedings Citation Index‐Science to February 2011. We scanned bibliographies of relevant publications and asked pharmaceutical companies for additional trials.

Selection criteria

We included all primary and secondary prevention randomised clinical trials on antioxidant supplements (beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium) versus placebo or no intervention.

Data collection and analysis

Three authors extracted data. Random‐effects and fixed‐effect model meta‐analyses were conducted. Risk of bias was considered in order to minimise the risk of systematic errors. Trial sequential analyses were conducted to minimise the risk of random errors. Random‐effects model meta‐regression analyses were performed to assess sources of intertrial heterogeneity.

Main results

Seventy‐eight randomised trials with 296,707 participants were included. Fifty‐six trials including 244,056 participants had low risk of bias. Twenty‐six trials included 215,900 healthy participants. Fifty‐two trials included 80,807 participants with various diseases in a stable phase. The mean age was 63 years (range 18 to 103 years). The mean proportion of women was 46%. Of the 78 trials, 46 used the parallel‐group design, 30 the factorial design, and 2 the cross‐over design. All antioxidants were administered orally, either alone or in combination with vitamins, minerals, or other interventions. The duration of supplementation varied from 28 days to 12 years (mean duration 3 years; median duration 2 years). Overall, the antioxidant supplements had no significant effect on mortality in a random‐effects model meta‐analysis (21,484 dead/183,749 (11.7%) versus 11,479 dead/112,958 (10.2%); 78 trials, relative risk (RR) 1.02, 95% confidence interval (CI) 0.98 to 1.05) but significantly increased mortality in a fixed‐effect model (RR 1.03, 95% CI 1.01 to 1.05). Heterogeneity was low with an I2‐ of 12%. In meta‐regression analysis, the risk of bias and type of antioxidant supplement were the only significant predictors of intertrial heterogeneity. Meta‐regression analysis did not find a significant difference in the estimated intervention effect in the primary prevention and the secondary prevention trials. In the 56 trials with a low risk of bias, the antioxidant supplements significantly increased mortality (18,833 dead/146,320 (12.9%) versus 10,320 dead/97,736 (10.6%); RR 1.04, 95% CI 1.01 to 1.07). This effect was confirmed by trial sequential analysis. Excluding factorial trials with potential confounding showed that 38 trials with low risk of bias demonstrated a significant increase in mortality (2822 dead/26,903 (10.5%) versus 2473 dead/26,052 (9.5%); RR 1.10, 95% CI 1.05 to 1.15). In trials with low risk of bias, beta‐carotene (13,202 dead/96,003 (13.8%) versus 8556 dead/77,003 (11.1%); 26 trials, RR 1.05, 95% CI 1.01 to 1.09) and vitamin E (11,689 dead/97,523 (12.0%) versus 7561 dead/73,721 (10.3%); 46 trials, RR 1.03, 95% CI 1.00 to 1.05) significantly increased mortality, whereas vitamin A (3444 dead/24,596 (14.0%) versus 2249 dead/16,548 (13.6%); 12 trials, RR 1.07, 95% CI 0.97 to 1.18), vitamin C (3637 dead/36,659 (9.9%) versus 2717 dead/29,283 (9.3%); 29 trials, RR 1.02, 95% CI 0.98 to 1.07), and selenium (2670 dead/39,779 (6.7%) versus 1468 dead/22,961 (6.4%); 17 trials, RR 0.97, 95% CI 0.91 to 1.03) did not significantly affect mortality. In univariate meta‐regression analysis, the dose of vitamin A was significantly associated with increased mortality (RR 1.0006, 95% CI 1.0002 to 1.001, P = 0.002).

Authors' conclusions

We found no evidence to support antioxidant supplements for primary or secondary prevention. Beta‐carotene and vitamin E seem to increase mortality, and so may higher doses of vitamin A. Antioxidant supplements need to be considered as medicinal products and should undergo sufficient evaluation before marketing.

Plain language summary

Antioxidant supplements for prevention of mortality in healthy participants and patients with various diseases

Previous research on animal and physiological models suggests that antioxidant supplements have beneficial effects that may prolong life. Some observational studies also suggest that antioxidant supplements may prolong life, whereas other observational studies demonstrate neutral or harmful effects. Our Cochrane review from 2008 demonstrated that antioxidant supplements seem to increase mortality. This review is now updated.

The present systematic review included 78 randomised clinical trials. In total, 296,707 participants were randomised to antioxidant supplements (beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium) versus placebo or no intervention. Twenty‐six trials included 215,900 healthy participants. Fifty‐two trials included 80,807 participants with various diseases in a stable phase (including gastrointestinal, cardiovascular, neurological, ocular, dermatological, rheumatoid, renal, endocrinological, or unspecified diseases). A total of 21,484 of 183,749 participants (11.7%) randomised to antioxidant supplements and 11,479 of 112,958 participants (10.2%) randomised to placebo or no intervention died. The trials appeared to have enough statistical similarity that they could be combined. When all of the trials were combined, antioxidants may or may not have increased mortality depending on which statistical combination method was employed; the analysis that is typically used when similarity is present demonstrated that antioxidant use did slightly increase mortality (that is, the patients consuming the antioxidants were 1.03 times as likely to die as were the controls). When analyses were done to identify factors that were associated with this finding, the two factors identified were better methodology to prevent bias from being a factor in the trial (trials with ‘low risk of bias’) and the use of vitamin A. In fact, when the trials with low risks of bias were considered separately, the increased mortality was even more pronounced (1.04 times as likely to die as were the controls). The potential damage from vitamin A disappeared when only the low risks of bias trials were considered. The increased risk of mortality was associated with beta‐carotene and possibly vitamin E and vitamin A, but was not associated with the use of vitamin C or selenium. The current evidence does not support the use of antioxidant supplements in the general population or in patients with various diseases.

Background

Description of the condition

Oxidative stress may play a role in the pathogenesis of cancer and cardiovascular disease, the leading causes of death in middle‐ and high‐income countries (Sies 1985; Poulsen 1998; Halliwell 1999). Diet provides numerous vitamins and trace elements that are essential for good health. Several observational studies have shown a significant positive association between higher intake of fruits and vegetables and reduced risk of chronic diseases (Block 1992; Ames 1993; Willcox 2004).

Description of the intervention

However, exactly which specific dietary constituents of fruits and vegetables might be beneficial is not clear. Furthermore, causal inferences are hard to establish from observational studies. Antioxidants have attracted the most attention as promising preventive agents. Fruits and vegetables are sources of numerous micronutrients and some of these, including beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium, have antioxidant potential. Many people take antioxidant supplements believing that doing so will improve their health (Balluz 2000; Millen 2004; Radimer 2004; Nichter 2006).

How the intervention might work

Whether antioxidant supplements are beneficial or harmful is uncertain (Ovesen 1984; Herbert 1997; Caraballoso 2003; Vivekananthan 2003; Bjelakovic 2004; Stanner 2004; Berger 2005; Miller 2005; Thomas 2006). Free radicals may play dual roles (Bjelakovic 2007b). Free radicals in moderate concentrations are essential mediators of reactions by which unwanted cells are deleted from the body. However, excessive antioxidants might interfere with some essential defensive mechanisms of our organism. In our previous Cochrane review we observed that antioxidant supplements seemed to increase mortality by about 4% (Bjelakovic 2007a; Bjelakovic 2008).

Why it is important to do this review

Since our previous Cochrane reviews (Bjelakovic 2004; Bjelakovic 2007a; Bjelakovic 2008), the effect of antioxidant supplements on mortality has been assessed in several large trials investigating the primary and secondary prevention of diseases (WACS 2007Low; PHS 2008Low; SELECT 2009Low). The present review is an update of the former reviews.

Objectives

Our aim was to assess the effect of antioxidant supplements (beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium) on overall mortality in primary or secondary prevention randomised clinical trials.

Methods

Criteria for considering studies for this review

Types of studies

All primary and secondary prevention randomised clinical trials, irrespective of trial design, blinding, publication status, publication year, or language, comparing antioxidant supplements (beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium) versus placebo or no intervention.

Types of participants

Adult participants (age 18 years or over) who were:

  • healthy participants or were recruited among the general population (primary prevention);

  • diagnosed with a specific disease that was in a stable phase (secondary prevention).

We excluded tertiary prevention trials, that is randomised trials in which antioxidant supplements were used to treat a specific disease or nutritional defects, like trials involving patients with acute, infectious, or malignant diseases (except non‐melanoma skin cancer).

We excluded trials including children and pregnant women since they may be in need of certain antioxidant supplements.

Types of interventions

We considered for inclusion trials that compared antioxidant supplements (that is, beta‐carotene, vitamin A, vitamin C, vitamin E, and selenium) at any dose, duration of treatment, and route of administration versus placebo or no intervention.

The antioxidants could have been administered:

  • separately or in any combination among themselves; or

  • in combination with other vitamins; or

  • in combination with trace elements without antioxidant function.

Concomitant interventions were allowed if used equally in both intervention arms of the trial.

Types of outcome measures

Our sole outcome measure was all‐cause mortality.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library (Issue 1, 2011), MEDLINE (1966 to February 2011), EMBASE (Excerpta Medica Database) (1985 to February 2011), and the Science Citation Index Expanded (1945 to February 2011) (Royle 2003).

We also searched the World Health Organization International Clinical Trials Registry Platform (ICTRP 2011) for ongoing trials. All search strategies are given in Appendix 1.

Searching other resources

We scanned bibliographies of relevant publications for additional trials.

We sent letters by post or e‐mail to major manufacturers of antioxidant supplements, that is CBH Qingdao Co, Ltd in China, DSM Nutritional Products in Switzerland, CVC4Health in USA, and BASF in Germany, asking for unpublished randomised trials.

Data collection and analysis

The present review was based on our protocol on antioxidant supplements for preventing gastrointestinal cancers (Bjelakovic 2003), adopted to assess overall mortality. An abbreviated and a full version of the review have previously been published (Bjelakovic 2007a; Bjelakovic 2008).

Selection of studies

Two of the three authors (GB and DN or RGS) independently assessed trial eligibility without blinding of the study authors. We listed excluded trials with the reasons for exclusion. Disagreement was resolved by discussion or in consultation with LLG or CG. We contacted authors of the trials for missing information. An adapted PRISMA flow‐diagram of study selection is included in the review (Moher 2009).

Data extraction and management

Participant characteristics, diagnosis, and interventions

From each trial we recorded first author; country of origin; country income category (low, middle, high) (World Bank 2011); number of participants; characteristics of participants: age range (mean or median) and sex ratio; participation rate; dropout rate; trial design (parallel, factorial, or cross‐over); type of antioxidant; dose; duration of supplementation; duration of follow‐up (that is, duration of intervention plus post‐intervention follow‐up); and co‐interventions.

Trial characteristics

We recorded the date, location, sponsor of the trial (known or unknown and type of sponsor), as well as publication status.

Assessment of risk of bias in included studies

Due to the risk of overestimation of beneficial intervention effects in randomised trials with unclear or inadequate methodological quality (Schulz 1995; Moher 1998; Kjaergard 2001; Wood 2008), we assessed the influence of the risk of bias on our results. We used the following domains: allocation sequence generation, allocation concealment, blinding, complete outcome data reporting, selective outcome reporting, and other apparent biases (Higgins 2011). The following definitions were used.

Allocation sequence generation 

‐ Low risk of bias: sequence generation was achieved using computer generated random numbers or a random number table, or similar. 
 ‐ Uncertain risk of bias: the trial was described as randomised but the method of sequence generation was not specified. 
 ‐ High risk of bias: the sequence generation method was not, or may not be, random. Quasi‐randomised studies, those using dates, names, or admittance numbers in order to allocate patients, were inadequate and were excluded for the assessment of benefits but not for assessing harms.

Allocation concealment

‐ Low risk of bias: allocation was controlled by a central and independent randomisation unit, sequentially numbered, opaque and sealed envelopes, or similar, so that intervention allocations could not have been foreseen in advance of or during enrolment. 
 ‐ Uncertain risk of bias: the trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of or during enrolment. 
 ‐ High risk of bias: if the allocation sequence was known to the investigators who assigned participants or if the study was quasi‐randomised. Quasi‐randomised studies were excluded for the assessment of benefits but not for assessing harms.

Blinding

‐ Low risk of bias: the trial was described as blinded, the parties that were blinded and the method of blinding were described, so that knowledge of allocation was adequately prevented during the trial. 
 ‐ Uncertain risk of bias: the trial was described as blind but the method of blinding was not described, so that knowledge of allocation was possible during the trial. 
 ‐ High risk of bias: the trial was not blinded, so that the allocation was known during the trial.

Incomplete outcome data

‐ Low risk of bias: the numbers and reasons for dropouts and withdrawals in all intervention groups were described, or if it was specified that there were no dropouts or withdrawals. 
 ‐ Uncertain risk of bias: the report gave the impression that there had been no dropouts or withdrawals but this was not specifically stated. 
 ‐ High risk of bias: the number of or reasons for dropouts and withdrawals were not described.

Selective outcome reporting

‐ Low risk of bias: pre‐defined or clinically relevant and reasonably expected outcomes were reported on. 
 ‐ Uncertain risk of bias: not all pre‐defined or clinically relevant and reasonably expected outcomes were reported on, or were not reported on fully, or it was unclear whether data on these outcomes were recorded or not. 
 ‐ High risk of bias: one or more clinically relevant and reasonably expected outcomes were not reported on; data on these outcomes were likely to have been recorded.

Other bias

‐ Low risk of bias: the trial appeared to be free of other components that could put it at risk of bias.  
 ‐ Uncertain risk of bias: the trial may or may not have been free of other components that could put it at risk of bias. 
 ‐ High risk of bias: there were other factors in the trial that could put it at risk of bias, eg, for‐profit involvement, authors have conducted other trials on the same topic, etc.

Trials with adequate assessments in all of the above mentioned risk of bias domains were considered as having low risk of bias.

Dealing with missing data

We tried to obtain relevant missing data from authors of the included trials. We performed evaluation of important numerical data such as screened, eligible, and randomised participants as well as intention‐to‐treat (ITT) and per protocol (PP) populations. We investigated attrition (that is, dropouts, losses to follow‐up, and withdrawals).

Dealing with duplicate publications

In the case of duplicate publications and companion papers of a primary trial, we tried to maximise the yield of information by simultaneous evaluation of all available data. In cases of doubt, the publication that reported the longest follow‐up (usually the most recent version) obtained priority.

Assessment of heterogeneity

We identified heterogeneity by visual inspection of the forest plots by using a standard χ2‐test and a significance level of α = 0.1. In view of the low power of such tests, we also examined heterogeneity with the I2 statistic (Higgins 2002), where I2 values of 50% and more indicate a substantial level of heterogeneity (Higgins 2011). When heterogeneity was found, we attempted to determine potential reasons for it by examining individual trial characteristics and those of subgroups of the main body of evidence.

Assessment of reporting biases

Funnel plots were used to assess the potential existence of bias (Lau 2006). There are a number of explanations for the asymmetry of a funnel plot, including true heterogeneity of effect with respect to trial size, poor methodological design and hence bias of small trials, and publication bias. We performed adjusted rank correlation (Begg 1994) and a regression asymmetry test for detection of bias (Egger 1997); a P < 0.10 was considered significant.

Data synthesis

We performed the meta‐analyses according to the recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). For the statistical analyses, we used RevMan (RevMan 2008), STATA 8.2 (STATA Corp, College Station, Texas), Sigma Stat 3.0 (SPSS Inc, Chicago, Ill), and Trial Sequential Analysis version 0.8 (TSA 2008; Thorlund 2011).

We analysed the data with both random‐effects (DerSimonian 1986) and fixed‐effect (DeMets 1987) model meta‐analyses. We presented the results of the random‐effects model analyses. When statistically significant results were obtained in either the random‐ or fixed‐effect model, we presented both analyses. Results were presented as the relative risk (RR) with 95% confidence interval (CI).

Random‐effects model meta‐regression analyses were performed to assess potential covariates that could predict intertrial heterogeneity, that is, the covariates that were statistically associated with the estimated intervention effects. The included covariates were risk of bias (low or high), type and dose of supplement, duration of prevention, and type of prevention (primary or secondary). We also performed subgroup analyses comparing the primary and secondary prevention trials. Furthermore, we performed sensitivity analyses excluding trials using small doses of antioxidant supplements in both the experimental and control study groups. The exclusion of trials using small doses of antioxidant supplements in both the experimental and control study groups was based on the fact that addition of, for example, a vitamin pill could be a confounder.

The influence of trials with zero events in the treatment or control group was assessed by re‐calculating the random‐effects model meta‐analyses with 0.5, 0.01, and 0.001 continuity corrections (Sweeting 2004; Bradburn 2007). We also performed additional meta‐analyses including one large hypothetical trial with one event in the treatment and control group and a sample size corresponding to the total number of participants in the zero events trials.

All our analyses followed the intention‐to‐treat principle. We accounted all of the participants for each trial and performed the analyses irrespective of how the original trialists had analysed the data. Participants lost to follow‐up were considered survivors. We also performed sensitivity analysis in which all participants lost to follow‐up were considered dead. For trials with a factorial design, we based our results on 'at the margins' analysis, comparing all groups that received antioxidant supplements with groups that did not receive antioxidant supplements (McAlister 2003). This entails a risk of interaction between the antioxidant and the other intervention(s) assessed, whether significant or not in the individual trial. Due to the risk of confounding in factorial trials assessing other interventions, we conducted post hoc sensitivity analysis including only factorial trial data, which could not be affected by such confounding (that is, 'inside the table' analysis) (McAlister 2003). In the trials with parallel group design with more than two arms and additional therapy, we compared only antioxidant intervention with placebo or no intervention. For cross‐over trials we included only data from the first period (Higgins 2011).

Comparison of intervention effects was conducted with a test of interaction (Altman 2003).

Trial sequential analyses

We conducted trial sequential analyses to reduce the risk of random error and prevent premature statements of superiority of the experimental or control intervention (Wetterslev 2008; Thorlund 2011). We performed a trial sequential analysis for all‐cause mortality with a type I error of 5%, type II error of 20% (80% power), and diversity‐adjusted required information size (Brok 2008; Wetterslev 2008; Brok 2009; Thorlund 2009; Wetterslev 2009; Thorlund 2011). We assumed an event proportion as estimated in the control group of our present review and an anticipated intervention effect of 5% relative risk reduction.

Results

Description of studies

Results of the search

We identified a total of 15,545 references of possible interest through searching the Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library (n = 3456), MEDLINE (n = 3388), EMBASE (n = 2124), Science Citation Index Expanded (n = 6515), Conference Proceedings Citation Index‐Science (n = 28), and reference lists (n = 34) (Figure 1). No reply about unpublished randomised trials was received from any of the contacted manufacturers of antioxidant supplements. To obtain additional information we wrote to authors of about 560 eligible trials. The authors of 140 trials (25%) responded. We excluded 14,134 duplicates and 401 clearly irrelevant references through reading the abstracts. Accordingly, 1010 references describing 615 trials were retrieved for further assessment. Of these, we excluded 47 studies because they were not randomised trials or did not fulfil our inclusion criteria. Reasons for exclusion are listed in the table 'Characteristics of excluded studies'.

1.

1

PRISMA flow diagram of identification of randomised trials for inclusion

We included 568 randomised trials. The authors of 481 trials did not report mortality (these trials are shown at http://ctu.rh.dk and listed under 'References to excluded studies'). The majority of these were small phase I or phase II trials with a short duration of follow‐up without assessment of clinical outcome measures. In nine trials there were deaths reported, but the authors did not report in which group of the trial and did not respond to our requests for additional information (Bussey 1982; Munoz 1985; Stich 1988; Berson 1993; Roncucci 1993; Bogden 1994; MacLennan 1995; Chandra 2001; Arad 2005).

In total, 78 randomised trials described in 473 references with 296,707 participants reported on mortality (Figure 1Table 1; Table 2).

1. Characteristics of included trials with low risk of bias.
Trial Design Number participants Mean age Suppl (Follow‐up)‐y Beta‐carotene (mg) Vitamin A (IU) Vitamin C (mg) Vitamin E (IU) Selenium (µg)
SCPS 1990 Parallel 1805 NA 5 (5) 50        
Murphy 1992 Parallel 109 NA 0.003 (0.25)   200000      
NIT2 1993 Parallel 3318 54 6 (6) 15 10000 180 60 50
PPS 1994 2 x 2 864 61 4 (4) 25   1000 440  
Pike 1995 Parallel 47 69 1 (1)   2666 90 45  
AMDS 1996 Parallel 71 72 1.5 (1.5) 12   750 200 50
CHAOS 1996 Parallel 2002 62 1.4 (1.4)       600  
NPCT 1996 Parallel 1312 63 4.5 (7.4)         200
PHS 1996 2 x 2 22,071 53 12 (12.9) 25        
SKICAP AK 1997 Parallel 2297 63 3.8 (3.8)   25000      
MINVITAOX 1999 2 x 2 725 84 2 (2) 6   120 16.5 100
NSCPT 1999 2 x 2 1621 49 4.5 (4.5) 30        
Correa 2000 2 x 2 x 2 976 51 6 (6) 30   2000    
Jacobson 2000 Parallel 112 42 0.5 (0.5) 12   500 400  
SPACE 2000 Parallel 196 65 1.42 (1.42)       800  
ALSRT 2001 Parallel 288 64 1 (1)       1000  
AREDS 2001 2 x 2 4757 68 6.3 (6.3) 15   500 400  
HATS 2001 2 x 2 160 53 3 (3) 25   1000 800 100
Graat 2002 2 x 2 652 NA 1 (1) 1.2 2000 60 272 25
HPS 2002 2 x 2 20,536 NA 5 (5) 20   250 660  
REACT 2002 Parallel 297 68 3 (3) 18   750 660  
VEAPS 2002 Parallel 353 56 3 (3)       400  
WAVE 2002 2 x 2 423 65 3 (3)     1000 800  
White 2002 Parallel 100 63 0.23 (0.23)     1000 223  
Wluka 2002 Parallel 136 64 2 (2)       500  
ASAP 2003 2 x 2 520 NA 6 (6)     250 272  
ATBC 2003 2 x 2 29,133 57 6.1 (14.1) 20     50  
Collins 2003 2 x 2 52 67 0.5 (2.5)       400  
Prince 2003 Crossover 61 58 0.25 (0.25) 3   150 74.5 75
Allsup 2004 Parallel 164 83 0.15 (0.5)   2666 120 60 60
CARET 2004 Parallel 18,314 58 4 (10) 30 25000      
DATOR 2004 Parallel 24 51 0.5 (4.5)       750  
LAST 2004 Parallel 61 75 1 (1) 10 2500 1500 500 200
Meydani 2004 Parallel 617 84 1 (1)       200 100
Mezey 2004 Parallel 51 48 0.25 (1)       1000  
VECAT 2004 Parallel 1193 66 4 (4)       500  
DATATOP 2005 2 x 2 800 61 2.6 (13)       2000  
Graf 2005 Parallel 160 58 1.5 (1.5)       5000  
HOPE TOO 2005 2 x 2 9541 66 4.5 (7)       400  
Limburg 2005 2 x 2 360 47 0.83 (0.83)         200
MAVIS 2005 Parallel 910 72 1 (1)   2666 60 10  
Mooney 2005 Parallel 284 37 1.25 (1.25)     500 400  
Tam 2005 Parallel 39 46 0.23 (2.67)     500 800  
WHS 2005 2 x 2 39,876 55 10.1 (10.1) 25     300  
Witte 2005 Parallel 32 NA 0.75 (0.75)   2666 500 400 50
MAVET 2006 Parallel 409 63 4 (4)       500  
UK PRECISE 2006 Parallel 501 67 0.5 (0.5)         200
Liu 2007 Parallel 763 85 1.6 (1.6) 16 1333 80 74 20
Plummer 2007 Parallel 1980 NA 3 (3) 18   750 600  
WACS 2007 2 x 2 x 2 8171 61 9.4 (9.4) 50   500 600  
ICARE 2008 Parallel 1434 69 1.5 (1.5)       400  
PHS 2008 2 x 2 x 2 14,641 64 8 (8) 50 5000 500 400 25
Garbagnati 2009 2 x 2 72 65 1 (1) 19   240 700  
Grieger 2009 Parallel 115 na 0.5 (0.5) 3   75 10  
SELECT 2009 2 x 2 35,533 62 5.5 (5.5)       400 200
SUVIMAX 2010 Parallel 13,017 49 7.54 (7.54) 6   120 33  

Abbreviation: 
 NA, not available. 
 Blank cells indicate that the supplement was not part of the trial.

2. Characteristics of included trials with high risk of bias.
Trial Design Number participants Mean age Suppl (Y) Beta‐carotene (mg) Vitamin A (IU) Vitamin C (mg) Vitamin E (IU) Selenium (µg)
Gillilan 1977 Crossover 52 57 0.5       1600  
Mckeown‐Eyssen 1988 Parallel 185 58 2     400 400  
Burns 1989 Parallel 19 81 0.1     200    
Penn 1991 Parallel 30 84 0.077   8000 100 50  
Chandra 1992 Parallel 96 74 1 16 1333 80 44 20
NIT1 1993 1/2 (2 x 2 x 2 x 2) 29584 NA 5.25 15 5000 120 33 50
de la Maza 1995 Parallel 74 50 1       500  
Takamatsu 1995 Parallel 147 47 6       136  
ter Riet 1995 2 x 2 88 NA 0.23     1000    
Hogarth 1996 2 x 2 106 83 0.083   8000 500    
ADCS 1 1997 2 x 2 341 73 2       2000  
Girodon 1997 2 x 2 81 84 2 6   120 15 100
Bonelli 1998 Parallel 304 NA 5   6000 180 30 200
GISSI 1999 2 x 2 11324 59 3.5       330  
de Waart 2001 Parallel 218 60 1.8       400  
PPP 2001 2 x 2 4495 64 3.6       330  
Stevic 2001 Parallel 28 57 1       1200 31.5
Sasazuki 2003 2 x 2 439 57 5 15   500    
Takagi 2003 Parallel 93 63 5       600  
ADCS 2 2005 Parallel 516 73 3       2000  
SIT 2006 2 x 2 x 2 3411 NA 3.25 15   500 200 75
CTNS 2008 Parallel 1020 68 9   5000 60 30 25

Abbreviation: 
 NA, not available. 
 Blank cells indicate that the supplement was not part of the trial. 
 The years of follow‐up are the same as the years of supplementation.

Included studies

The included trials are described in detail in the table 'Characteristics of included studies' and in Table 1; Table 2; Table 3; and Table 4.

3. Participants and outcome measures of included trials with a low risk of bias.
Trial Inclusion criteria Outcome measures Type of prevention
SCPS 1990 History of BCC or SCC Newly diagnosed BCC or SCC Secondary
Murphy 1992 Elderly nursing‐home residents Bacterial infections Secondary
NIT 2 1993 Esophageal dysplasia Cancer incidence, cancer mortality, all‐cause mortality Secondary
PPS 1994 Removed colorectal adenomas Newly diagnosed colorectal adenomas Secondary
Pike 1995 Elderly individuals Immune indices Primary
AMDS 1996 Age‐related macular degeneration Outcome of age‐related macular degeneration. Secondary
CHAOS 1996 Coronary artery disease Non‐fatal myocardial infarction and cardiovascular death Secondary
NPCT 1996 History of BCC or SCC Incidence of BCC and SCC, cancer incidence, cancer mortality, all‐cause mortality Secondary
PHS 1996 Male physicians Incidence of cancer and cardiovascular disease and all‐cause mortality Primary
SKICAP AK 1997 History of BCC or SCC Newly diagnosed BCC and SCC Secondary
MINVITAOX 1999 Institutionalized elderly patients Delayed‐type hypersensitivity skin response, humoral response to influenza vaccine, and infectious morbidity and mortality Secondary
NSCPT 1999 History of BCC or SCC Newly diagnosed BCC and SCC Secondary
Correa 2000 Multifocal atrophic gastritis with or without intestinal metaplasia and dysplasia Change of gastric precancerous lesions Secondary
Jacobson 2000 Heavy smokers DNA damage Primary
SPACE 2000 Stable haemodialysis patients with a documented medical history of CVD Acute myocardial infarction (fatal and nonfatal), ishaemic stroke, peripheral vascular disease, unstable angina, CVD mortality, all‐cause mortality Secondary
AREDS 2001 Aged‐related macular degeneration Increase in nuclear, cortical or posterior subcapsular opacity grades, cataract surgery, loss of visual acuity Secondary
Desnuelle 2001 Probable or definitive amyotrophic lateral sclerosis Change in functional status, survival, bulbar function Secondary
HATS 2001 Coronary artery disease Change in coronary stenosis, first cardiovascular event (death, myocardial infarction, stroke or revascularization) Secondary
Graat 2002 Elderly individuals Acute respiratory tract infections Primary
HPS 2002 Coronary and other occlusive arterial disease or diabetes Major coronary events, fatal and non‐fatal vascular events, cancer, other morbidity Secondary
REACT 2002 Cataract Cataract progression Secondary
VEAPS 2002 Healthy individuals (serum LDL cholesterol >3.37 mmol/L) Rate of change in the right distal common carotid artery intima media thickness Primary
WAVE 2002 Coronary artery disease Progression of coronary artery disease Secondary
White 2002 Patients with Barrett's oesophagus on long‐term acid suppression therapy Prevention of potentially pre‐malignant modifications to DNA in the human stomach Secondary
Wluka 2002 Knee osteoarthritis Change in cartilage volume Secondary
Collins 2003 Patients with peripheral arterial diseaseWalking ability and perceived quality of life Walking ability and perceived quality of life Secondary
Prince 2003 Primary biliary cirrhosis Change in patient fatigue Secondary
ASAP 2003 Healthy individuals (serum cholesterol >5 mmol/L) Progression of carotid atherosclerosis Primary
ATBC 2003 Male cigarette smokers Lung cancer and other major cancers, all‐cause and cause specific mortality, incidence of other disease Primary
Allsup 2004 Older institutionalised people Response to influenza vaccine Secondary
CARET 2004 Cigarette smokers, former smokers and workers exposed to asbestos Lung cancer, other cancers, mortality Primary
DATOR 2004 Type 1 diabetic patientsImpact on lipids and peroxidation during statin treatment Impact on lipids and peroxidation during statin treatment Secondary
LAST 2004 Age‐related macular degenerationVisual function Visual function Secondary
Meydani 2004 Elderly individuals Respiratory tract infections, emergency department visits, hospitalisation, and death Primary
Mezey 2004 Alcoholic hepatitis Clinical and laboratory parameters of liver function and markers of fibrogenesis Secondary
VECAT 2004 Early or no cataract Age related cataract Secondary
DATATOP 2005 Early Parkinson's disease not requiring levodopa Level of functional disability for initiation of levo‐dopa therapy Secondary
Graf 2005 Probable or definitive amyotrophic lateral sclerosis Survival Secondary
HOPE TOO 2005 History of cardiovascular disease or diabetes in the presence of at least one additional cardiovascular risk factor Cancer incidence, cancer deaths, major cardiovascular events, unstable angina, congestive heart failure, revascularization or amputation, all‐cause mortality Secondary
Limburg 2005 Patients with oesophageal dysplasia Change in histological grade of oesophageal dysplasia Secondary
MAVIS 2005 Elderly individuals irrespective of chronic illness Self reported days of infection, use of health services, quality of life Primary
Mooney 2005 Cigarette smokers Level of an intermediate cancer risk marker Primary
Tam 2005 Systemic lupus erythematosus Effects on markers of oxidative stress, antioxidant defence and endothelial function Secondary
WHS 2005 Female health professionals Invasive cancer, fatal and non‐fatal myocardial infarction, stroke, mortality Primary
Witte 2005 Stable chronic heart failure due to Ischaemic heart disease Left ventricular function, levels of pro‐inflammatory cytokines, quality of life Secondary
MAVET 2006 Male and female smokers Progression of carotid atherosclerosis Primary
Rayman 2006 General population Mood, quality of life, plasma selenium levels Primary
Liu 2007 Elderly institutionalised people Number of infections Primary
Plummer 2007 Population at high risk for stomach cancer Progression or regression of precancerous gastric lesions Secondary
WACS 2007 Women with a history of cardiovascular disease or three or more cardiovascular risk factors Cardiovascular disease morbidity and mortality Secondary
ICARE 2008 Patients with type 2 diabetes mellitus Myocardial infarction, stroke, and cardiovascular death Secondary
PHS 2008 US male physicians Cardiovascular diseases, cancer, and mortality Primary
Garbagnati 2009 Stroke survivors Clinical and functional status Secondary
Grieger 2009 Aged care residents Nutritional status, bone 
 quality and muscle strength Primary
SELECT 2009 Healthy men Prostate cancer incidence, incidence of other cancers, and mortality Primary
SUVIMAX 2010 General population Incidence of cancer and CVD and all‐cause mortality Primary

Abbreviations: 
 BCC ‐ basal cell carcinoma of the skin. 
 SCC ‐ squamous cell carcinoma of the skin. 
 SI conversion: to convert cholesterol values to mg/dL, divide by 0.0259.

4. Participants and outcome measures of included trials with a high risk of bias.
Trial Inclusion criteria Outcome measures Type of prevention
Gillilan 1977 Coronary artery disease Improvement of angina pectoris Secondary
Mckeown‐Eyssen 1988 Removed colorectal adenomas Newly diagnosed colorectal adenomas Secondary
Burns 1989 Senile dementia Progression of cognitive impairment and behavioural disturbance Secondary
Penn 1991 Elderly long‐stay patients Cell‐mediated immune function Secondary
Chandra 1992 Elderly individuals Infectuous morbidity Primary
NIT1 1993 General population Cancer incidence, cancer mortality, all‐cause mortality Primary
de la Maza 1995 Alcoholic cirrhosis Liver function, mortality, hospitalisation rates Secondary
Takamatsu 1995 General population Any illness Primary
ter Riet 1995 Nursing home patients with pressure ulcers Wound status and clinometric changes Secondary
Hogarth 1996 Elderly medical in‐patients Weight, serum albumin levels, activities of daily living, cognitive functioning, length of stay Secondary
ADCS 1 1997 Probable Alzheimer disease Death, institutionalisation, loss of ability to perform two of three basic activities of daily living Secondary
Girodon 1997 Elderly individuals Infectious morbidity Primary
Bonelli 1998 Removed colorectal adenomas Newly diagnosed colorectal adenomas Secondary
GISSI 1999 Recent myocardial infarction All‐cause mortality, non‐fatal myocardial infarction, non‐fatal stroke, cardiovascular death Secondary
de Waart 2001 Male cigarette smokers Progression of atherosclerosis Primary
PPP 2001 Elderly with at least one of the major cardiovascular risk factors Cardiovascular death, non‐fatal myocardial infarction and stroke, all‐cause mortality, total cardiovascular events, angina pectoris, transient ischaemic attacks, peripheral artery disease, revascularization procedures Primary
Stevic 2001 Probable or definitive amyotrophic lateral sclerosis Survival and rate of disease progression Secondary
Sasazuki 2003 Chronic atrophic gastritis Blood pressure Secondary
Takagi 2003 Liver cirrhosis Tumour free survival and cumulative survival rate Secondary
ADCS 2 2005 Amnestic mild cognitive impairment Alzheimer's disease Secondary
SIT 2006 General population Prevalence of dysplasia, gastric cancer, chronic atrophic gastritis, intestinal metaplasia Primary
CTNS 2008 Early cataract or no cataract Nuclear, cortical, or posterior subcapsular cataract opacity grades or cataract surgery. Secondary
Trial characteristics

Out of the 78 included trials, 46 trials used parallel‐group design, 30 trials used factorial design (25 trials 2 x 2; 4 trials 2 x 2 x 2; 1 trial half replicate of 2 x 2 x 2 x 2) and two trials used a cross‐over design (Pocock 2004).

In 63 trials (81%), the antioxidants were provided at no cost from pharmaceutical companies. In the rest of the trials funding was not reported.

The trials were conducted in Europe, North and South America, Asia, and Australia. Seventy‐one trials came from high‐income countries, four trials (de la Maza 1995; Correa 2000Low; Stevic 2001; Plummer 2007Low) from upper middle income countries, and three trials came from lower middle income countries (NIT1 1993; NIT2 1993Low; SIT 2006).

Participants

A total of 296,707 participants were randomly assigned in the 78 trials reporting mortality. The number of participants in each trial ranged from 19 to 39,876. The mean age was 63 years (range 18 to 103 years). The mean proportion of women was 46% in the 73 trials reporting sex (Table 1; Table 2).

Twenty‐six trials were primary prevention trials including 215,900 healthy participants; 52 trials were secondary prevention trials including 80,807 participants with gastrointestinal (n = 11 trials), cardiovascular (n = 10), neurological (n = 8), ocular (n = 6), dermatological (n = 5), rheumatoid (n = 2), renal and cardiovascular (n = 1), endocrinological (n = 2), or unspecified (n = 7) diseases. The main outcome measures in the primary prevention trials were cancer and mortality (cause‐specific and all‐cause mortality), and in the secondary prevention trials the outcome measures were progression of disease and mortality (cause‐specific and all‐cause mortality) (Table 4; Table 3).

Experimental interventions

Antioxidants were administered either alone or in combination with vitamins, minerals, or other interventions (Table 1; Table 2). All antioxidant supplements were administered orally. The doses and regimens of the antioxidant supplements were: beta‐carotene 1.2 to 50.0 mg (mean 19.2 mg; median 16 mg), vitamin A 1333 to 200,000 IU (mean 17,491 IU; median 5000 IU), vitamin C 60 to 2000 mg (mean 459 mg; median 400 mg), vitamin E 10 to 5000 IU (mean 539 IU; median 400 IU), and selenium 20 to 200 µg (mean 94 µg; median 75 µg) daily or on alternate days for 28 days to 12 years (mean 3.0 years; median 2.0 years). In one trial (Murphy 1992Low) antioxidants were applied in a single dose and participants were followed up for three months thereafter. The mean duration of follow‐up in all trials was 3.4 years (range 28 days to 14.1 years) (Table 1; Table 2).

Beta‐carotene was tested in 31 trials, vitamin A in 18, vitamin C in 41, vitamin E in 64, and selenium in 24 trials.

The antioxidant supplements were given in the following combinations:

  • beta‐carotene and vitamin A;

  • beta‐carotene and vitamin C;

  • beta‐carotene and vitamin E;

  • vitamin A and vitamin C;

  • vitamin C and vitamin E;

  • vitamin E and selenium;

  • selenium and zinc;

  • beta‐carotene, vitamin C, and vitamin E;

  • beta‐carotene, vitamin C, vitamin E, and selenium;

  • beta‐carotene, vitamin C, vitamin E, selenium, and zinc;

  • vitamin A, vitamin C, vitamin E;

  • vitamin A, vitamin C, vitamin E, selenium, and zinc;

  • vitamin A, vitamin C, vitamin E, selenium, methionine, and ubiquinone.

Control interventions

Seventy‐four trials used placebo and four trials used no intervention in the control group (ter Riet 1995; GISSI 1999; PPP 2001; Takagi 2003).

Concomitant interventions

In 17 trials, participants were supplemented with different mixtures of antioxidants as well as with vitamins and minerals without antioxidant properties (Burns 1989; Chandra 1992; NIT1 1993; NIT2 1993Low; Pike 1995Low; Hogarth 1996; AMDS 1996Low; Graat 2002Low; Allsup 2004Low; LAST 2004Low; MAVIS 2005 Low; Witte 2005Low; Liu 2007Low; CTNS 2008; PHS 2008Low; Garbagnati 2009Low; Grieger 2009Low).

In nine trials, the experimental and control groups were supplemented with vitamins and minerals (with or without antioxidant properties) (Gillilan 1977; Murphy 1992Low; Takamatsu 1995; ter Riet 1995; HATS 2001Low; Sasazuki 2003; Meydani 2004Low; DATATOP 2005Low; ADCS 2 2005). In seven of the trials, the supplementation was with vitamin E 4 IU (Takamatsu 1995; Meydani 2004Low), vitamin A 1000 IU (Murphy 1992Low), vitamin C 20 mg and 50 mg (ter Riet 1995; Sasazuki 2003), riboflavin 10 mg (Gillilan 1977), or niacin 100 mg (HATS 2001Low).

In some of the factorial designed trials, other interventions were administered to some of the participants in the antioxidant experimental arms and in the control arms. In the trials with factorial or parallel‐group design, the additional interventions tested were multivitamins and minerals; ubiquinone; L‐methionine; omega‐3 polyunsaturated fatty acids; citrus bioflavonoid complex; quercetin, bilberry extract, rutin (bioflavonoids); taurine; N‐acetyl cysteine; L‐glutathione; aged garlic; deprenyl‐selegiline (selective monoamine oxidase B inhibitor); donepezil (acetylcholinesterase inhibitor); riluzole (modulator of glutamatergic neurotransmission); amoxicillin, metronidazole (antibiotics); bismuth subsalicylate; omeprazole (proton‐pump inhibitor); aspirin; simvastatin (cholesterol‐lowering drug); celecoxib (inhibitor of cyclooxygenase); and ramipril (angiotensin‐converting enzyme inhibitor).

All‐cause mortality

Sixty‐nine of the trials published data on mortality. Mortality data were obtained from authors of a further nine trials (Jacobson 2000Low; ALSRT 2001Low; de Waart 2001; White 2002Low; Collins 2003Low; Sasazuki 2003; Allsup 2004Low; DATOR 2004Low; Tam 2005Low). In total, 78 trials reported on mortality.

Excluded studies

The reasons for exclusion of studies are given in the table 'Characteristics of excluded studies'.

Risk of bias in included studies

Fifty‐six of the 78 trials (72%) had low risk of bias, that is they had adequate generation of the allocation sequence, adequate allocation concealment, adequate blinding, adequate reporting, and were free of other bias. For an overview of the included trials with low risk of bias see Table 1 and Table 3.

Twenty‐two trials had one or more unclear or inadequate bias risk domains. For an overview of the included trials with high risk of bias see Table 2 and Table 4.

Seventy‐six trials reported losses to follow‐up. There was not a substantial difference in the losses to follow‐up between the intervention group and the control group (4872 out of 156,206 (3.12%) versus 3416 out of 107,599 (3.17%)).

Effects of interventions

Mortality in all trials

A total of 21,484 of 183,749 participants (11.7%) randomised to antioxidant supplements and 11,479 of 112,958 participants (10.2%) randomised to placebo or no intervention died. In the random‐effects model meta‐analysis, antioxidant supplements had no significant effect on mortality (RR 1.02, 95% CI 0.98 to 1.05). In the fixed‐effect model meta‐analysis, antioxidant supplements significantly increased mortality (RR 1.03, 95% CI 1.01 to 1.05). There was no significant intertrial heterogeneity (I2 = 12%) (Analysis 1.1).

1.1. Analysis.

1.1

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 1 Mortality in trials with a low or high risk of bias.

Sensitivity analyses taking trials with zero events into account

We included 15 trials with zero mortality in one arm. To account for the potential influence of these trials, we calculated the RR with 0.5, 0.01, and 0.001 as empirical continuity corrections (Sweeting 2004; Bradburn 2007). The random‐effects model RRs for the three continuity corrections were: RR 1.014 (95% CI 0.983 to 1.046); RR 1.027 (95% CI 1.002 to 1.052); RR 1.033 (95% CI 1.01 to 1.055), respectively. The fixed‐effect models with the same continuity corrections were all showing significantly increased mortality in the antioxidant group and an RR of 1.031 (95% CI 1.009 to 1.053) in all three analyses.

Overall, 481 trials had zero mortality in both the experimental and control groups. These trials were excluded from the meta‐analyses using RR as the association measure. The total number of participants in these trials was about 42000. Therefore we performed exploratory analyses adding an imagined trial with 1 death and 21,000 participants in each intervention group. The influence of zero events trials on our final result was not noticeable.

Analyses of bias risk

Inspection of the funnel plot in Figure 2 suggested potential bias (asymmetry). The adjusted‐rank correlation test (P = 0.48) found no significant evidence of bias. A regression asymmetry test (P = 0.00) found significant evidence of bias.

2.

2

Funnel plot of comparison: 1 Antioxidants versus placebo/no intervention, outcome: 1.1 Mortality in trials with a low or high risk of bias.

Meta‐regression analyses

Univariate meta‐regression analyses revealed that both trials with low risk of bias (RR 1.14, 95% CI 1.03 to 1.26, P = 0.006) and dose of vitamin A (RR 1.0006, 95% CI 1.0002 to 1.001, P = 0.002) were associated with a significantly higher estimated intervention effect on mortality. None of the other covariates (dose of beta‐carotene, dose of vitamin C, dose of vitamin E, dose of selenium, and duration of supplementation) were significantly associated with the estimated intervention effect on mortality.

In multivariate meta‐regression analysis including all covariates, trials with low risk of bias were significantly associated with a higher intervention effect on mortality (RR 1.14, 95% CI 1.04 to 1.26, P = 0.006, and dose of selenium was associated with a significantly lower estimated intervention effect on mortality (RR 0.9993, 95% CI 0.9987 to 0.9999, P = 0.022). None of the other covariates were significantly associated with the estimated intervention effect on mortality.

Intervention effects according to bias risk of trials (Analysis 1.1) 
 In trials with low risk of bias, mortality was significantly increased in the supplemented group (RR 1.04, 95% CI 1.01 to 1.07, P = 0.004, I2 = 4%). In trials with high risk of bias, mortality was significantly decreased in the supplemented group (RR 0.91, 95% CI 0.85 to 0.98, P = 0.02, I2 = 0%).

The difference between the estimate of antioxidants on mortality in trials with low risk of bias and trials with high risk of bias was statistically significant by the test of interaction (Z = 3.5, P = 0.0005).

Trial sequential analysis of 56 trials with low risk of bias assessing antioxidant supplements versus placebo was constructed based on a mortality of 10.6% in the control group, a relative risk reduction of 5% with antioxidant supplements, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 35%). The diversity‐adjusted required information size was 276,918 participants. The trial sequential analysis revealed that the cumulative Z‐curve (blue line) crossed the trial sequential monitoring boundary (red line) in 2005 after the 44th trial (Figure 3). Subsequently 12 trials have been published.

3.

3

Trial sequential analysis of 56 trials with low risk of bias assessing antioxidant supplements versus placebo was constructed based on a mortality of 10.6% in the control group, a relative risk reduction of 5% of antioxidant supplements, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 35%). The diversity‐adjusted required information size was 276,918 participants. The trial sequential analysis revealed that the cumulative Z‐curve (blue line) crossed the trial sequential monitoring boundary (red line) in 2005 after the 44th trial. Subsequently 12 trials have been published.

Sensitivity analysis assuming that all participants lost to follow‐up were dead (Analysis 1.2)

In trials with low risk of bias, mortality was not significantly increased in the supplemented group (RR 1.02, 95% CI 0.99 to 1.04, P = 0.14, I2 = 9%). In trials with high risk of bias mortality was significantly decreased in the supplemented group (RR 0.92, 95% CI 0.86 to 0.99, P = 0.03, I2 = 0%). Overall, antioxidant supplements had no significant effect on mortality (RR 1.01, 95% CI 0.99 to 1.03, I2 = 4%).

Random‐effects and fixed‐effect model meta‐analyses

For an overview of the effects of the different antioxidant supplements on mortality in a random‐effects or fixed‐effect model see Table 5 and Table 6.

5. Effects of antioxidant supplements versus placebo or no intervention ‐ random‐effects model.
Supplement Number of trials Participants number RR (95%CI) Heterogeneity (%)
Beta‐carotene given singly or in combination with other antioxidant supplements 31 195,503 1.02, 0.98 to 1.07 34
Beta‐carotene given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 26 173,006 1.05, 1.01 to 1.09 18
Beta‐carotene given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 5 22,497 0.81, 0.62 to 1.07 31
Vitamin A given singly or in combination with other antioxidant supplements 18 61,190 1.04, 0.96 to 1.13 26
Vitamin A given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 12 41,144 1.07, 0.97 to 1.18 27
Vitamin A given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 6 20,046 0.96, 0.85 to 1.07 0
Vitamin C given singly or in combination with other antioxidant supplements 41 90,191 1.01, 0.97 to 1.05 0
Vitamin C given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 29 65,942 1.02, 0.98 to 1.07 0
Vitamin C given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 12 24,249 0.93, 0.84 to 1.04 0
Vitamin E given singly or in combination with other antioxidant supplements 64 211,957 1.02, 0.99 to 1.04 0
Vitamin E given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 46 171,244 1.03, 1.00 to 1.05 0
Vitamin E given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 18 40,713 0.92, 0.85 to 0.99 0
Selenium given singly or in combination with other antioxidant supplements 24 86,150 0.96, 0.91 to 0.1.01 0
Selenium given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 17 62,740 0.97, 0.91 to 1.03 0
Selenium given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 7 23,410 0.91, 0.81 to 1.01 0
All antioxidant supplements given singly or in combination with other antioxidant supplements 78 296,707 1.02, 0.98 to 1.05 12
6. Effects of antioxidant supplements versus placebo or no intervention ‐ fixed‐effect model.
Supplement Number of trials Participants number RR (95%CI) Heterogeneity (%)
Beta‐carotene given singly or in combination with other antioxidant supplements 31 195,503 1.05, 1.02 to 1.07 34
Beta‐carotene given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 26 173,006 1.06, 1.03 to 1.08 21
Beta‐carotene given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 5 22,497 0.88, 0.78 to 0.99 31
Vitamin A given singly or in combination with other antioxidant supplements 18 61,190 1.08, 1.03 to 1.13 26
Vitamin A given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 12 41,144 1.11, 1.05 to 1.16 27
Vitamin A given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 6 20,046 0.95, 0.85 to 1.07 0
Vitamin C given singly or in combination with other antioxidant supplements 41 90,191 1.01, 0.96 to 1.05 0
Vitamin C given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 29 65,942 1.02, 0.98 to 1.07 0
Vitamin C given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 12 24,249 0.92, 0.83 to 1.03 0
Vitamin E given singly or in combination with other antioxidant supplements 64 211,957 1.01, 0.99 to 1.04 0
Vitamin E given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 46 171,244 1.03, 1.00 to 1.05 0
Vitamin E given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 18 40,713 0.92, 0.85 to 0.99 0
Selenium given singly or in combination with other antioxidant supplements 24 86,150 0.95, 0.90 to 0.1.01 0
Selenium given singly or in combination with other antioxidant supplements ‐ trials with low risk of bias 17 62,740 0.97, 0.91 to 1.03 0
Selenium given singly or in combination with other antioxidant supplements ‐ trials with high risk of bias 7 23,410 0.91, 0.81 to 1.01 0
All antioxidant supplements given singly or in combination with other antioxidant supplements 78 296,707 1.03, 1.01 to 1.05 12

Primary and secondary prevention (Analysis 1.3)

In primary prevention trials with low risk of bias, mortality was not significantly increased in the supplemented group in the random‐effects model analysis (RR 1.03, 95% CI 0.97 to 1.08, P = 0.36, I2 = 46.9%) but significantly increased in the fixed‐effect analysis (RR 1.05, 95% CI 1.02 to 1.08, P < 0.0003).

In secondary prevention trials with low risk of bias, mortality was not significantly influenced by supplements (RR 1.03, 95% CI 0.99 to 1.07, P = 0.21, I2 = 0%).

In primary prevention trials with high risk of bias, mortality was not significantly influenced by supplements (RR 0.93, 95% CI 0.84 to 1.03, P = 0.19, I2 = 0%).

In secondary prevention trials with high risk of bias, mortality was significantly reduced by supplements (RR 0.90, 95% CI 0.81 to 0.99, P = 0.04, I2 = 12%).

Meta‐regression analysis did not find significant differences in the estimated intervention effects in primary and secondary prevention trials.

Sensitivity analyses excluding trials with co‐administration of additional supplements to the experimental group (Analysis 1.4)

Sensitivity analyses excluding 18 trials that used co‐interventions in the form of extra vitamins or trace elements with or without antioxidant functions in the experimental group did not noticeably change our results. In the 43 trials with low risk of bias antioxidant supplements significantly increased mortality (RR 1.04, 95% CI 1.01 to 1.07, P = 0.007, I2 = 9%).

Sensitivity analyses excluding trials with co‐administration of additional antioxidant supplements to both the experimental and control groups (Analysis 1.5)

Sensitivity analyses excluding 15 trials that used co‐interventions in the form of extra vitamins in both the experimental and control groups did not noticeably change our results. In the 48 trials with low risk of bias antioxidant supplements significantly increased mortality (RR 1.04, 95% CI 1.00 to 1.07, P = 0.04, I2 = 14%).

Sensitivity analysis excluding factorial trials testing collateral interventions (Analysis 1.6)

Sensitivity analysis excluding 26 factorial trials testing collateral interventions which could lead to potential confounding did not noticeably change our results. In the 38 trials with low risk of bias, mortality was significantly increased in the supplemented group (RR 1.10, 95% CI 1.05 to 1.15, P = 0.0002, I2 = 0%).

Sensitivity analysis excluding trials with co‐administration of additional supplements to experimental and control groups, and factorial trials testing collateral interventions (Analysis 1.7)

Sensitivity analysis excluding 42 trials with co‐administration of additional supplements to the experimental and control groups, and factorial trials testing collateral interventions did not noticeably change our results. In the 27 trials with low risk of bias, mortality was significantly increased in the supplemented group (RR 1.12, 95% CI 1.06 to 1.18, P < 0.0001, I2 = 0%).

The estimate of mortality risk in the 27 trials with low risk of bias without potential confounding interventions (RR 1.12, 95% CI 1.06 to 1.18) compared to the estimate of mortality risk in the 56 trials with low risk of bias with potential confounding (RR 1.04, 95% CI 1.02 to 1.07) was significantly increased (Z = ‐2.47; P = 0.013).

Sensitivity analyses according to type of antioxidant supplement

Beta‐carotene

In a random‐effects model meta‐analysis beta‐carotene used singly or in combination with other antioxidants significantly increased mortality in 26 trials with low risk of bias (RR 1.05, 95% CI 1.01 to 1.09, I2 = 21%) (Analysis 1.8; Table 5).

1.8. Analysis.

1.8

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 8 Mortality in beta‐carotene trials with a low or high risk of bias.

A fixed‐effect model meta‐analysis found a significant harmful effect of beta‐carotene in trials with low risk of bias (RR 1.06, 95% CI 1.03 to 1.08) (Table 6).

Trial sequential analysis of 26 trials with low risk of bias assessing beta‐carotene versus placebo was constructed based on a mortality of 11.1% in the control group, a relative risk reduction of 5% with beta‐carotene, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 54%). The diversity‐adjusted required information size was 244,756 participants.The trial sequential analysis revealed that the cumulative Z‐curve (blue line) crossed the trial sequential monitoring boundary (red line) in 2007 after the 22nd trial (Figure 4). Subsequently four trials have been published.

4.

4

Trial sequential analysis of 26 trials with low risk of bias assessing beta‐carotene versus placebo was constructed based on a mortality of 11.1% in the control group, a relative risk reduction of 5% of beta‐carotene, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 54%). The diversity‐adjusted required information size was 244,756 participants.The trial sequential analysis revealed that the cumulative Z‐curve (blue line) crossed the trial sequential monitoring boundary (red line) in 2007 after the 22nd trial. Subsequently four trials have been published.

Vitamin A

In a random‐effects model meta‐analysis vitamin A used singly or in combination with other antioxidants had no significant effect on mortality in 12 trials with low risk of bias (RR 1.07, 95% CI 0.97 to 1.18, I2 = 27% ) (Analysis 1.9; Table 5).

1.9. Analysis.

1.9

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 9 Mortality in vitamin A trials with a low or high risk of bias.

A fixed‐effect model meta‐analysis found a significant harmful effect of vitamin A in trials with low risk of bias (RR 1.11, 95% CI 1.05 to 1.16) (Table 6).

Trial sequential analysis of 12 trials with low risk of bias assessing vitamin A versus placebo was constructed based on a mortality of 14% in the control group, a relative risk reduction of 5% with vitamin A, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 65%). The diversity‐adjusted required information size was 108,645 participants.The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the trial sequential monitoring boundary (red line) (Figure 5).

5.

5

Trial sequential analysis of 12 trials with low risk of bias assessing vitamin A versus placebo was constructed based on a mortality of 13.6% in the control group, a relative risk reduction of 5% of vitamin A, a type I error of 5%, and a type II error of 20% (80% power). There was diversity (D2 = 65%). The diversity‐adjusted required information size was 108,645 participants.The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the trial sequential monitoring boundary (red line). Neither did the cumulative Z‐curve reach the area of futility, which was not constructed by the program due to too little information.

Vitamin C

In a random‐effects model meta‐analysis vitamin C used singly or in combination with other antioxidants had no significant effect on mortality in 29 trials with low risk of bias (RR 1.02, 95% CI 0.98 to 1.07, I2 = 0%) (Analysis 1.10; Table 5).

1.10. Analysis.

1.10

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 10 Mortality in vitamin C trials with a low or high risk of bias.

A fixed‐effect model meta‐analysis found no significant effect of vitamin C in trials with low risk of bias (RR 1.02, 95% CI 0.98 to 1.07) (Table 6).

Trial sequential analysis of 29 trials with low risk of bias assessing vitamin C versus placebo was constructed based on a mortality of 9.3% in the control group, a relative risk reduction of 5% with vitamin C, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity (D2 = 0%). The required information size was 552,959 participants. The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the trial sequential monitoring boundary (red line) (Figure 6).

6.

6

Trial sequential analysis of 29 trials with low risk of bias assessing vitamin C versus placebo was constructed based on a mortality of 9.3% in the control group, a relative risk reduction of 5% of vitamin C, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity. The required information size was 552,959 participants.The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the trial sequential monitoring boundary (red line). Neither did the cumulative Z‐curve reach the area of futility, which was not constructed by the program due to too little information.

Vitamin E

In a random‐effects model meta‐analysis vitamin E used singly or in combination with other antioxidants significantly increased mortality in 46 trials with low risk of bias (RR 1.03, 95% CI 1.00 to 1.05, I2 = 0%) (Analysis 1.11; Table 5).

1.11. Analysis.

1.11

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 11 Mortality in vitamin E trials with a low or high risk of bias.

A fixed‐effect model meta‐analysis found no significant effect of vitamin E in trials with low risk of bias (RR 1.03, 95% CI 1.00 to 1.05) (Table 6).

Trial sequential analysis of 46 trials with low risk of bias assessing vitamin E versus placebo was constructed based on a mortality of 10.3% in the control group, a relative risk reduction of 5% with vitamin E, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity (D2 = 0%). The required information size was 374,427 participants. The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the monitoring boundary (red line) (Figure 7).

7.

7

Trial sequential analysis of 46 trials with low risk of bias assessing vitamin E was constructed based on a mortality of 10.3% in the control group, a relative risk reduction of 5% of vitamin E, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity. The required information size was 374,427 participants. The trial sequential analysis revealed that the cumulative Z‐curve (blue line) did not cross the monitoring boundary (red line). Neither did the cumulative Z‐curve reach the area of futility, which was not constructed by the program due to too little information.

Selenium

In a random‐effects model meta‐analysis selenium used singly or in combination with other antioxidants had no significant effect on mortality in 17 trials with low risk of bias (RR 0.97, 95% CI 0.91 to 1.03, I2 = 0%) (Analysis 1.12; Table 5).

1.12. Analysis.

1.12

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 12 Mortality in selenium trials with a low or high risk of bias.

A fixed‐effect model meta‐analysis found no significant effect of selenium in trials with low risk of bias (RR 0.97, 95% CI 0.91 to 1.03, I2 = 0%) (Table 6).

Trial sequential analysis of 17 trials with low risk of bias assessing selenium versus placebo was constructed based on a mortality of 5% in the control group, a relative risk reduction of 6.4% with selenium, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity (D2 = 0%). The required information size was 572,856 participants.The trial sequential analysis revealed that the cumulative Z‐curve did not cross the trial sequential monitoring boundary (Figure 8).

8.

8

Trial sequential analysis of 17 trials with low risk of bias assessing selenium versus placebo was constructed based on a mortality of 5% in the control group, a relative risk reduction of 6.4% of selenium, a type I error of 5%, and a type II error of 20% (80% power). There was no diversity. The required information size was 572,856 participants.The trial sequential analysis revealed that the cumulative Z‐curve did not cross the trial sequential monitoring boundary. Neither did the cumulative Z‐curve reach the area of futility, which was not constructed by the program due to too little information.

Discussion

Summary of main results

Our systematic review contains a number of findings. Beta‐carotene, vitamin A, and vitamin E given singly or combined with other antioxidant supplements significantly increased mortality. There is no evidence that vitamin C or selenium may increase longevity. We confirmed that trials with inadequate bias control significantly overestimate intervention effects (Schulz 1995; Moher 1998; Kjaergard 2001; Bjelakovic 2004; Bjelakovic 2006; Bjelakovic 2008). The detrimental effect of antioxidant supplements became significantly more pronounced when we excluded all trials with potential confounding. Our findings support and extend our previous findings regarding antioxidant supplements and increased mortality (Bjelakovic 2004; Bjelakovic 2007a; Bjelakovic 2008).

Compared to our previous review (Bjelakovic 2007a; Bjelakovic 2008), the number of included trials in the present review is expanded with 11 new trials (16.4%) adding another 64,157 participants (27.6%). Moreover, we have obtained updated results of longer follow‐up from two large‐scale randomised trials (SIT 2006; SUVIMAX 2010Low). In spite of these expansions our results remain largely the same.

Fixed‐effect model and random‐effects model meta‐analyses

The fixed‐effect model meta‐analysis assumes that the true intervention effect is the same in every randomised trial, that is, the effect is fixed across trials. The random‐effects model assumes that the effects being estimated based on the different randomised trials differ but follow some general distribution. When there is no heterogeneity (I2 = 0%), then fixed‐effect and random‐effects model meta‐analyses tend to give the same result. With increasing heterogeneity, the estimated intervention effect as well as the corresponding 95% confidence interval will differ between the two models.

The meta‐analyses we conducted included a heterogeneous set of randomised trials, for example, healthy participants or patients, single antioxidant supplement or combination, short duration of follow‐up or long duration. These aspects could strengthen the argument for only employing the random‐effects model. We were requested to do so in our sister publication of the present review in JAMA (Bjelakovic 2007a). The standard random‐effect model used in RevMan Analysis (RevMan 2008) is the DerSimonian and Laird method, which models the known differences between trials by incorporating a variance parameter tau2 to account for across‐trial variation (DerSimonian 1986). Adoption of the random‐effects model in meta‐analysis permits extension of inferences to a broader population of studies than the fixed‐effect model does, which excludes the parameter tau2 from the model.

The use of the random‐effects model may, however, come at a price. If there is between trial heterogeneity then the weight of the large trials (usually providing more realistic estimates of intervention effects) becomes less. At the same time, the weight of small trials (usually providing more unrealistic estimates of intervention effects due to 'bias' (systematic errors) and 'chance' (random errors)) increases. We, therefore, also performed our meta‐analyses with the fixed‐effect method. In all meta‐analyses the pooled estimate of increased mortality in the antioxidant‐supplemented group became more pronounced (Table 6). More of the meta‐analyses became more significant or changed from non‐significant to significant detrimental effects of vitamin A, beta‐carotene, and vitamin E (Table 6).

The choice of statistical model for performing meta‐analysis of sparse data is important (Sweeting 2004; Bradburn 2007). Because many methods are based on large sample approximations, they may be unsuitable when events are rare. Bradburn et al found that no method gives completely unbiased estimates (Bradburn 2007). At event rates below 1%, the Peto odds ratio method appears to be the least biased and the most powerful method when there is no substantial imbalance in treatment and control group sizes within trials, and treatment effects are not exceptionally large. Bradburn et al (Bradburn 2007) also demonstrated that the Peto odds ratio works well up to event rates around 10%. The calculation avoids addition of 0.5 event adjustments or any other adjustment. When we applied the Peto odds ratio (a fixed‐effect model analysis), we found even stronger support for detrimental effects of the supplements (for all 78 trials: OR 1.04, 95% CI 1.01 to 1.07; for the 56 low‐bias risk trials: OR 1.06, 95% CI 1.03 to 1.09). Again, our random‐effects model analyses are supported and extended by the fixed‐effect model analyses.

Strengths

Our review offers a number of strengths. It follows the overall plans of a published, peer reviewed Cochrane protocol (Bjelakovic 2003), taking into consideration our previous findings in a systematic review on antioxidant supplements for preventing gastrointestinal cancers (Bjelakovic 2004) and requests for having all preventive trials assessed (Forman 2004). Our review represents a comprehensive review of the topic, including 78 randomised trials with more than a quarter of a million participants. This increases the precision and power of our analyses (Higgins 2011).

Previous meta‐analyses of preventive trials of antioxidant supplements have included less information (lung cancer, 4 trials with 109,394 participants (Caraballoso 2003); cardiovascular diseases, 8 trials with 138,113 participants (Vivekananthan 2003); gastrointestinal cancers, 14 trials with 170,525 participants (Bjelakovic 2004); colorectal adenoma, 8 trials with 17620 participants (Bjelakovic 2006); cancer or pre‐invasive lesions, 7 trials with 5112 participants (Davies 2006); mortality, 19 trials with 135,967 participants (Miller 2005); as well as their efficacy and safety, 5 trials with 47,289 participants (Huang 2006)). Previous meta‐analyses either found neutral effects of the supplements (Caraballoso 2003; Vivekananthan 2003; Bjelakovic 2006; Davies 2006; Huang 2006) or reported a significantly increased mortality (Caraballoso 2003; Vivekananthan 2003; Bjelakovic 2004; Miller 2005; Bjelakovic 2007a; Bjelakovic 2008).

We conducted a thorough review with methodology following the recommendations of The Cochrane Collaboration (Higgins 2011) and findings of methodological studies (Schulz 1995; Moher 1998; Kjaergard 2001). More than two thirds of the included trials with more than 240,000 participants fall in to the group of low risk of bias trials. This highlights the validity of our results (Schulz 1995; Moher 1998; Kjaergard 2001). Antioxidant supplements not only seem to be one of the most researched topics in the world, they also seem to be one of the most adequately researched questions. Usually, only a small proportion of trials use adequate methodologies (Gluud 2006a; Gluud 2006b).

Our meta‐analyses had little trial heterogeneity. This increases the trustworthiness of our findings. Our analyses were robust to sensitivity analyses involving different imputations of mortality in the zero‐event intervention groups. We gave full account of all 481 identified trials assessing the supplements having zero events in both intervention groups. These trials were mostly assessing short‐term supplement administration and surrogate outcome measures. Our results were robust to exploratory analyses, adding an imagined trial with 21000 participants and one death in each intervention group. Accordingly, the increased mortality does not seem to be an artefact created by exclusion of trials with zero events in both intervention groups (Sweeting 2004; Bradburn 2007). Furthermore, all‐cause mortality should generally be connected with unbiased estimates (Wood 2008).

Our estimates of increased mortality in low‐bias risk trials increased significantly when we excluded factorial trials as well as other trials with collateral interventions. These trials may all suffer from potential confounding from the collateral interventions. This highlights the potential dramatic public health consequences of our results.

A large number of unpublished trials on supplements may exist. Their results are more likely to have been either neutral or negative than to have shown beneficial effects (Dickersin 2003). Accordingly, our estimates of increased mortality are likely to be conservative. Again, this highlights the potential dramatic public health consequences of our results.

We also performed trial sequential analyses to estimate the risk of random errors in the cumulative meta‐analyses and to prevent premature statements of superiority or inferiority of antioxidant supplements (Brok 2008; Wetterslev 2008; Brok 2009; Thorlund 2009; Wetterslev 2009; Thorlund 2011). Overall, the finding of increased mortality with the five assessed antioxidant supplements together and with beta‐carotene did not seem to be due to a random error. The trial sequential analyses for vitamin A, E, C, and selenium demonstrated that we may not have sufficient evidence as none of these analyses crossed any of the alpha monitoring boundaries or any of the beta monitoring boundaries (are of futility). These areas were not created by the program due to insufficient information (Thorlund 2011). One should discuss, however, how much evidence one would require when dealing with potential harm. On the one hand, harmful effects can also occur due to random error and therefore sufficient information needs to be assessed to demonstrate harm beyond reasonable doubt. On the other hand, when the evidence is pointing towards harm, then maybe ethical considerations should prevail and proof beyond any doubt may not be necessary.

Limitations

Our systematic review has several limitations. As with all systematic reviews, our findings and interpretations are limited by the quality and quantity of available evidence on the effects of specific supplements on mortality. The examined populations varied. The effects of supplements were assessed in the general population or in patients with gastrointestinal, cardiovascular, neurological, skin, ocular, renal, endocrinological, rheumatoid, and undefined diseases in a stable phase. These populations mostly came from countries without overt deficiencies of specific supplements. Accordingly, we are unable to assess how antioxidant supplements affected mortality in populations with specific needs.

Most trials assessed combinations of different supplements, which reflects the way supplements are marketed, sold, and taken by people (Balluz 2000; Millen 2004; Radimer 2004; Nichter 2006). As a result, we have compared antioxidants with different properties, given at different doses and for varying durations, singly or combined. We are aware of the potential risks in assessing together the effects of different types of antioxidants with different mechanisms of action, biotransformation, and bioavailability.

The methodological quality of some of the trials was assessed using the published reports, which may not reflect the actual design and risk of bias of the trials. Only some authors responded to our requests for further information.

Nevertheless, since it is likely that these trials had different degrees of methodologic rigor, our overall analyses should be evaluated with care as they include data from trials in which bias is likely as well as data from trials in which bias is less likely. Furthermore, our review includes several trials in which we cannot exclude confounding by other interventions examined in these trials. (By excluding data from such trials our relative mortality risk rose from 1.04 to 1.12 in low‐bias risk trials, an increase from 4% to 12%.)

Clinical speculations

There are pros (Palace 1999; Vertuani 2004; Kawanishi 2005) and cons (Maxwell 1999) in the literature about vitamin A being an antioxidant. Most trials assessed combinations of different supplements, which reflects the way supplements are marketed, sold, and taken by people (Balluz 2000; Millen 2004; Radimer 2004; Nichter 2006).

All available non‐enzymatic antioxidants work differently in the human body, and most of them exert effects that are non‐antioxidant. We are not able to point to the specific biochemical mechanisms behind the detrimental effects. We found that trials examining the individual supplements singly were rare. It has been suggested that antioxidant supplements may show interdependency and may have effects only if given in combination (Hercberg 1998).

A recent review by Ristow and Schmeissner suggests that antioxidant supplements may reduce the life span of organisms and that the 'free radical theory of aging' may not be correct (Ristow 2011). In fact, the authors suggest that reactive oxygen species promote health and longevity (Ristow 2011).

Most trials investigated the effects of supplements administered at higher doses than those commonly found in a balanced diet, and some of the trials used doses well above the recommended daily allowances and even above the upper tolerable intake levels (Anonymous 2000a; Anonymous 2000b) (see Figure 9 for overview of recommended dietary allowance, tolerable upper intake level, and experimental doses and the regimen used). Our meta‐regression analyses revealed significant effects of dose of beta‐carotene, vitamin A, and selenium on mortality. The duration of supplementation and follow‐up differed among the trials. However, we found no significant effect of treatment duration on our results.

9.

9

Recommended dietary allowance, tolerable upper intake level, experimental doses, and regimen used in antioxidant supplements

We only assessed all‐cause mortality. We are not able to determine the cause of the increased mortality. It is likely that increased cancer and cardiovascular mortality are the main reasons for the increased all‐cause mortality (Caraballoso 2003; Vivekananthan 2003; Bjelakovic 2004; Lawson 2007). Further study of causes of mortality is needed. Our results extend the evidence in previous reviews (Caraballoso 2003; Vivekananthan 2003; Bjelakovic 2004; Miller 2005; Bjelakovic 2006; Davies 2006; Huang 2006; Bjelakovic 2007a; Bjelakovic 2008) and guidelines (Ritenbaugh 1999; Atkins 2002; McKevith 2003) suggesting that antioxidant supplements may not be beneficial.

Beta‐carotene, administered singly or in combination with other antioxidants, significantly increased mortality in trials with low risk of bias. The trial sequential analysis supported this finding. Recent studies have suggested that beta‐carotene may act as a co‐carcinogen (Lee 2003; Paolini 2003).

In random‐effects model meta‐analysis vitamin A used singly or in combination with other antioxidants had no significant effect on mortality, although there was a trend toward a harmful effect. A fixed‐effect model meta‐analysis found a significant harmful effect of vitamin A. Furthermore, we observed an association which was significant between vitamin A dose and mortality. Recent research revealed that vitamin A can cause oxidative damage to deoxyribonucleic acid (Murata 2000).

We found that vitamin E given singly or combined with four other antioxidants significantly increased mortality in trials with low risk of bias. This is in agreement with a previous meta‐analysis (Miller 2005). The chance that vitamin E may provide benefit seems low (Brown 2005; Devaraj 2005; Guallar 2005). The trial sequential analysis revealed that we might need more randomised trials assessing the influence of vitamin E on mortality to attain firm evidence or to discard such an intervention effect, with the required information size.

The trials in which vitamin C was applied singly or in different combinations with beta‐carotene, vitamin A, vitamin E, and selenium found no significant effect on mortality. According to the confidence intervals, small beneficial or large harmful effects cannot be excluded. Studies have demonstrated that vitamin C may act as both a pro‐oxidant and as an antioxidant in vivo (Podmore 1998; Duarte 2005).

Selenium given singly or in combination with other supplements had no significant effect on mortality. Recently, a randomised trial and an observational study have shown that selenium may carry health risks (Bleys 2007a; Bleys 2007b; Stranges 2007).

Our findings contradict the findings of observational studies claiming that antioxidants improve health (Machlin 1987; Diplock 1994; van Poppel 1997; Diplock 1998). Considering that more than 10% to 20% of the adult population (80 million to 160 million people) in North America and Europe may consume the assessed supplements (Balluz 2000; Millen 2004; Radimer 2004; Nichter 2006) the public health consequences could be substantial.

There are several possible explanations for the increased mortality induced by antioxidant supplements. Although oxidative stress has a hypothesised role in the pathogenesis of many chronic diseases it may be the consequence of pathological conditions (Halliwell 2000). By eliminating free radicals from our organism, we interfere with some essential defensive mechanisms like apoptosis, phagocytosis, and detoxification (Simon 2000; Salganik 2001; Kimura 2005). Recent evidence suggests that inhibition of reactive oxygen species formation in cells decreases the life span of nematodes (Schulz 2007). Antioxidant supplements are not subjected to the same rigorous toxicity studies as other pharmaceutical agents (Bast 2002). Better understanding of mechanisms and actions of antioxidants in relation to a potential disease is needed (Ratnam 2006). As suggested, antioxidant supplements may interfere with reactive oxygen species and interfere with health and longevity (Ristow 2011).

Because we examined only the influence of antioxidant supplements, our findings should not be translated to potential effects of fruits and vegetables (Dragsted 2006). Furthermore, we did not examine the treatment effect of antioxidant supplements (tertiary prevention) in specific patient groups or the preventive effects of antioxidant supplements for patient groups with verified specific need for antioxidant supplements. However, we only examined a selected group of commonly used antioxidants. We cannot exclude that other substances with antioxidant properties may have neutral or beneficial effects. Other systematic reviews should address these issues.

Authors' conclusions

Implications for practice.

We have found no convincing evidence that antioxidant supplements decrease mortality. Even more, beta‐carotene and perhaps, vitamin E and vitamin A seem to increase mortality. Therefore, we cannot recommend the use of antioxidant supplements as a primary and secondary preventive measure in the population groups studied in the present review.

Implications for research.

Our review highlights the necessity for national and international laws and regulations that require that anything sold to the public claiming health benefits should be subjected to adequate assessment of benefits and harms before market release.

The significant association between unclear or inadequate methodological quality and overestimation of intervention effects has again focused on the need for more objective assessment of preventive and therapeutic interventions.

The published trials that were included in the systematic review lacked important information. We suggest that researchers involved in future trials should seriously consider adopting the CONSORT Statement (CONSORT ‐ Consolidated Standards of Reporting Trials at www.consort‐statement.org).

Feedback

Study employed inappropriate statistical analysis, 5 June 2008

Summary

Study employed inappropriate statistical analysis 
 Steve Hickeyi, Claus Hancke, Rob Verkerkii, Gert Schuitemakeriii, Andrew Hickey, Hilary Roberts, and Len NoriegaiiFaculty of Computing Engineering and technology, Staffordshire University, Beaconside, Stafford, England. 
 iiAlliance for Natural Health, Dorking, Surrey, England. 
 iiiInternational Society for Orthomolecular Medicine, Ortho Institute, Toronto, Canada. 
 
 The review by Bjelakovic et al. (2008) has statistical inconsistencies, which bring its conclusions into doubt. Moreover, its statistical assumptions are biologically unsound. Furthermore, Bjelakovic et al. used non‐blind selection of trials and post hoc selection of statistical tests. 
 
 1. The study selected the trials in a non‐blind fashion, thereby introducing bias. Blinding is crucially important, as previous knowledge may cause reviewers to distort comparisons.1 More importantly, the authors have not addressed this critical objection, despite it having been published as a specific and direct refutation of the earlier version of this review.2 This study is thus a statement of the opinion of the authors and depends on their prior prejudice. 
 
 2. The results of this conflicting review should be considered a tentative modification of the prior probability of substantial positive effects (Bayes’ theorem). Since the claimed effects are relatively small, and the N value is large, the importance of these effects may be considered minimal in the light of existing knowledge. 
 
 3. The presentation of results appears biased. For example, one of the main reported results claimed increased risk with beta‐carotene if selenium was not included. We might ask what the results would be if selenium were included, or if vitamins C and E were excluded, and so on. The analyses in the last section (10.13‐10.16) exclude selenium, but each of the antioxidants studied, that is vitamin A (which is not an antioxidant) and the other interventions, could have been excluded. However, results for the exclusion of selenium alone are reported. The authors’ suggestion that this is not post hoc design because it is based on their previous review on antioxidants and gastrointestinal cancer, appears to be a circular argument. The current review contains 10 trials of antioxidant supplements in gastrointestinal cancers, and the studies in the two reviews overlap. 
 
 4. Did using only the first period from crossover trials introduce bias? To show that this is not the case the results from the full datasets could have been provided. 
 
 5. The review reported no effect on mortality, in the more appropriate random effects model. The fixed effects model is inappropriate in this heterogeneous group of studies involving multiple interventions. It is not acceptable that such heterogeneity is ignored.3 Pooled studies must be compatible. The use of two such analytical approaches (tests) should be balanced by a decrease in the acceptable confidence level, which was not done. 
 
 6. The review does not state how many statistical tests on subgroups were actually performed, nor how many results were unreported. Failure to include a full description of the analyses invalidates all conclusions, as they could have arisen from repeated testing. 
 
 7. The analyses performed included multiple subgroups and comparisons, yet the authors report no specific calculation to support the validity of this procedure. Again, the potential for bias is substantial. 
 
 8. The reported statistics are inconsistent; for example, the abstract claims: low‐bias risk trials on selenium found no significant effect on mortality (RR 0.91, 95% CI 0.76 to 1.09) but analysis 01.12 gives “0.90 [0.80, 1.01]” for low risk trials and exactly the same output results for high risk trials. Moreover, the combined result showed a significant benefit “0.90 [0.83, 0.98]”. 
 
 9. It is invalid to assume that weighting studies based on the N value alone minimizes bias, especially if a small number of large studies receive a high weighting. Systematic bias from the methods employed occurs in experiments regardless of their scale. For example, on page 185, two studies, ATBC and CARET, involving smokers (and asbestos workers), were weighted 47.3% of the total. Overall, the presentation of statistical results in the review suggests a high degree of experimenter choice. The review does not provide sufficient detail of the actual and totality of tests performed to evaluate the results with true statistical meaning. 
 
 The conclusions cannot be considered as more than the prior prejudice of the authors. Cochrane should make clear the subjective nature of this review.

1 Higgins J.P.T. Green S. (2006) Cochrane Handbook for Systematic Reviews of Interventions 4.2.6 [updated September 2006]. In: The Cochrane Library, Issue 4, John Wiley & Sons, Ltd. 
 2 S. Hickey H.J. Roberts and L.A. Noriega (2007) Poor methodology in meta‐analysis of 
 vitamins, JOM, 22(1), 8–10. 
 3 Hemilä H. (2007) Antioxidant Supplements and Mortality. Vol. 298 No.

Reply

We have read the letter from Hickey et al with interest. Overall, we disagree that our statistical analyses are inappropriate. As described in our review, our analyses were planned in advance. We do not understand why our assumptions are ‘biologically unsound’. We employ methodology of reviewing systematically all randomised trials we could identify. This is the soundest scientific method incorporating biological knowledge regarding interpretation of beneficial and harmful effects of interventions. We used the Cochrane well‐established procedure for selecting trials for inclusion in our review and no one can accuse us of performing a biased selection of trials. The selection of statistical analyses was not post hoc, but ac‐cording to our protocol and to the standard methodology within the Cochrane Hepato‐Biliary Group. 
 
 We know that our conclusions have stirred a debate and have received several comments from people who disagree with our approach and suggest a number of different subgroup analyses, sensitivity analyses, and other post‐hoc analyses to test the overall result. However, since systematic reviews with meta‐analyses are observational, although based on data that were originally gathered in a prospective and controlled manner, we believe that it is crucial to maintain a rigorous approach. Therefore, we have not changed our analytical strategy, but will of course consider all relevant comments in future reviews. Below please also see our point‐to‐point reply to the comments.

1. The selection of trials was based on criteria specified in our protocol and described in the methods section of our review. None of the authors of the review have participated in clinical trials that were excluded or included in the present review. We were of course familiar with some of the literature at the protocol stage. There were no financial, academic, or personal interests on the side of the authors that may be construed as a conflict of interest. Furthermore, to the best of our knowledge there is no clear evidence demonstrating the negative effects of selecting trials in a non‐blinded manner. This is what is usually conducted in Cochrane systematic reviews. And where are – by the way ‐ the trials that we wrongly included and excluded? 
 
 When results like our present become published, many people may react based on different backgrounds. Hickey and co‐authors accuse us for not having responded to a previous critical objection published by them in JOM, ie, Journal of Orthomolecular Medicine. First, we do not have access to this journal. Second, none of the authors have sent us their objections. Third, Hickey and co‐authors should know that the usual academic procedure is to submit such objections to JAMA, which in 2007 published an abbreviated version of our review. Had they done so, we would of course have responded to any of their objections.

2. We have assessed the evidence regarding the primary and secondary preventive effects of antioxidant supplements with traditional meta‐analytic methodology of randomised clinical trials. Randomised clinical trials – and especially those having low risk of bias – are to be found at the top of the evidence hierarchy. We did not employ Baysian meta‐analyses. That Hickey and co‐authors might have had prior probabilities being more positive towards antioxidant supplements than a neutral prior could be due to them placing too much confidence in results of observational studies and in basic research findings from in vivo and in vitro studies. The literature is loaded with examples where evidence from randomised trials does not concur with evidence from lower levels of the evidence hierarchy. That antioxidant supplements could be another of these examples does not come as a surprise to us.

3. Our 2004 Cochrane Hepato‐Biliary Group systematic review on antioxidant supplements including 14 randomised trials, among which only 9 trials provided data on overall mortality, led us to observe an increased mortality in participants on antioxidant supplements. At that time we were informed in an Editorial accompanying our paper in The Lancet that this conclusion could be wrong due to the fact that we had only included the randomised trials that looked on antioxidant supplement prevention of gastrointestinal cancers. Our present Cochrane Hepato‐Biliary Group systematic review on antioxidant supplements is a response to the request for a broader meta‐analysis, including all randomised preventive trials on antioxidant supplements that report mortality. We do, therefore, agree with Hickey and co‐authors that the results of our previous review influenced our present review, now including 67 randomised trials. It is also correct that our analyses excluding selenium trials can be seen as a post‐hoc decision. However, it should be seen as a post‐hoc decision following our findings in our 2004 Cochrane Hepato‐Biliary Group systematic review on antioxidant supplements. It is not a post‐hoc decision taken following the analyses of data in our present review. We think this distinction is of central importance to the inferences drawn from our analyses. So, when we write: “The sensitivity analysis removing selenium trials from our analysis to evaluate their influence on our conclusions was therefore not a post hoc decision”, we should have added perhaps “taken after the analyses of trials in the present review”. Of course, we let us informed of our previous results. Isn’t this the whole meaning of doing research? 
 
 In retrospect, we are not so sure that we did a fair mix of antioxidants by including selenium. We feel that it is fair science to take this understanding into consideration, now that we have looked at a larger group of antioxidant trials. If we do not learn from previous mistakes, where would we end up? 
 
 Hickey and co‐authors do not consider vitamin A as an antioxidant. We refer the reader to our discussion on that topic in our review. 
 
 Hickey and co‐authors point to the fact that there is an overlap between our primary systematic review on antioxidant supplements for prevention of cancer and our present review on antioxidants on prevention of mortality. This overlap is not at all surprising or wrong. We would have been accused of slicing up the evidence if we had excluded previously reviewed trials that contained information on mortality in our present review. 
 
 4. Among the 67 included trials, only two trials (Gililan 1977; Prince 2003) are of cross‐over design. We did not include data from the second period of cross‐over trials to avoid mixing acute effects with more protracted effects (ie, carry over effects). This could have biased our analyses. The exclusion of the second phase from cross‐over trials is an unlikely cause of bias. First, the vast majority of the cross‐over trials were trials with relatively short duration and, therefore, not very likely to inform us on mortality data. Second, as antioxidants seem to increase mortality, the inclusion of the second phase of cross‐over trials would have risked our results to become biased towards no difference (when in fact a difference exists in reality). Third, the approach we used on cross‐over trials is the approach selected in the majority of Cochrane reviews including cross‐over trials (Lathyris DN, Trikalinos TA, Ioannidis JP. Evidence from crossover trials: empirical evaluation and comparison against parallel arm trials. Int J Epidemiol. 2007;36(2):422‐30).

5. The random‐effects model analyses provided a more conservative estimate of the effects of antioxidants than the fixed‐effect model analyses, based on the fact that the former analysis puts more weight on small trials (which are more often biased) compared to the latter analyses, which weight more the results from larger trials. Such larger trials are considered trustworthier both considering the risk for systematic error and risk for random error. 
 
 When we look at the analyses based on the 47 low‐bias risk trials, we would like to draw attention to the fact that our random‐effects model and our fixed‐effect model analyses fully concur: RR 1.05, 95% CI 1.02 to 1.08, P = 0.003 in the random‐effects model compared to RR 1.05, 95% CI 1.03 to 1.08, P = 0.00001 in the fixed‐effect model. In these analyses, on which we place our largest confidence, we find no substantial heterogeneity (I2 = 7.5%). So, we do not understand the accusation that we ignored heterogeneity and that we did not see increased mortality in the antioxidant supplemented group by using the random‐effects model. 
 
 6. We did not exclude any analyses. We agree that we conducted a number of analyses and this may make it difficult to evaluate all of our findings. We also agree that meta‐analyses as presently conducted both within and outside The Cochrane Collaboration run the risk of reaching statistically significant results due to repeated testing of accumulating data in cumulative meta‐analyses (Wetterslev J, Thorlund K, Brok J, Gluud C. Trial sequential analysis may establish when firm evidence is reached in cumulative meta‐analysis. J Clin Epidemiol 2008;61:64‐75). 
 
 In response to this criticism, we would like to draw attention to the following. First, the P‐values we observed were generally small and unlikely to be affected by, eg, Bonferroni correction. Second, we observed a detrimental effect of antioxidant supplements. It is more likely that there are unpublished trial results with detrimental data than unpublished trials with positive results. Therefore, significant negative findings in meta‐analyses should be considered firmer evidence than if we were dealing with positive observations. This is due to the strong publication bias that appears to affect most areas of clinical research. We are in the process of analysing our data with trial sequential analyses. 
 
 7. We agree that subgroup analyses may open for biased results. The grouping of trials according to bias‐risk was planned in our protocol. The subgroup analyses excluding trials with potential confounders were conducted to present as fair comparisons between antioxidant supplements and placebos as possible – running the risk of loosing the statistical power that meta‐analysis introduces. From these analyses, we got the clear impression that these ‘ fairest’ comparisons only substantiated the detrimental effect of antioxidant supplements observed in our primary analyses (relative risk of death due to antioxidant supplements after exclusion of all trials with potential confounding 1.16, 95% CI 1.09 to 1.23, P < 0.00001 without significant heterogeneity (I2 = 0.0%)). This is equal to an increased mortality caused by antioxidant supplements of 16%. 
 
 8. Hickey and co‐authors are correct. The intervention effects regarding selenium is wrongly quoted in the abstract: RR 0.91, 95% CI 0.76 to 1.09 should become RR 0.90, 95% CI 0.80 to 1.01, as we correctly report in the results section. We have now amended this mistake. 
 
 We do not advice to base conclusions that include high‐bias risk trials. Such results are likely to be biased. 
 
 9. Contrary to what Hickey and co‐authors seem to assume we have put much effort into weighting both systematic errors and random errors in our reviews. It is through putting focus on both types of errors that we obtain clinical research results that are internally valid and can be used for discussing external validity. Hickey and co‐authors note that both the ATBC trial and the CARET trial carry a large weight. We notice that having these trials included does not cause substantial heterogeneity (I2 = 7.5%), and both trials included almost 50 000 of the about 181 000 participants in the low‐bias risk trials. So no wonder the two trials carry a large weight. In this post hoc decision‐making process, should these two trials be excluded due to the smokers and asbestos workers participating in them or should they stay because they both fulfilled our a priory‐defined inclusion criteria? 
 
 To elucidate the impact of the two trials on our findings, we excluded them from the analysis of the 47 low‐bias risk trials. Comparing the intervention effect of antioxidant supplements in the remaining 45 randomised trials did not show any significant difference compared to the meta‐analysis of the effect in the ATBC and CARET trials (test of interaction, z = ‐1.44, P = 0.15). When we excluded the CARET trial from the 19 trials without any potential confounders and compared the remaining 18 trials to the results of the CARET trial, we did not observe significant differences (test of interaction, z = ‐0.22, P = 0.83). 
 
 So, we do not think our analyses are biased by any prejudice, but as always regarding bias: others may be better to judge than oneself. In a similar vein, we would like to draw the readers’ attention to the previous publications of Hickey and co‐authors as well as the institutions they work. These pieces of information make us ask: ‘What kind of bias may they have?’

Contributors

Christian Gluud, 
 Lise Lotte Gluud, 
 Dimitrinka Nikolova, 
 Rosanna Simonetti, 
 Goran Bjelakovic.

Subjective, selective, and biased, 5 June 2008

Summary

Subjective, selective, and biased 
 Gert Schuitemakeri, Bo Jonssonii, Stephen Lawsoniii, Steve Hickeyiv, Len Noriegaiv, Hilary Roberts, Damien DowningviInternational Society for Orthomolecular Medicine, Ortho Institute, Toronto, Canada. 
 iiDepartment of Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden. 
 iiiLinus Pauling Institute, Oregon State University, Corvallis, OR, USA. 
 ivFaculty of Computing Engineering and Technology, Staffordshire University, Beaconside, Stafford, England. 
 v British Society for Ecological Medicine, England. 
 
 The review on antioxidant supplements and mortality by Bjelacovic et al (2008) essentially is a repeated publication of a review that has been highly criticized. The criticisms cover both the background biology and the statistical methods employed. 
 
 Main conclusions 
 The report’s main conclusion, “we found no evidence to support antioxidant supplements for primary or secondary prevention”, is not derived from the nature of the study, which was on mortality. The sole outcome measure in this review was all‐cause mortality. 
 
 The report’s secondary finding “vitamin A, beta‐carotene, and vitamin E may increase mortality.” is in the author’s conclusions. This result was only produced by statistical manipulation when the effects of selenium were excluded, and the number of statistical tests was not reported. 
 
 The report’s third finding, “future randomized trials could evaluate the potential effects of vitamin C and selenium for primary and secondary prevention”, is inappropriate, given the nature of this study of mortality. The additional statement “such trials should be closely monitored for potential harmful effects” implies bias and a lack of appreciation that such factors are normally monitored in well‐designed trials. 
 
 The report’s final main finding, “antioxidant supplements need to be considered medicinal products and should undergo sufficient evaluation before marketing”, is a political statement that is not justified by the scientific content of this review. 
 
 Statistical problems 
 1. The authors refer to the previous publication of their review but ignore the detailed scientific and methodological criticisms that it attracted.1,2 
 
 2. A meta‐analysis of mortality should not be used as a vehicle to promote a controversial viewpoint on the role of nutrition in primary or secondary prevention of disease. Moreover, the authors of such studies should be required to answer the existing major criticisms in their work before producing a repeated publication. 
 
 The study pooled study data from both the sick and the healthy and included several other interventions that increased the variability in the data. This increased variability could hide positive effects. The sick had a variety of conditions each of which needed to be considered separately in the design. Pooling these groups is not biologically valid because of the different sources of variation in the data. Pooling sick and healthy groups confuses nutrition and pharmacology. 
 
 3. Studies with no deaths were excluded, 405 studies showed no deaths compared with the 67 included studies, and relative statistics were reported. Any inference about the role of supplements in causing such deaths is unwarranted; there may have been deaths that had no relation to supplement use. There was no valid calculation to demonstrate that gross bias did not result. In the classic book “How to Lie with Statistics”, Darrel Huff used relative measure to show how statistics can mislead.3

4. The report of the “sensitivity analysis” involving 14 studies with no deaths to show that excluding such studies had no effect was flawed. The studies were not listed. Similarly, the reports of the analysis with one death were insufficiently well described for the reader to evaluate their meaning. 
 
 5. "Participants lost to follow‐up were considered survivors." There is no justification for this assumption, no estimate of the proportions in each of the groups for each statistical test employed, and no calculation of its impact on the results. 
 
 6. The selection of trials defined as “high bias” or “low bias”, with inadequate criteria, gives differing results. Since the selection process was not blind, these results are unreliable, since a greater bias than that controlled could arise from the subjective selection. 
 
 7. Two large trials with positive results (GISSI and NIT1) were allocated to the high‐risk group, the reasons are not clear and may be inappropriate. 
 
 Pharmacology and nutrition 
 1. The use of placebo controls in study selection is inappropriate and poor science, as placebo effects would not cause or prevent death, which is a definitive and objective outcome.4 
 
 2. Vitamin A is not a dietary antioxidant. The author’s suggestion that vitamin A has antioxidant activity is spurious, as numerous dietary substances have some redox activity. 
 
 3. Dietary antioxidants have complex redox mechanisms and it is oversimplistic to assume that consumption of a particular dose or molecular form of an “antioxidant” will produce a physiological antioxidant effect.5 
 
 4. The antioxidant supplements studied were a small subset of available dietary supplements Several redox active substances, such as rutin, coenzyme Q10, and iron, used in the included studies were not addressed in the analysis. 
 
 5. The doses of Vitamin C studied were too small and too infrequent to be effective.6 Taken alone, this point invalidates the vitamin C element of the study. Claims for the effects of vitamin C relate to doses 10‐100 times larger than those in the Bjelakovic paper. 
 
 6. The paper fails to distinguish between different forms of selenium, such as sodium selenite or methylselenocysteine, which have different pharmacological and redox effects. Without a description of the different forms, the results are unclear and confusing. 
 
 7. The term “vitamin E”, as used in the paper, is vague. Vitamin E is a generic term for members of two families of molecules, the tocopherols and the tocotrienols. In addition, several other lipid‐soluble antioxidants show vitamin E activity. The tocopherols consist of a set of four molecules, alpha‐, beta‐, delta‐, and gamma‐tocopherols, and the tocotrienols are similarly grouped. These molecules are subdivided further, in terms of molecular configuration. Taking alpha‐tocopherol as an example, the naturally occurring RRR‐alpha‐tocopherol is usually called d‐alpha‐tocopherol. Synthetic “dl‐alpha‐tocopherol” consists of roughly equal amounts of the eight possible stereoisomers (RRR, RRS, RSR, RSS, SSS, SSR, SRS and SRR) and may contain several additional unnatural molecules. Basic pharmacology indicates that each of these molecules has specific effects in the body and thus the indiscriminate results for “vitamin E” are misleading.

8. The authors’ suggestion that the effect of antioxidants is one of the most adequately researched areas indicates a lack of knowledge of fundamental questions in redox biology and medicine.

9. Trials that included children and pregnant women were excluded from the study “since they may be in need of certain antioxidant supplements”. This introduces bias against antioxidants.

10. The authors claim that unpublished studies of antioxidants are more likely to be negative than positive. They provide no direct or specific evidence to support this claim. The equally valid counter argument is that some commercial sources of funding would prefer not to publish trials favouring antioxidants to profitable drugs. 
 
 Conclusions 
 As Sir Bradford Hill pointed out in his landmark “rules” paper,7 epidemiology should rigorously conform to the constraints of basic science and physiology. This study by Bjelakovic is a particularly confused application of statistical methods, as it involves a varied intake of antioxidant supplements, other nutrients, and drugs, in heterogeneous populations of both sick and healthy people. With such diversity, the meaning of results is unclear and the study’s conclusions have little validity. 
 
 The authors may wish to return to their analysis to provide sufficient information for an objective assessment of their results to be made. Without such a re‐analysis, this review may be considered biased. 
 
 1 Hickey S. Roberts H.J. Noriega L.A. (2007) Poor methodology in meta‐analysis of vitamins, JOM, 22(1), 8–10. 
 2 Albanes D. (2007) Antioxidant Supplements and Mortality. JAMA, 298(4), 402‐403. 
 3 Huff D. (1993) How to Lie With Statistics, W. W. Norton & Company. 
 4 Hróbjartsson A, Gøtzsche PC. Placebo interventions for all clinical conditions. Cochrane Database of Systematic Reviews 2003, Issue 1. Art. No.: CD003974. 
 5 Block G. Jensen C.D. Morrow J.D. Holland N. Norkusc E.P. Milneb G.L. Hudesa M. 
 Dalvia T.B. Crawford P.B. Fung E.B. Schumacherd L. Harmatz P. (2008) The effect of vitamins C and E on biomarkers of oxidative stress depends on baseline level, Free Radical Biology and Medicine, doi:10.1016/j.freeradbiomed.2008.04.005. 
 6 S. Hickey H.J. Roberts and R.F. Cathcart (2005) Dynamic flow, JOM, 20(4), 237‐244. 
 7 Hill A.B. (1965) The environment and disease: Association or causation? Proc Roy Soc Med, 58, 295‐300.

Reply

As stated in our review, an abbreviated version of the review was published in JAMA (Bjelakovic G, Nikolova D, Gluud LL, Simonetti RG, Gluud C. Mortality in randomized trials of antioxidant supplements for primary and secondary prevention: Systematic review and meta‐analysis. JAMA. 2007;297(8):842‐857). We have previously answered to all critical comments raised to that paper (Gluud LL, Bjelakovic G, Nikolova D, Simonetti RG, Gluud C. Antioxidant supplements and mortality—Reply. JAMA. 2007;298(4):402‐403).

On 'Main conclusions'We have responded to major parts of these allegations in the letter by Hickey et al above, and we refer readers to those responses. 
 
 We are well aware of the correct way to monitor trials dealing with potential harmful interventions. The fact is, however, that the vast majority of trials are insufficiently monitored. For example, only 27% of the clinical trials (470) used a data monitoring committee and only 7% (116) reported some form of interim analysis out of 1772 randomised trials published in eight major journals during 2000 to 2005 (Tharmanathan P, Calvert M, Hampton J, Freemantle N. The use of interim data and Data Monitoring Committee recommendations in randomized controlled trial reports: frequency, implications and potential sources of bias. BMC Med Res Methodol. 2008;8:12). 
 
 We agree with Schuitemaker and co‐authors that our request for more proper regulatory overview is urgently needed concerning antioxidant supplements, and that our statement may be seen as a political statement. We expressed this with the hope that some politicians and regulatory authorities will act as brave individuals and not servants of the industry.

On 'Statistical problems' 
 1. We have responded to the criticism raised in JAMA (Gluud LL, Bjelakovic G, Nikolova D, Simonetti RG, Gluud C. Antioxidant supplements and mortality—Reply. JAMA. 2007;298(4):402‐403). This is the normal and usual academic procedure. Schuitemaker and co‐authors accuse us for not having responded to a previous critical objection published by them in JOM, ie, Journal of Orthomolecular Medicine. First, we do not have access to this journal. Second, none of the authors have sent us their objections. Third, Schuitemaker and co‐authors should know that the usual academic procedure is to submit such objections to JAMA, which in 2007 published an abbreviated version of our review. Had they done so, we would, of course, have responded to any of their objections. 
 
 2. We produced the abbreviated version of our review for the JAMA and the long version for The Cochrane Library more or less in parallel. The short version is about 15 printed pages long, and had a much shorter review process. The long version is about 200 printed pages long and had a much longer and elaborate review process. That Cochrane Reviews are published both as electronic versions in The Cochrane Library with sister publications in paper journals are an often encountered procedure. 
 
 It was part of our protocol that we would include both healthy participants and patients without active diseases. The effect of antioxidants did not differ significantly between the two subgroups. 
 
 As every trial is accounted for and displayed we do not understand the accusation that we are hiding positive effects. 
 
 3. We have dealt with this issue extensively in our review. Trials without deaths were excluded. First, they provide very little information. Second, most of these trials were not ‘preventive trials’, but rather explanatory trials assessing the impact of antioxidant supplements on potential surrogate outcome measures. Third, we assessed the influence of trials with zero events in the treatment or control group by re‐calculating the random‐effects meta‐analyses with 0.5, 0.05, and 0.005 continuity corrections. Fourth, we also performed additional meta‐analyses including one large hypothetical trial with one event in the treatment and control group and a sample size corresponding to the total number of participants in the 405 zero events trials. All these analyses confirmed the results of our primary analyses.

4. We have dealt with this issue extensively in our review as well as above. We refer the reader to these for further explanation. We do not know what Schuitemaker and co‐authors refer to when they speak of “sensitivity analysis involving 14 studies”. 
 
 5. Schuitemaker and co‐authors point to the fact that we considered any drop‐outs as survivors. First, all low‐bias risk trials had adequate reporting of follow‐up. The details are reported in Table of included trials. Here, Schuitemaker and co‐authors can see trial for trial how well follow‐up was reported. All low‐bias risk trials had adequate reporting of follow‐up. Second, the trials having losses to follow up usually had few losses. Third, the outcome in question was all‐cause mortality, an outcome that could usually be determined in the countries in which the trials were performed. Fourth, we had chosen the option to consider drop‐outs as survivors at the protocol stage – which is among the options one may choose according to 16.1.2 of the Cochrane Collaboration Handbook (Higgins JPT, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions 4.2.5 [updated May 2005]. In: The Cochrane Library, Issue 3, 2005. Chichester, UK: John Wiley & Sons, Ltd.). Considering that antioxidant supplements seem to increase mortality, we think we might have biased our analyses for the benefit of antioxidant supplements. 
 
 We will acknowledge that missing data may bias meta‐analyses and when updating this review we shall examine this aspect in greater detail. 
 
 6. We think Schuitemaker and co‐authors erred on these points and recommend them to read the Cochrane Collaboration Handbook (Higgins JPT, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions 4.2.5 [updated May 2005]. In: The Cochrane Library, Issue 3, 2005. Chichester, UK: John Wiley & Sons, Ltd.).

7. Again we think Schuitemaker and co‐authors erred on these points and recommend them to read our criteria for including a trial among the low‐bias risk trials as well as the Table describing the included studies. Here, we describe that GISSI was not placebo controlled (ie, there was no intervention in the control group). Therefore, this trial was considered having a high risk of bias. In the Table describing the included studies we also describe that NIT1 had inadequate follow‐up as losses to follow‐up were not reported. Therefore, this trial was considered a high risk of bias trial. 
 
 On 'Pharmacology and nutrition' 
 1. We think Schuitemaker and co‐authors erred on these points and re‐commend that they re‐read Hróbjartsson and Gøtzsche’s article on placebo. Placebo is not given to cause or prevent death. Placebo is given to blind participants, investigators, and assessors to avoid reporting bias, collateral intervention bias, and outcome assessment bias. 
 
 2. Schuitemaker and co‐authors do not consider vitamin A as an antioxidant. We refer the reader to our discussion on that topic in our review. 
 
 3. We agree with Schuitemaker and co‐authors that the antioxidant field has been influenced by a measure of naivety. We think their criticism should go to the investigators and companies behind the conducted trials – we are only reviewing and meta‐analysing the evidence that we are able to find. 
 
 4. We agree. We only looked at a handful of antioxidant supplements. We still think that we assessed some of the more commonly used antioxidants, which can be witnessed by the large number of randomised trials that we identified. Meta‐analyses cannot cover all interventions. We strongly support further systematic reviews of other antioxidant supplements.

5. But claims are not valid proof.

6. We may agree. But again, we followed our protocol to look at selenium irrespective of form.

7. We may agree. But again, we followed our protocol to look at ‘vitamin E’ irrespective of form.

8. We may obtain consensus when we explain that what we meant was that this area was one of the best researched areas when we take into account the very large number of low‐bias risk trials with large participant groups. Hereby both systematic error and random error are contained, and internal validity maximised. Where else does one find 47 trials offering such high internal validity?

9. Not at all. We are not dealing with children, pregnant women, or patients with active disease because some of these groups may be in need of one or more antioxidant supplements.

10. The problem with publication bias has been known for more than 50 years. This bias usually affects trials with neutral and harmful findings (Dickersin K, Rennie D. Registering clinical trials. JAMA. 2003;290(4):516‐23). Now Schuitemaker and co‐authors suggest that the pharmaceutical industry has withheld positive studies showing benefits of antioxidant supplements. We do not think this is very likely. First, the industry supporting the included trials had produced a median of 6 publications per trial (range 1 to 44 publications per trial). This is an extraordinary marketing achievement, considering that randomised trials are usually only published twice (Gluud C, Nikolova D. Likely country of origin in publications on randomised controlled trials and controlled clinical trials during the last 60 years. Trials. 2007;8:7). Second, the same industry has been sued and ordered to pay very large fines due to cartel price fixing (and it has been able to pay) (Connor JM. Global Price Fixing. Second Ed., Springer‐Verlag, Berlin Heidelberg, 2007). Witnessing the exorbitant large fines, we are quite confident that this industry has not had a net loss on these products. But we must admit, we do not know. 
 
 On 'Conclusions' 
 The recommendations that Sir Bradford Hill outlined in 1965 – before the introduction of systematic reviews – did not say that basic science and physiology should trump systematic reviews of low‐bias risk trials. Although basic science and physiology should be studied intensely, it may never make us ignore empirical research results. 
 
 As stated above, we intend to analyse our data from different aspects in future updates to try to accommodate the concerns that have been raised.

Contributors

Christian Gluud, 
 Lise Lotte Gluud, 
 Dimitrinka Nikolova, 
 Rosanna Simonetti, 
 Goran Bjelakovic.

Antioxidant supplements for prevention of mortality in healthy participants and patients with various diseases, 14 September 2008

Summary

Feedback by Enrico Magosso*, Kah Hay Yuen*, Yogheswaran Gopalan* *School of Pharmaceutical Sciences Universiti Sains Malaysia, 11800 Penang Malaysia 
 correspondence to: magosso.fd08@student.usm.my

It is indeed impressive the authors have put together this review involving over 230,000 patients in 67 trials. The antioxidants considered in the trials were vitamin A, vitamin E (mostly as alpha‐tocopherol), beta‐carotene, vitamin C and selenium, used either alone or in combination. 
 
 However, this range of antioxidants has to be considered as extremely limited compared to those found in nature and other natural compounds with antioxidant properties are increasingly being described. For example vitamin E is the generic name given to a family of 8 homologue compounds comprising alpha, beta‐, gamma‐ and delta‐tocopherols as well as alpha‐, beta‐, gamma‐ and delta‐tocotrienols. These homologues have been shown to exert individual functions (Brigelius‐Flohè and Traber, 1999; McIntyre et al., 2000; Schaffer et al., 2004). Tocopherols and tocotrienols share similar structural features of the chromanol ring or 'head' and are similarly named as alpha‐, beta‐, gamma‐ and delta‐ depending on the number and position of the methyl groups attached to the 'head'. The main chemical difference between tocopherols and tocotrienols lies in the phytyl chain or 'tail', saturated in the former and unsaturated in the latter. Natural alpha‐tocopherol occurs only as d‐ (or RRR‐) isomer, while synthetic alpha‐tocopherol (that is derived from petroleum) is the d,l‐racemic mixture. Tocotrienols are of natural origin and are exclusively d‐isomers. The majority of investigations have used the form of vitamin E that is alpha‐tocopherol (mostly of synthetic origin) to the extent that alpha‐tocopherol and vitamin E became synonymous.

Sen and co‐authors have highlighted the tangible differences, in efficacy as well as in toxicity, between tocopherols and tocotrienols. The following passage is taken from Sen et al. (2006): 'An expanding body of evidence support that members of the vitamin E family are functionally unique. In recognition of this fact, title claims in manuscripts should be limited to the specific form of vitamin E studied. For example, evidence for toxicity of a specific form of tocopherol in excess may not be used to conclude that high‐dosage "vitamin E" supplementation may increase all‐cause mortality. Such conclusion incorrectly implies that tocotrienols are toxic as well under conditions where tocotrienols were not even considered.'

Similar arguments apply to beta‐carotene, which is a single homologue of a range of about 600 carotenoids found in nature, 50 of which exert the role of vitamin A precursors and occurring as cis/trans racemic mixture at a variable ratio (Schieber and Carle, 2005; Krinsky and Johnson, 2005). Lyn (2000) in his review suggested 'the efficient uptake of synthetic all‐trans beta‐carotene [?] appears to make the synthetic form more desirable for effective absorption. But the tendency of synthetic beta‐carotene to alter normal serum trans/cis ratios in favor of the trans‐isomer may not be a beneficial effect' and that 'the consequences of using all‐trans synthetic beta‐carotenes are at this point unknown'. It has been suggested the negative outcome in clinical trials involving beta‐carotene might be ascribed to the use of the purified, synthetic form (Ben‐Amotz and Levy, 1996; Lyn, 2000). 
 In the review entitled 'The use of antioxidant therapies during chemotherapy', Drisko and co‐authors (2003) highlighted the importance of natural mixed carotenoids, suggesting that 'the use of synthetic beta‐carotene as a single agent rather than natural mixed carotenoids may actually promote cancer formation'. 
 
 In all but a handful of studies considered in this review synthetic alpha‐tocopherol and synthetic beta‐carotene were used. 
 In addition to many examples in which different isomers of the same compound present different level of activity and toxicity, the FDA does not register new pharmaceuticals without chiral definition (FDA, 1992; FDA, 1995; FDA 1997). 
 
 We believe certain aspects of the review should consider the following to reflect objectively several facts: 
 1) The origin of the antioxidants, either synthetic or natural, was not mentioned with regard to the included studies. Thus assuming that there are not biological differences between the two sources. 
 2) Tocotrienols were not present in any of the preparations administered in the studies considered, but the authors did not differentiate them from the generic name of vitamin E or alpha‐tocopherol and thus shared the same detrimental effect on survival rate. 
 3) Since the authors' conclusion as currently stated all but prohibit further antioxidant trials it will be helpful to perform a subgroup analysis by natural or synthetic origin of the antioxidants administered and the range of antioxidants to which the conclusion is applicable needs to be stated. 
 
 References: 
 Ben‐Amotz A, Levy Y (1996). Bioavailability of a natural isomer mixture compared with synthetic all‐trans β‐carotene in human serum. Am.J.Cl. Nutr. 63:729‐734 
 Brigelius‐Floh? R, Traber MG (1999). Vitamin E: function and metabolism. FASEB J. 13: 1145‐1155 
 Drisko JA, Chapman J, Hunter VJ (2003). The use of antioxidant therapies during chemotherapy. Gynec. Onc. 88: 434‐439 
 Food and Drug Administration (1992). 
 http://www.fda.gov/cder/guidance/stereo.htm 
 Food and Drug Administration (1995). http://www.fda.gov/cder/guidance/phase1.pdf 
 Food and Drug Administration (1997). http://www.fda.gov/cder/guidance/old038fn.pdf 
 Krinsky NI, Johnson EJ (2005). Carotenoid actions and their relation to health and disease. Molecular Asp. Med. 26: 459‐516 
 Lyn P (2000). Beta‐Carotene: The Controversy Continues. Altern. Med. Rev. 5 (6): 530‐545 
 McIntyre B, Briski KP, Gapor A, Sylvester PW (2000). Antiproliferative and apoptotic effects of tocopherols and tocotrienols on preneoplastic and neoplastic mouse mammary epithelial cells. P.S.E.B.M. 224: 292‐301 
 Schaffer S, Moller WE, Eckert GP (2005). Tocotrienols: constitutional effects in aging and disease. J.Nutr. 135: 151‐154 
 Schieber A, Carle R (2005). Occurrence of carotenoids cis‐isomers in food: Technological, analytical and nutritional implications. Trends Food Sc.Tech. 16: 416‐422 
 Sen CK, Khanna S, Roy S (2006). Tocotrienols: Vitamin E beyond tocopherols. Life Sci. 78 (18): 2088‐98 
 
 Submitter has modified conflict of interest statement: 
 EM, KHY, YG research on tocotrienols has been funded by Malaysian Palm Oil Board and supported by pharmaceutical industry.

Reply

Magosso, Yuen, and Gopalan wrote: 
 It is indeed impressive the authors have put together this review involving over 230,000 patients in 67 trials. The antioxidants considered in the trials were vitamin A, vitamin E (mostly as alpha‐tocopherol), beta‐carotene, vitamin C and selenium, used either alone or in combination. 
 
 ‐ We thank Magosso, Yuen, and Gopalan for these most positive comments.

Magosso, Yuen, and Gopalan wrote: 
 However, this range of antioxidants has to be considered as extremely limited compared to those found in nature and other natural compounds with antioxidant properties are increasingly being described. For example vitamin E is the generic name given to a family of 8 homologue compounds comprising alpha, beta‐, gamma‐ and delta‐tocopherols as well as alpha‐, beta‐, gamma‐ and delta‐tocotrienols. These homologues have been shown to exert individual functions (Brigelius‐Flohè and Traber, 1999; McIntyre et al., 2000; Schaffer et al., 2004). Tocopherols and tocotrienols share similar structural features of the chromanol ring or 'head' and are similarly named as alpha‐, beta‐, gamma‐ and delta‐ depending on the number and position of the methyl groups attached to the 'head'. The main chemical difference between tocopherols and tocotrienols lies in the phytyl chain or 'tail', saturated in the former and unsaturated in the latter. Natural alpha‐tocopherol occurs only as d‐ (or RRR‐) isomer, while synthetic alpha‐tocopherol (that is derived from petroleum) is the d,l‐racemic mixture. Tocotrienols are of natural origin and are exclusively d‐isomers. The majority of investigations have used the form of vitamin E that is alpha‐tocopherol (mostly of synthetic origin) to the extent that alpha‐tocopherol and vitamin E became synonymous. Sen and co‐authors have highlighted the tangible differences, in efficacy as well as in toxicity, between tocopherols and tocotrienols. The following passage is taken from Sen et al. (2006): 'An expanding body of evidence support that members of the vitamin E family are functionally unique. In recognition of this fact, title claims in manuscripts should be limited to the specific form of vitamin E studied. For example, evidence for toxicity of a specific form of tocopherol in excess may not be used to conclude that high‐dosage "vitamin E" supplementation may increase all‐cause mortality. Such conclusion incorrectly implies that tocotrienols are toxic as well under conditions where tocotrienols were not even considered.' 
 
 ‐ We are aware of the facts mention above. However, we could not include trials with tocotrienols, gama tocopherol, or any other form of vitamin E because such randomised trials have not been published. The majority of the trials conducted tested alpha‐tocopherol. We will in future updates try to highlight this issue.

Magosso, Yuen, and Gopalan wrote: 
 Similar arguments apply to beta‐carotene, which is a single homologue of a range of about 600 carotenoids found in nature, 50 of which exert the role of vitamin A precursors and occurring as cis/trans racemic mixture at a variable ratio (Schieber and Carle, 2005; Krinsky and Johnson, 2005). Lyn (2000) in his review suggested 'the efficient uptake of synthetic all‐trans beta‐carotene [?] appears to make the synthetic form more desirable for effective absorption. But the tendency of synthetic beta‐carotene to alter normal serum trans/cis ratios in favor of the trans‐isomer may not be a beneficial effect' and that 'the consequences of using all‐trans synthetic beta‐carotenes are at this point unknown'. It has been suggested the negative outcome in clinical trials involving beta‐carotene might be ascribed to the use of the purified, synthetic form (Ben‐Amotz and Levy, 1996; Lyn, 2000). 
 In the review entitled 'The use of antioxidant therapies during chemotherapy', Drisko and co‐authors (2003) highlighted the importance of natural mixed carotenoids, suggesting that 'the use of synthetic beta‐carotene as a single agent rather than natural mixed carotenoids may actually promote cancer formation'. 
 
 ‐ The answer is as above. We included trials with beta‐carotene that we were able to identify. We will in future updates try to highlight the raised issues. 
 
 Magosso, Yuen, and Gopalan wrote: 
 In all but a handful of studies considered in this review synthetic alpha‐tocopherol and synthetic beta‐carotene were used. 
 
 ‐ We included the trials that we were able to identify according to our protocol. Further trials and systematic reviews have to access whether there are certain benefits or harms connected to ‘synthetic’ as well as ‘natural’ vitamins. 
 
 Magosso, Yuen, and Gopalan wrote: 
 In addition to many examples in which different isomers of the same compound present different level of activity and toxicity, the FDA does not register new pharmaceuticals without chiral definition (FDA, 1992; FDA, 1995; FDA 1997). 
 
 ‐ It would be much better for the FDA and other regulatory agencies to require that dietary supplements sold to the public claiming health benefits are subjected to adequate assessment of benefits and harms before market release, similar to any other drug.

Magosso, Yuen, and Gopalan wrote: 
 We believe certain aspects of the review should consider the following to reflect objectively several facts: 
 1) The origin of the antioxidants, either synthetic or natural, was not mentioned with regard to the included studies. Thus assuming that there are not biological differences between the two sources. 
 
 ‐ We mentioned the form of antioxidant used in the table 'Characteristics of included studies'. The origin of the antioxidants was mentioned in the majority of the trials, but in some of them not.

Magosso, Yuen, and Gopalan wrote: 
 2) Tocotrienols were not present in any of the preparations administered in the studies considered, but the authors did not differentiate them from the generic name of vitamin E or alpha‐tocopherol and thus shared the same detrimental effect on survival rate. 
 
 ‐ We were not able to identify such trials. In case that at any future review updates, we identify randomised trials with tocotrienols, we will include them in our meta‐analyses. As already stated, we will address this issue in further updates.

Magosso, Yuen, and Gopalan wrote: 
 3) Since the authors' conclusion as currently stated all but prohibit further antioxidant trials it will be helpful to perform a subgroup analysis by natural or synthetic origin of the antioxidants administered and the range of antioxidants to which the conclusion is applicable needs to be stated. 
 
 ‐ If we identify enough number of trials with natural antioxidant supplements, we can perform subgroup analyses. However, we are of the opinion that we have fulfilled the main Cochrane criterion to look at the totality of evidence for the effects of a specific intervention. If you are aware of any randomised trials with ‘natural’ antioxidant supplements that is not included in our analyses, we will appreciate to receive this information from you. 
 
 A large population based randomised clinical trial (Physician Health Study II1) has recently been completed. Its authors concluded that neither vitamin E nor vitamin C supplementation reduced the risk of major cardiovascular events. A second trial (SELECT2) stopped supplementation after the Data and Safety Monitoring Committee reviewed the available data and found that selenium and vitamin E supplements failed to prevent prostate cancer. Participants taking vitamin E had a small increase in prostate cancer, while participants taking only selenium were more likely to develop diabetes. These results strongly support the findings of our review. 
 
 1. Sesso HD, Buring JE, Christen WG, Kurth T, Belanger C, MacFadyen J, Bubes V, Manson JE, Glynn RJ, Gaziano JM. Vitamins E and C in the prevention of cardiovascular disease in men: the Physicians' Health Study II randomized controlled trial. JAMA 2008;300:2123‐33. 
 2. http://www.crab.org/select/ (Assessed 19 November 2008).

Contributors

Goran Bjelakovic, 
 Christian Gluud, 
 Dimitrinka Nikolova.

Antioxidants are not identical and their effects are not uniform over the population, 24 August 2009

Summary

Antioxidants are not identical and their effects are not uniform over the population 
 Harri Hemilä, MD, PhD 
 Department of Public Health, University of Helsinki, Helsinki, Finland. e‐mail harri.hemila@helsinki.fi 

Submitter agrees with the default conflict of interest statement: I certify that I have no affiliations with or involvement in any organization or entity with a financial interest in the subject matter of my feedback.

Reply

Dr. Harri Hemilä wrote:

There are two fundamental problems with the review by Bjelakovic and coworkers. First, the authors are combining apples and oranges. Second, the authors ignore the evidence indicating that vitamin E effect is not uniform over the population.  

First, let us consider an analogy. If a researcher is interested in the effect of antibiotics on mortality caused by infections, and combines all antibiotics to a single broad category of ‘antibiotics’, and pools all ‘antibiotic trials’ together, people with basic background in clinical microbiology would consider such a project silly. Different antibiotics kill different bacteria, there is geographic and social group variation in the occurrence of pathogenic bacteria, the usage of antibiotics generates resistant strains, etc., etc. The biology of antibiotics is very complex. It is obvious that a single universal estimate for ‘antibiotic effect’ is meaningless.

Antioxidants are also a heterogeneous group. Vitamin C is water soluble, vitamin E is fat soluble, selenium is an inorganic element, beta‐carotene is not an essential nutrient, etc. Moreover, in model systems small‐molecule antioxidants are oxidized at different speeds. For example, activated neutrophils in plasma first oxidize vitamin C, whereas the oxidation of urate starts only after vitamin C has been consumed, and the level of vitamin E remains virtually unchanged [1]. From the point of view of biology, there is no basis to consider that all antioxidants are similar enough to justify the pooling of all trials with compounds belonging to the broad category of ‘antioxidants’. In a proper analysis, each antioxidant should be analyzed specifically.

‐ Response:   

We thank Dr Hemilä for the comments. We agree that ‘antioxidants’ are a heterogeneous group of substances. Vitamin E and the other antioxidant supplements may have different effects in different populations. Accordingly, we have accounted for the statistical heterogeneity in our analyses. Several potentially interesting subgroup analyses may be considered, but the included trials do not report the necessary data. When performing systematic reviews, the decision to split or lump a heterogeneous group of trials may be difficult. Splitting trials into various subgroups may provide more focused answers, but will also increase the risk of spurious results. Lumping a heterogeneous group of trials provides a less focused answer, but may increase the external validity. In the present review, we decided to focus on antioxidants supplements to provide an overall picture of the potential effects. Our results provide information that may form basis for future trials and reviews.

We have analysed antioxidant supplements combined (beta‐carotene; vitamin A; vitamin C; vitamin E; selenium) as well as individually.

Dr. Harri Hemilä wrote:

Bjelakovic and coworkers label vitamin A as an antioxidant. Halliwell and Gutteridge write in the latest edition of their monograph, that "vitamin A can do the same [scavenging some oxidants, as beta‐carotene does] but no data exist supporting such a role in vivo?" [2 p. 177]. Of course, Halliwell and Gutteridge may be wrong, but given the depth of their familiarity with the antioxidant literature, Bjelakovic and coworkers should give explicit reference(s) to reject that statement when labelling vitamin A as an antioxidant. Most biomolecules are oxidized by hydroxyl radical, but it is not reasonable to thereby label all of them as antioxidants. Halliwell and Gutteridge suggest that the definition of "antioxidant" should require that it delays, prevents or removes oxidative damage [2 p. 80‐81]. Thus, a systematic review focusing on antioxidants should consider and define what is meant by an "antioxidant". Bjelakovic writes that "there are pros and cons in the literature about vitamin A being antioxidant", which in my opinion is not a scientific argument to justify labelling vitamin A as an antioxidant. Furthermore, Bjelakovic and coworkers write in Discussion that "recent research revealed that vitamin A can cause oxidative damage to DNA". Thus, there is lack of logic when claiming that vitamin A has been shown to be a pro‐oxidant, but include it in the review restricted to antioxidants on the basis of conflicting opinions.

‐ Response:   

Based on previous evidence, we have classified vitamin A as an antioxidant supplement. The potential antioxidant and pro‐oxidant potentials of vitamin A are clarified in the discussion section of our review. We kindly refer the reader to this discussion.

Multivitamins/multiminerals (containing vitamin A) are the most commonly used dietary supplements. What is important is the effect of vitamin A on mortality irrespective of specific biological effect.  

Dr. Harri Hemilä wrote:

Analyzing trials by the specific biochemical substance that is tested would make the review much more logical. If a review focuses on, say, vitamin A and mortality, there is no need to firmly decide whether vitamin A is an antioxidant or not. The problem of defining ‘antioxidant’, and deciding whether some of them is pro‐oxidant under certain condition is avoided if the substances are analyzed by the biochemical definitions. In such an approach, antioxidant and pro‐oxidant concepts could be left to the Discussion section for giving plausible biological explanations for the observed effects.                                          

An important goal in modern biomedicine is specificity. The strength of the randomized trial is that the difference between the trial groups can be specifically attributed to the intervention that is being tested. However, when the intervention varies from a single antioxidant to the combinations of diverse antioxidants, and includes 11 trials in which “participants were supplemented with different mixtures of antioxidants as well as with vitamins and mineral without antioxidant properties”, we lose specificity because of the apples and oranges problem.

‐ Response:   

We agree that specificity is an important goal. This is why we report meta‐analyses of the selected five antioxidant supplements together (beta‐carotene; vitamin A; vitamin C; vitamin E; selenium) as well as meta‐analyses of them individually (beta‐carotene; vitamin A; vitamin C; vitamin E; selenium). Our intention by also using meta‐analyses of the individual supplements was to evade the critic of mixing ‘apples and pears’ or ‘apples and oranges’.

Dr. Harri Hemilä wrote:

In Table 6, Bjelakovic and coworkers calculate the specific effect of vitamin A (95% CI for the RR: 0.84 to 1.68; N=2406) and vitamin E (95% CI for the RR: 0.98 to 1.05; N=41341), but these specific effect estimates are hidden from the reader. In the Discussion, Bjelakovic and coworkers state that “beta‐carotene, vitamin A, and vitamin E given singly or combined with other antioxidant supplements significantly increase mortality”, which is false and misleads those readers who skip Table 6 and only look at the Discussion. The above confidence intervals show that vitamins A and E singly do no significantly increase mortality. In the Abstract, Bjelakovic does not give the specific estimates for vitamins A and E, but gives RR estimates based on studies with scores of other antioxidants including beta‐carotene, which has been known to increase mortality since the publication of the ATBC and CARET trials. A reader of the Abstract cannot figure out that the given RR‐values don't tell us anything about the specific effects of vitamins A and E. Thus, even the Abstract is misleading.

‐ Response:

As described in the methods section, we described our results of the overall, subgroup, and sensitivity analyses. We are surprised that Hemilä accuses us from hiding any information the information from the random‐effects model analyses were clearly presented inTable 5and the corresponding analyses from the fixed‐effect analyses were presented inTable 6. Now Hemilä is especially interested in certain subgroup results. Such data may be misleading as they may not have the necessary power and precision. The data we present in our abstract are those having the best power and precision and hence represent findings with most external validity. 

Second, when Bjelakovic and coworkers first published the review in JAMA, I pointed out that the effect of vitamin E on respiratory infections was heterogeneous in the large scale ATBC Study [3,4]. Vitamin E had no overall effect on the incidence of the common cold or pneumonia, but the effects were significantly modified by age and smoking [5,6]. Although heterogeneity in the effect on respiratory infections does not directly imply that the effect on mortality must be heterogeneous, such a possibility should not be dismissed. If the effect of vitamin E is heterogeneous, then a single estimate for effect can be meaningless. However, Bjelakovic ignores this issue.

Motivated by our findings on respiratory infections, we analyzed the effect of vitamin E on the mortality of ATBC participants and found strong evidence that the effect of vitamin E on total mortality was also heterogeneous [7,8]. Vitamin E had no effect on those who had low dietary vitamin C intake; however, among those who had high dietary vitamin C intake, vitamin E increased mortality in young and decreased mortality in old participants. Close to half of the participants fell to those groups in which vitamin E effect was inconsistent with the average effect of the whole study population. When the average effect of a large trial is misleading for half of the study participants, it seems obvious that calculating and presenting a single universal ‘estimate for vitamin E effect’ is an unsound approach.

Although heterogeneity in vitamin E effect on mortality does not directly imply that the effect of vitamin C and beta‐carotene must be heterogeneous, such a possibility should not be ignored. In fact, we can even turn the argument around. Given the strong evidence that vitamin E effect is heterogeneous, why should we accept such a premise that the effects of other antioxidants are uniform over the population. If we assume that the effects of antioxidants are heterogeneous, further studies should try to identify and characterize the subpopulations where the antioxidants might be beneficial, rather than calculating a fictionally accurate average effect on all people. Lack of uniformity in vitamin C effect is suggested by the interaction between vitamin E supplementation and dietary vitamin C intake. Although dietary vitamin C has a high level of correlation with other substances in fruit and vegetables, the other substances did not explain the modification of vitamin E effect in the ATBC cohort [7].

‐ Response:  

Dr Hemilä reports some interesting subgroup analyses from a randomised trial. The data suggest a possible beneficial effect of vitamin E given in combination with vitamin C seen in certain populations.  Although we cannot exclude such an effect, we are unable to analyse the question as only trial‐level data were available for our meta‐analyses. In our review, we have clarified that the effect of antioxidant supplementation might not be uniform across the population.

Bjelakovic and coworkers write in their Discussion that “our analyses had little trial heterogeneity. This increases the trustworthiness of our findings”. I cannot see any justification for such an argument. If there is a strong premise that the effect of, say, vitamin E should be uniform over the population, in such a case observing little heterogeneity is consistent with our expectations. However, as noted above, there is no basis for such a premise in general, and in the case of vitamin E it was firmly refuted [7]. The level of heterogeneity is no measure of ‘trustworthiness’.

Bjelakovic and coworkers state in their Discussion that “adoption of the random‐effects model in meta‐analysis permits extension of inferences to a broader population of studies than the fixed‐effect model does”, which is incorrect. If there is heterogeneity, we do not know to whom the calculated overall estimate applies, and this problem does not disappear by using the random‐effects model. In the random‐effects model the confidence interval is wider, but that does not help us to understand what are the characteristics that modify the effect: to whom there is effect and to whom not. When there is evidence of heterogeneity, the main focus should be on trying to understand any sources of heterogeneity that are present [9].

‐ Response:

In systematic reviews, different sources of intertrial heterogeneity may exist including clinical, methodological, and statistical heterogeneity.17 In our review, we found little evidence of intertrial heterogeneity in our meta‐analyses. In our meta‐regression analyses, the risk of bias and the type of antioxidant supplement were the only significant predictors of intertrial heterogeneity. In the trials with a low risk of bias, the antioxidant supplements significantly increased mortality (RR 1.05, 95% CI 1.02 to 1.08).18 We have discussed differences between random‐effects and fixed‐effect models of meta‐analysis in the section Discussion.3

Bjelakovic and coworkers also conclude that the effect of antioxidants is uniform over the duration of supplementation: “we found no significant effect of treatment duration on our results” (Discussion). Our analysis of the individual‐level data of the ATBC study refuted also this conclusion. In young participants who had high dietary vitamin C intake, vitamin E supplementation had no effect over the first 3.3 years, but thereafter increased mortality by 38% [7]. Adding the two different vitamin E effects significantly improved the Cox model (P=0.007). Correlation of treatment effect with the average duration of supplementation at the trial level is too crude a method to examine the time dependency of supplementation effects. In epidemiology, ‘ecological fallacy’ means thinking that relationships observed for the averages for groups necessarily hold for individuals. Thus, Bjelakovic's conclusion that treatment duration has no effect on the effect of antioxidant supplementation is an example of the ecological fallacy.

‐ Response:  

We found a number of trials assessing similar intervention regimens to the one assessed in the ATBC trial. Based on our meta‐regression analyses stratified by the intervention regimen, the duration of supplementation was not a predictor of the estimated intervention effect.

Finally, Bjelakovic and coworkers were ambitious when covering all antioxidants plus vitamin A trials. Such wide coverage requires lots of work and easily leads to errors in the extraction of data, and to the lack of time to read the papers and learn the context of the trials. As a reflection of this problem, Bjelakovic and coworkers wrote half‐a‐page erratum to their JAMA paper [10]. Nevertheless, in the 2008 version of the Cochrane review Bjelakovic still includes the Chandra 1992 trial [11] in their analysis, even though it had been shown to be fabricated several years earlier. The story should be familiar to everyone who follows the major journals [12‐15]. In the reference list, under the citation of the Chandra 1992 report, Bjelakovic cites the Lancet letter [12]. Apparently, Bjelakovic and coworkers lacked time to read the Lancet letter to see that there would have been good reasons to exclude the 1992 study from analysis.

‐ Response:

The most commonly used dietary supplements in adults are multivitamins/multiminerals.19 The daily use of vitamin A, C, and E also increased significantly during the last decades. Assessing the effects of these interventions is, therefore, important. We are aware that there may be limited bias control in the trial by Chandra and co‐workers, but had to include the trial in our overall assessment. We did not include the trial by Chandra that was retracted in 2005 (Chandra RK. Effect of vitamin and trace‐element supplementation on cognitive function in elderly subjects. Nutrition 2001;17:709‐12). The second trial by the same group, published in Lancet in 1992 was included in our meta‐analysis. Excluding this trial with 96 participants and 2 deaths in the placebo arm did not change our overall results.

Bailar criticized the meta‐analysis approach in general and gave examples of severe errors in five influential meta‐analyses [16]. In particular, Bailar criticized the 'job‐shop' approach: a group of researchers picks a topic, rushes to collect trials and pools their results, without making themselves familiar with the biology and other relevant context of the topic. Lack of considering the differences between antioxidants, labelling vitamin A as an antioxidant (while simultaneously claiming that vitamin A is a pro‐oxidant), ignoring the evidence of heterogeneity in vitamin E effect, the large number of errors in the first version of the meta‐analysis [10], the inclusion of the Chandra 1992 trial; all these indicate to me that there is a severe 'job‐shop' type of problem in Bjelakovic's review.

‐ Response:  

We included vitamin A, vitamin C, vitamin E, beta‐carotene, and selenium based on their proven antioxidant function. These antioxidants were chosen after extensive discussion of the antioxidant literature. Moreover, these antioxidants were accepted following extensive peer reviewer and editorial assessments of our Cochrane Hepato‐Biliary Group protocol published in 2003.20As this protocol covered more than cancers in the liver and biliary tract ‐ but focused on all gastrointestinal cancers ‐ we had our protocol quality assessed and approved by the Editorial Teams of The Cochrane Upper Gastrointestinal Diseases Group, The Cochrane Inflammatory Bowel Diseases Group, and The Cochrane Colorectal Cancer Diseases Group as well as The Cochrane Hepato‐Biliary Group. So quite contrary to what Dr. Hemilä seems to imagine, we find the ‘job shop’ description the least well description of the extensive process we went through before assembling and extracting any data. Our present systematic review is an extension of this 2003 protocol now focusing on all‐cause mortality.

There is evidence that not only vitamin A but also vitamin C and beta‐carotene possess pro‐oxidant function.21,22

I do not disagree with Bjelakovic and coworkers about the main conclusions. So far, there is no good evidence indicating that ordinary people would benefit from taking antioxidant supplements for the purpose of reducing mortality. This conclusion can be reached by reading the major trial reports separately, without calculating a fictional pooled antioxidant effect. In this respect, pooling of the results does not give us any additional understanding. Although the evidence of heterogeneity in the vitamin E effect on mortality is strong [7], I do not think that it justifies practical conclusions yet. Rather, the complexity encourages caution in drawing conclusions and patience in waiting for further research.

‐ Response:  

Based on the results of recently completed (PHS II; WACS)23,24 and prematurely terminated randomised trials (SELECT)25 as well as earlier meta‐analyses26,27 we do not suggest further randomised trials testing the effect of antioxidant supplementation for primary prevention or secondary prevention. It may well be that there may be subgroup of patients with active diseases that may potentially benefit from certain antioxidants but that need proper assessments in randomised clinical trials as well as systematic reviews of such trials.

Bailar commented that meta‐analysis can aid in filling in the second and third decimal places once the questions are clear but it is a poor tool for developing new concepts, new hypotheses [16]. The Bjelakovic and coworkers meta‐analysis implies that there is no justification for further research on vitamin E and mortality because the particularly narrow confidence interval (0.98 to 1.05) firmly rejects any substantial benefits. In contrast, our analysis of the ATBC Study suggests a path that should be explored: does the combination of vitamins E and C improve the health of some subpopulations of elderly males. In this respect, my conclusions significantly diverge from those of Bjelakovic. Therefore the two problems discussed above are fundamentally important.

‐ Response:

We agree that dogmas like the one that antioxidants, especially vitamin E supplementation, might be beneficial for human population, is sometimes very difficult to disaffirm.

In recent years, several major studies of vitamins and supplements have produced disappointing results. One recent example is the Selenium and Vitamin E Cancer Prevention Trial (SELECT) originally scheduled to end in 2011. SELECT was terminated in October, 2008 over a disproportionally high incidence in prostate cancer in participants in the trial who were taking vitamin E.25

References:

1. Frei B, Stocker R, Ames BN. Antioxidant defences and lipid peroxidation in human blood plasma. Proc Natl Acad Sci USA 1988; 85:9748‐52.

http://www.pnas.org/content/85/24/9748

2. Halliwell B, Gutteridge JMC. Free Radicals in Biology and Medicine, 4th ed. Oxford University Press, Oxford, 2007.

3. Bjelakovic G, Nikolova D, Gluud LL, Simonetti RG, Gluud C. Mortality in randomized trials of antioxidant supplements for primary and secondary prevention: systematic review and meta‐analysis. JAMA 2007;297:842‐57. http://dx.doi.org/10.1001/jama.297.8.842

4. Hemilä H. Antioxidant supplements and mortality. JAMA 2007;298:401. http://dx.doi.org/10.1001/jama.298.4.401‐a

5. Hemilä H, Virtamo J, Albanes D, Kaprio J. The effect of vitamin E on common cold incidence is modified by age, smoking and residential neighborhood. J Am Coll Nutr 2006;25:332‐9.

http://www.jacn.org/cgi/content/abstract/25/4/332

6. Hemilä H, Virtamo J, Albanes D, Kaprio J. Vitamin E and beta‐carotene supplementation and hospital‐treated pneumonia incidence in male smokers. Chest 2004;125:557‐65.http://dx.doi.org/10.1378/chest.125.2.557

7. Hemilä H, Kaprio J. Modification of the effect of vitamin E supplementation on the mortality of male smokers by age and dietary vitamin C. Am J Epidemiol 2009;169:946‐53.

http://dx.doi.org/10.1093/aje/kwn413

8. Hemilä H. Vitamin E is likely to affect mortality even at low doses. Clin Trials 2009;6:392‐3.

http://dx.doi.org/10.1177/1740774509340211

9. Thompson SG. Why sources of heterogeneity in meta‐analysis should be investigated. BMJ 1994;309: 1351?5. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2541868

10. Bjelakovic G. Corrections. JAMA 2008;299:765‐6. http://jama.ama‐assn.org/cgi/content/full/299/7/765

11. Chandra RK. Effect of vitamin and trace‐element supplementation on immune responses and infection in elderly subjects. Lancet 1992;340:1124‐7. http://dx.doi.org/10.1016/0140‐6736(92)93151‐C

12. Carpenter KJ, Roberts S, Sternberg S. Nutrition and immune function: a 1992 report. Lancet 2003;361:2247‐8. http://dx.doi.org/10.1016/S0140‐6736(03)13755‐5

13. White C. Three journals raise doubts on validity of Canadian studies. BMJ 2004;328:67.

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=314042

14. Smith R. Investigating the previous studies of a fraudulent author. BMJ 2005;331:288‐91.

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1181274

15. Meguid MM. Retraction [of a Chandra study based on the Lancet 1992 data]. Nutrition 2005;21:286.

http://dx.doi.org/10.1016/j.nut.2004.12.002

16. Bailar JC. The practice of meta‐analysis. J Clin Epidemiol 1995;48:149‐57. http://dx.doi.org/10.1016/0895‐4356(94)00149‐K.

17. Higgins JPT, Green S (editors). Cochrane Handbook for Systematic Reviews of Interventions Version 5.0.1 [updated September 2008]. The Cochrane Collaboration, 2008. Available from www.cochrane‐handbook.org.

18. Bjelakovic G, Nikolova D, Gluud LL, Simonetti RG, Gluud C. Antioxidant supplements for prevention of mortality in healthy participants and patients with various diseases. Cochrane Database of Systematic Reviews 2008, Issue 2. Art. No.: CD007176. DOI:10.1002/14651858.CD007176.

19. Radimer K, Bindewald B, Hughes J, Ervin B, Swanson C, Picciano MF. Dietary supplement use by US adults: data from the National Health and Nutrition Examination Survey, 1999‐2000. Am J Epidemiol 2004;160:339‐49.

20. Bjelakovic G, Nikolova D, Simonetti R. Antioxidants for preventing gastrointestinal cancers (protocol for a Cochrane Review). The Cochrane Database of Systematic Reviews 2003, Issue 2.

21. Podmore ID, Grifiths HR, Herbert KE, Mistry N, Mistry P, et al. Vitamin C exhibits pro‐oxidant properties. Nature 1998;392:559.

22. Paolini M, Abdel‐Rahman SZ, Sapone A, Pedulli GF, Perocco P, Cantelli‐Forti G, et al. Beta‐carotene: a cancer chemopreventive agent or a co‐carcinogen?. Mutation Research 2003;7719:1­­‐6.

23. Sesso HD, Buring JE, Christen WG, Kurth T, Belanger C, MacFadyen J, et al. Vitamins E and C in the prevention of cardiovascular disease in men: the Physicians' Health Study II randomized controlled trial. JAMA 2008;300:2123‐33.

24. Cook NR, Albert CM, Gaziano JM, Zaharris E, MacFadyen J, Danielson E, et al. A randomized factorial trial of vitamins C and E and beta carotene in the secondary prevention of cardiovascular events in women: results from the Women's Antioxidant Cardiovascular Study.  Arch Intern Med 2007;167:1610‐8.

25. Lippman SM, Klein EA, Goodman PJ, Lucia MS, Thompson IM, Ford LG, et al. Effect of selenium and vitamin E on risk of prostate cancer and other cancers: the Selenium and Vitamin E Cancer Prevention Trial (SELECT). JAMA 2009;301:39‐51.

26. Miller ER 3rd, Pastor‐Barriuso R, Dalal D, Riemersma RA, Appel LJ, Guallar E. Meta‐analysis: high‐dosage vitamin E supplementation may increase all‐cause mortality. Ann Intern Med 2005;142:3746.

27. Vivekananthan DP, Penn MS, Sapp SK, Hsu A, Topol EJ. Use of antioxidant vitamins for the prevention of cardiovascular disease: meta‐analysis of randomised trials. Lancet 2003;361:201723.  

Contributors

Goran Bjelakovic, 
 Christian Gluud,Lise Lotte GluudRosa Simonetti, 
 Dimitrinka Nikolova.

What's new

Date Event Description
16 January 2012 New citation required but conclusions have not changed Compared to our previous review (Bjelakovic 2008), the number of included trials in the present review is expanded with 11 new trials (16.4%) adding another 64,157 participants (27.6%). Moreover, we have obtained updated results of longer follow‐up in two large‐scale randomised trials (SIT 2006; SUVIMAX 2010Low). In spite of the addition of the 11 new trials and updated follow‐up results of two of the included trials, our results remained largely the same.
16 January 2012 New search has been performed We have now updated the review, published in 2008 (Bjelakovic 2008). The searches were extended to February 2011.

Notes

The protocol for this review was published with a title 'Antioxidants for preventing gastrointestinal cancers'. The content of this review has already been published in JAMA (Mortality in randomized trials of antioxidant supplements for primary and secondary prevention: systematic review and meta‐analysis. JAMA 2007;297(8):842‐857) with corrections appearing in JAMA 2008 (Data Errors in: Mortality in randomized trials of antioxidant supplements for primary and secondary prevention: systematic review and meta‐analysis. JAMA 2008;299(7):765‐766). The present version of the review incorporates all these corrections. Furthermore, we have realised that we included a quasi‐randomised study in our JAMA version. This study has been excluded from our Cochrane review. This has not materially changed the results.

We have now updated the review published in 2008 (Bjelakovic 2008). We extended our searches until February 2011. Compared to our previous review (Bjelakovic 2008), the number of included trials in the present review has increased with 11 new trials (16.4%) adding another 64,157 participants (27.6%). Moreover, we have obtained updated results of longer follow‐up on two large‐scale randomised trials (SIT 2006; SUVIMAX 2010Low). In spite of these expansions, our results remained largely the same.

Acknowledgements

We thank the participants who entered the trials and the investigators who conducted them. We thank authors who kindly responded to our requests for further information on the trials they were involved in. We thank Yan Gong, MD, MIH, PhD, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark, for assistance with statistical analyses and Sarah Louise Klingenberg, Cochrane Hepato‐Biliary Group, Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark, for help with paper copies of articles. We thank the Editors and peer reviewers of JAMA for helpful suggestions for improvements for the sister JAMA version of our review. We also thank JAMA for having pointed out a data extraction error, which led us to reexamine all data extraction. Errors identified during this process have been corrected in the present review. We thank Ronald L Koretz, Contact Editor of the CHBG, Kurinchi Gurusamy, Editor of the CHBG, as well as other CHBG Editors and Consumer Representatives of The CHBG for very helpful comments on the previous version of this review. This benefited the clarity of our review. The Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen University Hospital, Copenhagen, Denmark provided monetary support for the review.

Review 
 Peer reviewers: Kurinchi Gurusamy, UK; Abe Fingerhut, France; Edgar Miller, USA. 
 Contact Editor: Ronald L Koretz, USA.

Appendices

Appendix 1. Search strategies

Database Timespan Search strategy Number of references
Cochrane Central Register of Controlled Trials (CENTRAL) in The Cochrane Library Issue 1, 2011. #1 ((antioxidant* or vitamin*) and supplement*) NOT child* 3796 3456
MEDLINE (Ovid SP) 1950 to February 2011. 1. ((antioxidant* or vitamin*) and supplement*).mp. [mp=protocol supplementary concept, rare disease supplementary concept, title, original title, abstract, name of substance word, subject heading word, unique identifier]
2. limit 1 to (("young adult (19 to 24 years)" or "adult (19 to 44 years)" or "young adult and adult (19‐24 and 19‐44)" or "middle age (45 to 64 years)" or "middle aged (45 plus years)" or "all aged (65 and over)" or "aged (80 and over)") and (female or humans or male))
3. (random* or blind* or placebo* or meta‐analysis).mp. [mp=protocol supplementary concept, rare disease supplementary concept, title, original title, abstract, name of substance word, subject heading word, unique identifier]
4. 2 and 3
3388
EMBASE (Ovid SP) 1980 to February 2011. 1. ((antioxidant* or vitamin*) and supplement*).mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer]
2. limit 1 to (human and male and female and (adult <18 to 64 years> or aged <65+ years>))
3. (random* or blind* or placebo* or meta‐analysis).mp. [mp=title, abstract, subject headings, heading word, drug trade name, original title, device manufacturer, drug manufacturer]
4. 2 and 3
2124
Science Citation Index Expanded 1900 to February 2011. # 1 29,732 TS=(((antioxidant* or vitamin*) and supplement*) NOT child*)
# 2 >100,000 TS=(random* or blind* or placebo* or meta‐analysis)
# 3 6,515 #2 AND #1
6515

Data and analyses

Comparison 1. Antioxidants versus placebo or no intervention.

Outcome or subgroup title No. of studies No. of participants Statistical method Effect size
1 Mortality in trials with a low or high risk of bias 78 296707 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.98, 1.05]
1.1 Trials with low risk of bias 56 244056 Risk Ratio (M‐H, Random, 95% CI) 1.04 [1.01, 1.07]
1.2 Trials with high risk of bias 22 52651 Risk Ratio (M‐H, Random, 95% CI) 0.91 [0.85, 0.98]
2 Mortality in 76 trials with a low or high risk of bias with assigned fatalities to all of the dropouts 76 263805 Risk Ratio (M‐H, Random, 95% CI) 1.01 [0.99, 1.03]
2.1 Trials with low risk of bias 55 240738 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.99, 1.04]
2.2 Trials with high risk of bias 21 23067 Risk Ratio (M‐H, Random, 95% CI) 0.92 [0.86, 0.99]
3 Mortality in primary and secondary prevention trials with a low or high risk of bias 78 296707 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.98, 1.05]
3.1 Primary prevention trials with a low risk of bias 19 177868 Risk Ratio (M‐H, Random, 95% CI) 1.03 [0.97, 1.08]
3.2 Secondary prevention trials with a low risk of bias 37 66188 Risk Ratio (M‐H, Random, 95% CI) 1.03 [0.99, 1.07]
3.3 Primary prevention trials with a high risk of bias 7 38032 Risk Ratio (M‐H, Random, 95% CI) 0.93 [0.84, 1.03]
3.4 Secondary prevention trials with a high risk of bias 15 14619 Risk Ratio (M‐H, Random, 95% CI) 0.90 [0.81, 0.99]
4 Mortality after excluding trials administrating extra supplements in the antioxidant group 60 249526 Risk Ratio (M‐H, Random, 95% CI) 1.03 [0.99, 1.06]
4.1 Trials with low risk of bias 43 227700 Risk Ratio (M‐H, Random, 95% CI) 1.04 [1.01, 1.07]
4.2 Trials with high risk of bias 17 21826 Risk Ratio (M‐H, Random, 95% CI) 0.91 [0.82, 1.00]
5 Mortality after excluding trials with extra supplements for both intervention groups 63 283619 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.99, 1.05]
5.1 Trials with low risk of bias 48 233249 Risk Ratio (M‐H, Random, 95% CI) 1.04 [1.00, 1.07]
5.2 Trials with high risk of bias 15 50370 Risk Ratio (M‐H, Random, 95% CI) 0.93 [0.86, 1.00]
6 Mortality after excluding factorial trials with potential confounding 52 60544 Risk Ratio (M‐H, Random, 95% CI) 1.01 [0.94, 1.08]
6.1 Trials with low risk of bias 38 52955 Risk Ratio (M‐H, Random, 95% CI) 1.10 [1.05, 1.15]
6.2 Trials with high risk of bias 14 7589 Risk Ratio (M‐H, Random, 95% CI) 0.84 [0.73, 0.97]
7 Mortality after excluding factorial trials with potential confounding and trials with extra supplements 36 53542 Risk Ratio (M‐H, Random, 95% CI) 1.01 [0.92, 1.11]
7.1 Trials with low risk of bias 27 46783 Risk Ratio (M‐H, Random, 95% CI) 1.12 [1.06, 1.18]
7.2 Trials with a high risk of bias 9 6759 Risk Ratio (M‐H, Random, 95% CI) 0.84 [0.73, 0.98]
8 Mortality in beta‐carotene trials with a low or high risk of bias 31 195503 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.98, 1.07]
8.1 Trials with a low risk of bias 26 173006 Risk Ratio (M‐H, Random, 95% CI) 1.05 [1.01, 1.09]
8.2 Trials with a high risk of bias 5 22497 Risk Ratio (M‐H, Random, 95% CI) 0.81 [0.62, 1.07]
9 Mortality in vitamin A trials with a low or high risk of bias 18 61190 Risk Ratio (M‐H, Random, 95% CI) 1.04 [0.96, 1.13]
9.1 Trials with a low risk of bias 12 41144 Risk Ratio (M‐H, Random, 95% CI) 1.07 [0.97, 1.18]
9.2 Trials with a high risk of bias 6 20046 Risk Ratio (M‐H, Random, 95% CI) 0.96 [0.85, 1.07]
10 Mortality in vitamin C trials with a low or high risk of bias 41 90191 Risk Ratio (M‐H, Random, 95% CI) 1.01 [0.97, 1.05]
10.1 Trials with a low risk of bias 29 65942 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.98, 1.07]
10.2 Trials with a high risk of bias 12 24249 Risk Ratio (M‐H, Random, 95% CI) 0.93 [0.84, 1.04]
11 Mortality in vitamin E trials with a low or high risk of bias 64 211957 Risk Ratio (M‐H, Random, 95% CI) 1.02 [0.99, 1.04]
11.1 Trials with a low risk of bias 46 171244 Risk Ratio (M‐H, Random, 95% CI) 1.03 [1.00, 1.05]
11.2 Trials with a high risk of bias 18 40713 Risk Ratio (M‐H, Random, 95% CI) 0.92 [0.85, 0.99]
12 Mortality in selenium trials with a low or high risk of bias 24 86150 Risk Ratio (M‐H, Random, 95% CI) 0.96 [0.91, 1.01]
12.1 Trials with a low risk of bias 17 62740 Risk Ratio (M‐H, Random, 95% CI) 0.97 [0.91, 1.03]
12.2 Trials with a high risk of bias 7 23410 Risk Ratio (M‐H, Random, 95% CI) 0.91 [0.81, 1.01]

1.2. Analysis.

1.2

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 2 Mortality in 76 trials with a low or high risk of bias with assigned fatalities to all of the dropouts.

1.3. Analysis.

1.3

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 3 Mortality in primary and secondary prevention trials with a low or high risk of bias.

1.4. Analysis.

1.4

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 4 Mortality after excluding trials administrating extra supplements in the antioxidant group.

1.5. Analysis.

1.5

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 5 Mortality after excluding trials with extra supplements for both intervention groups.

1.6. Analysis.

1.6

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 6 Mortality after excluding factorial trials with potential confounding.

1.7. Analysis.

1.7

Comparison 1 Antioxidants versus placebo or no intervention, Outcome 7 Mortality after excluding factorial trials with potential confounding and trials with extra supplements.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

ADCS 1 1997.

Methods Alzheimer's Disease Cooperative Study (ADCS 1)
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America.
Number of participants randomised: 341, older than 18 years, mean age 73 years, 65% females from 23 centres participating in the Alzheimers Disease Co‐operative Study
Inclusion criteria: patients with probable Alzheimer's disease of moderate severity, as measured by a Clinical Dementia Rating of 2, free of other central nervous system diseases, were not taking psychoactive medications, and were residing either at home or in a supervised setting with a care giver but not in a skilled‐nursing facility.
Exclusion criteria: none stated.
Interventions The patients were randomly assigned to receive:
group 1: selegiline 5 mg (n = 87),
group 2: alpha‐tocopherol 1000 IU (n = 85),
group 3: selegiline and alpha‐tocopherol (n = 85),
group 4: placebo (n = 84).
Selegiline was given in a dose of 5 mg twice a day, and a racemic mixture of dl‐alpha‐tocopherol was given in a dose of 1000 IU twice a day. Both agents were given in the morning and in the afternoon for two years.
Outcomes The primary outcome measure was: the time to the occurrence of any one of the following outcomes: death; institutionalisation; loss of the ability to perform at least two of three basic activities of daily living (ie, eating, grooming, using the toilet), as measured by part 2 of the Blessed Dementia Scale; and severe dementia, defined as a Clinical Dementia Rating of 3.
Secondary outcome measures were: measures of cognition, function, behaviour, and the presence or absence of extrapyramidal signs.
Notes Compliance was monitored in two ways. At each visit, unused medication was returned and the pills were counted. Measures to counter poor compliance included additional phone contact or review of the correct medication dosing schedule with the appropriate caregivers. Compliance was also monitored with surveillance of serum tocopherol concentrations, and the level of selegiline was monitored by measuring amphetamine, its major metabolite, in the urine.
Compliance with treatment was good. Urine samples were available from 318 patients for analysis of amphetamine levels. The proportion of patients with positive tests for selegiline was 93% in the combined group, 98% in the selegiline group, 11% in the alpha‐tocopherol group, and 13% in the placebo group. Serum samples were available from 332 patients. The proportion of patients with positive tests for alpha‐tocopherol was 91% in the combined group, 93% in the alpha‐tocopherol group, 9% in the selegiline group, and 12% in the placebo group.
Follow‐up was conducted one month after enrolment and at three‐month intervals for the remainder of the two‐year trial period. At each interval, every effort was made to assess primary and secondary outcomes, regardless of whether an outcome measure had been reached or the medication had been discontinued.
The losses to follow‐up were 7% in the placebo group, 5% in the selegiline group, 9% in the alpha‐tocopherol group, and 6% in the selegiline and alpha‐tocopherol group.
Study agents were supplied by Somerset Pharmaceuticals, Tampa, Fla. (Selegiline) and Hoffmann‐LaRoche, Nutley, N.J. (alpha tocopherol).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ADCS 2 2005.

Methods Alzheimer`s Disease Cooperative Study (ADCS 2).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (three intervention groups).
Participants Country: United States and Canada.
Number of participants randomised: 769, 417 (54%) men and 352 (46%) women, aged 55‐90 years, mean age 72.9 years.
Inclusion criteria: amnestic mild cognitive impairment (MCI), age 55‐90 years, inclusive, study informant available, mini‐mental state examination (MMSE) 24‐30, adequate vision and hearing for neuropsychological testing, normal vitamin B12 level and thyroid function studies and non‐reactive rapid plasma reagin, electrocardiogram normal or no clinically significant abnormalities, a clinical dementia rating (CDR) of 0.5, all participants and study informants signed written consent
Exclusion criteria: significant cerebral vascular disease, depression, central nervous system infarct, infection or focal lesions of clinical significance on computed tomography or magnetic resonance Imaging, medical diseases or psychiatric disorders that could interfere with study participation, pregnant, lactating or of child bearing potential, taking vitamin supplements, other supplements or a multi‐vitamin, restrictions on concomitant medication usage, including those with significant cholinergic or anticholinergic effects or potential adverse effects on cognition.
Interventions Participants were randomly assigned to receive:
group 1: 2000 IU of vitamin E, placebo donepezil, and a multivitamin daily (n = 257);
group 2: 10 mg of donepezil, placebo vitamin E, and a multivitamin daily (n = 253);
group 3: placebo vitamin E, placebo donepezil, and a multivitamin daily (n = 259);
over a period of 3 years.
The multivitamin contained 15 IU of vitamin E.
The initial dose of donepezil was 5 mg daily, and the dose was increased to 10 mg after six weeks.
The initial dose of vitamin E was 1000 IU daily, and the dose was increased to 2000 IU (1000 IU twice daily) after six weeks. If a participant had difficulty tolerating the higher dose of vitamin E or donepezil, the investigator could reduce the dose of either medication temporarily and then rechallenge with the higher dose.
On verification by a central review committee that a participant met clinical criteria for Alzheimer's disease, the participant stopped taking donepezil or matching placebo in a blinded fashion and was offered open‐label donepezil until he or she completed the study at month 36.
Outcomes The primary outcome measure was: the time to the development of possible or probable Alzheimer's disease.
The secondary outcome measures were: the scores for the mini‐mental state examination (MMSE); the Alzheimer's Disease Assessment Scale, cognitive subscale (ADASCog); the global CDR (Clinical Dementia Rating); the CDR sum of boxes (the sum of individual CDR domain scores); the ADCS Mild Cognitive Impairment Activities of Daily Living Scale; the Global Deterioration Scale; and a neuropsychological battery of tests.
Notes Compliance with treatment is not described.
A total of 230 (29.9%) participants discontinued treatment during the double‐blind phase; 92 in the donepezil group, 72 in the vitamin E group, and 66 in the placebo group.
Supported by a grant from Pfizer and Eisai. DSM Nutritional Products donated the vitamin E.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias High risk There are other factors in the trial that could put it at risk of bias (for‐profit involvement).

Allsup 2004Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Great Britain.
Number of participants randomised: 164, mean age 83 (37% males).
Inclusion criteria: older people (60 or older) living in nursing and residential homes able to give informed consent (abbreviated mental test score > 7), not having neoplastic disease, and not prescribed immunosuppressant medication at the time of recruitment.
Exclusion criteria: taking multivitamin supplements, vitamin C, or vitamin B, previous adverse reaction to influenza vaccine.
Interventions Participants were randomly assigned to receive:
group 1: multivitamin and trace element supplement (vitamin A 2666 IU, vitamin D3 400 IU, vitamin E 60 mg, vitamin B1 1.2 mg, vitamin B2 1.4 mg, vitamin B6 3.0 mg, nicotinamide 14 mg, folic acid 0.6 mg, vitamin B12 200 µg, biotin 30 mg, calcium pantothenate 5 mg, vitamin C 120 mg, iron 12 mg, zinc 14 mg, copper 2 mg, iodine 150 µg, manganese 1 mg, chromium 50 µg, selenium 60 µg, molybdenum 100 µg, calcium 240 mg, and magnesium 100 mg, n = 81;
group 2: placebo, n = 83.
Tablets were taken over an 8‐week period.
Influenza vaccine was administered 4 weeks after tablet commencement.
Outcomes The primary outcome was: response to influenza vaccine.
Notes The nursing staff at each home were responsible for the administration of tablets.
Analysed for primary outcome were 61 (75.3%) participants in the active and 57 (68.7%) participants in the placebo group. Overall, 20 participants in the active and 26 participants in the placebo group were lost to follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit. The identities of placebo and supplement were kept with the manufacturer (Recip AB, Arsta, Sweden) and were not revealed to the researchers until all data had been analysed.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ALSRT 2001Low.

Methods Amyotrophic Lateral Sclerosis riluzole‐tocopherol study (ALSRT).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: France.
Number of participants randomised: 288, 158 men and 130 women, mean age 64 years.
Inclusion criteria: probable or definitive amyotrophic lateral sclerosis (ALS), according to El Escorial criteria, over 18 years of age at recruitment, to be able to fully understand the study information, have been treated with riluzole 950 mg b.i.d. for at least three months without presenting side effects.
Exclusion criteria: signs of dementia and/or major psychiatric disorders, another concomitant serious disease, or handicap likely to interfere with their assessment of survival, forced vital capacity of less than 60%, monoclonal gammopathy, conduction blocks of motor nerves on electromyography, hepatic or renal disfunction, pregnancy or breast feeding, creatinine plasma concentration above 200 µM, alanine aminotransferase and/or aspartate transaminase activity greater than twice the upper limit of the normal range, known hepatic disease, taking drugs known to be hepatotoxic, enzyme‐inducing or enzyme‐inhibiting, taking vitamin E.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E 500 mg plus riluzole 50 mg (n = 144);
group 2: placebo (n = 144) plus riluzole 50 mg;
one capsule (vitamin E) and one tablet (Riluzole) daily for one year.
Outcomes The primary outcome measure was: change in functional status of each patient using the modified Norris limb scale.
The secondary outcome measures were: survival (defined as the time to death or tracheostomy), bulbar function assessed with the Norris bulbar scale (total possible score 39) and manual muscle testing.
Notes Compliance with treatment was checked by measuring the plasma vitamin E levels.
Compliance with treatment good. In the vitamin E group, a highly significant increase in plasma levels of vitamin E was observed after 3 months of treatment.
No losses to follow‐up. One hundred and forty‐six patients (74 in the placebo group and 72 in the alpha‐tocopherol group) did not complete the one‐year treatment period.
Study agents were supplied by: alpha‐tocopherol (Toco 500R) Laboratories Rhone‐Poulenc Rorer (now trading under the name Laboratories Aventis); riluzole Rhone‐Poulenc Rorer.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

AMDS 1996Low.

Methods Age Related Macular Degeneration Study (AMDS)
Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States.
Number of participants randomised: 71, 66 men (93%) and 5 women (7%), mean age 72 years.
Inclusion criteria: monocular one line drop of acuity, not attributable to cataract, amblyopia, systemic or ophthalmic disease, loss of macular reflex, Retinal Pigment Epithelium disruption and drusen observable by 90 degrees lens stereoscopic evaluation.
Exclusion criteria: greater than 1 year prior use of vitamins, veterans who were former prisoners of war and veterans who were chronic alcoholics with tobacco/nutritional amblyopia or gastrointestinal absorption disorders.
Interventions Patients were randomly assigned to receive:
group 1: antioxidants (beta‐carotene 6 mg; vitamin E 200 IU; vitamin C 750 mg; citrus bioflavonoid complex 125 mg; quercitin 50 mg; bilberry extract 5 mg; rutin 50 mg; zinc picolinate 12.5 mg; selenium 50 µg; taurine 100 mg; n‐acetyl cysteine 100 mg; l‐glutathione 5 mg; vitamin B2 25 mg; chromium 100 µg (n = 39);
group 2: starch placebo (n = 32);
for a period of 18 months.
Outcomes The primary outcome was: non‐exudative age‐related macular degeneration.
Notes Compliance with treatment was not described.
Attrition data were as follows: 71 patients at baseline, 61 patients at 6 months, 59 patients at 12 months, and 59 patients at 18 months. Overall 2 participants from active intervention arm and 1 participant from placebo arm were lost to follow‐up.
Trial agents were provided by Twin Laboratories, Inc., Ronkonkoma, NY.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

AREDS 2001Low.

Methods Age Related Eye Disease Study (AREDS).
Randomised, double‐blind, placebo‐controlled trial.
Participants Country: United States of America.
Number of participants randomised: 4757; 56% female, aged 55 to 80 years, median age 68 years.
Inclusion criteria: participants free of any illness or condition that would make long‐term follow‐up or compliance with study medications unlikely or difficult. On the basis of fundus photographs graded by a central reading centre, best corrected visual acuity, and ophthalmologic evaluations, participants were enrolled in one of several AMD categories. At least one eye of each participant was free from eye disease that could complicate assessment of AMD, lens opacity progression, or visual acuity, and that eye could not have had previous ocular surgery (other than cataract surgery).
Exclusion criteria: illness or disorders (eg, history of cancer with a poor 7‐year prognosis, major cardiovascular or cerebrovascular event within the last year, haemochromatosis) that would make long term follow‐up or compliance with the study protocol unlikely or difficult. Persons bilaterally aphakic or pseudophakic were ineligible for AMD category one.
Interventions Participants were divided into two clinical trials: 
 AMD Trial (n = 128) and 
 Cataract Trial (n = 4629).
In the Cataract Trial participants were randomly assigned to receive:
group 1: placebo (n = 1456);
group 2: zinc 80 mg (as zinc oxide) (n = 869);
group 3: beta‐carotene 15 mg, vitamin C 500 mg, vitamin E 400 IU, (n = 1451);
group 4: beta‐carotene 15 mg, vitamin C 500 mg, vitamin E 400 IU and zinc (n = 853).
Two study medication tablets were to be taken each morning and two each evening, to meet the total daily dose requirement. Tablets were to be taken with food to avoid potential irritation of an empty stomach by zinc.
Participants were followed up for an average of 6.3 years.
Outcomes Primary outcome measures were: an increase from baseline in nuclear, cortical, or posterior subcapsular opacity grades or cataract surgery, and at least moderate visual acuity loss from baseline (> 15 letters).
Notes Compliance with treatment was checked by random serum assessments. Compliance with treatment was excellent. Overall adherence was estimated to be 75% or greater (participants took > 75% of their study tablets) for 70% of participants at five years. At 60 months, 20% of participants (20% both for current smokers and former or non‐smokers) reported taking some multivitamin supplement containing at least one of the study medication. In 1994 and 1996, AREDS participants were informed of the results of the Alpha‐Tocopherol, Beta Carotene Cancer Prevention Study and the Beta‐Carotene and Retinol Efficacy Trial suggesting potential harmful effects of beta‐carotene among smokers.
About 2.3% of participants were lost to follow‐up. The rate of participants withdrawal from the trial medication was 14% by 60 months and 15% by the end of trial. This rates include participants lost to follow‐up and current smokers, 24% of whom withdrew from the trial medication after the results from the clinical trials of beta‐carotene and lung cancer were announced. Overall, the vital status was known for 4753 out of 4757 participants.
The trial was supported from Bausch & Lomb Inc, Rochester, NY.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Centrally by the co‐ordinating centre using the on‐site computers, with procedures to protect the integrity of randomisation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment. Multiple levels of data encryption ensured the integrity of the treatment assignment files.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ASAP 2003Low.

Methods The Antioxidant Supplementation in Atherosclerosis Prevention (ASAP) Study.
Randomised, partially double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: Finland. Number of participants randomised: 520, 256 men and 264 postmenopausal women, smoking and non smoking, aged 45 to 69 years with serum cholesterol > 5 mmol/L (193 mg/dL). Inclusion criteria: participants with hypercholesterolaemia defined as serum cholesterol levels > 5 mmol/L (193 mg/dL). Exclusion criteria: regular intake of antioxidants, acetosalicylate, or any other drug with antioxidative properties, severe obesity (body mass index >32 kg/m2), type 1 diabetes, uncontrolled hypertension (sitting diastolic blood pressure >105 mm Hg), any condition limiting mobility, or severe disease shortening life expectancy. Premenopausal women and those taking oral oestrogen therapy were also excluded.
Interventions The study consisted of 8‐week dietary counselling and placebo lead‐in phase, a 3‐year double‐masked treatment period, and a 3‐year open treatment period. The participants were randomly allocated to receive twice daily with meal:
group 1: d‐alpha tocopherol 91 mg (corresponding to 100 mg of d‐alpha‐tocopheryl acetate and 136 IU of vitamin E) (n = 130);
group 2: 250 mg slow‐release vitamin C (n = 130);
group 3: both d‐alpha‐tocopherol and slow‐release ascorbic acid in a single tablet (CellaVie), (n = 130);
group 4: placebo only (n = 130);
for a period of 6 years.
Outcomes The primary outcome measure was: progression of carotid atherosclerosis.
Notes Compliance with treatment was checked by random serum assessments. Of the 390 participants randomised to supplementation, 335 continued the study after 3 years and 256 (76.4%) took the supplements as instructed for 6 years, whereas 62 participants stopped the supplements during the first 3 study years and additional 18 participants during the last 3 study years. The mean plasma alpha‐tocopherol and ascorbate concentration increased in 6 years in the group randomised to supplementation, and in the non supplemented group decrease.
Of the 520 participants randomised, 440 (84.6%) completed the study and underwent the six‐year re‐examination. overall, 55 participants in the three vitamin groups and 25 participants in the placebo group were lost to follow‐up.
Ferrosan A/S, Denmark, provided the vitamin supplements.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ATBC 2003Low.

Methods Alpha‐Tocopherol, Beta‐Carotene Cancer Prevention Study (ATBC).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: Finland. Number of participants randomised: 29,133 males. Inclusion criteria: male smokers (five or more cigarettes daily), aged 50 to 69 years, averaged 57.2 years of age at study entry, who lived in south‐western Finland. Exclusion criteria: men with a prior cancer or with other serious illness, or who used vitamin E, vitamin A, or beta‐carotene supplements in excess of predefined doses (> 20 mg, > 20000 IU, or > 6 mg, respectively), or anticoagulants.
Interventions Participants were randomly assigned in four groups to receive:
group 1: alpha‐tocopherol 50 mg (n = 7286);
group 2: beta‐carotene 20 mg (n = 7282);
group 3: alpha‐tocopherol and beta‐carotene, (n = 7278);
group 4: placebo (n = 7287); daily for five to eight years (median 6.1 years).
All participants took a single capsule daily. The four trial intervention groups were well balanced for all baseline characteristics evaluated. The two‐by‐two factorial design allowed assessment of the two intervention agents independently, with one‐half of participants receiving alpha‐tocopherol (n = 14,564) and the other half not (n = 14,569); similarly, half of the participants received beta‐carotene (n = 14,560) and half did not (n = 14,573).
The study was conducted between 1985 and 1993 (mean 6.1 years). The active intervention continued through April 30, 1993 and postintervention follow‐up until April 30, 2001. Mean follow‐up time regarding incident cancers and cause‐specific deaths was 12.1 years and overall mortality 14.1 years.
Outcomes The primary outcome measure was: incidence of lung cancer. Secondary outcome measures were: incidence of other major cancers, overall and cause specific mortality and incidence of other diseases.
Notes Compliance with treatment was assessed by counts of the remaining capsules at each visit, by measurement of serum alpha‐tocopherol and beta‐carotene levels after three years of supplementation, and by measurements in random serum samples throughout the study. Compliance with treatment was excellent with four out of five active participants taking more than 95% of the scheduled capsules. Dropout rate and compliance were similar between all four groups.
There were no losses to follow‐up.
All capsules were supplied by Hoffmann‐La Roche, Basel, Switzerland. Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Bonelli 1998.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Italy.
Number of patients randomised: 304, patients who had histologically confirmed adenomatous polyps removed from the colon, aged 25 to 75 years.
Inclusion criteria: age 25 to 75 years, at least one histologically confirmed colorectal adenoma endoscopically removed with a clean colon as a result of the work up.
Exclusion criteria: Familiar adenomatous polyposis, inflammatory bowel diseases, polypectomy performed more than six months before randomisation, adenoma with invasive carcinoma, previous colorectal resection, invasive cancer at any site, life‐threatening chronic heart, liver or kidney diseases, current use of vitamin or calcium supplements, mental disability precluding informed consent to participate and adherence to the treatment, patients with 10 adenomas or more and those with large sessile adenomas (3 cm or more in diameter). The clean colon after polypectomy was assessed by means of total colonoscopy. When a total colonoscopy was not feasible a double contrast barium enema was performed. Colonoscopy was scheduled on years one, three and five after randomisation.
Interventions Patients were randomly assigned to receive:
group 1: selenium 200 µg (l‐selenemethionine), zinc 30 mg, vitamin A 6000 IU, vitamin C 180 mg, vitamin E 30 mg (n = 147);
group 2: placebo (n = 157);
daily for 5 years.
Outcomes The primary outcome measure was: occurrence of metachronous adenomas detected at follow‐up endoscopic examinations.
Notes The overall 5 year actuarial compliance to the treatment was 51%. Of the 304 randomised patients, 233 (76.6%) underwent at least one endoscopic follow up examination: 117 in the active compound group and 116 in the placebo group.
Of 304 patients randomised, 233 (76.6%) underwent at least one endoscopic follow‐up examination. Overall, 30 participants in the active treatment group and 41 participant in the placebo group were lost to follow‐up.
Active intervention and placebo were provided by Pharma Nord.
Additional information obtained through personal communication with authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Burns 1989.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United Kingdom.
Number of participants randomised: 19 elderly patients suffering from senile dementia, mean age 81 years, 89% females.
Inclusion criteria: senile dementia.
Exclusion criteria: serious physical illness (malignancy, severe crippling arthritis or stroke affecting the ability to eat) and those who could not cooperate in a mental or physical examination.
Interventions Participants were randomly assigned to receive:
group 1: vitamin C 200 mg, vitamin B1 100 mg, vitamin B2 10 mg, vitamin B3 400 mg, and vitamin B6 10 mg (n = 10);
group 2: placebo (n = 9)
orally, daily, for a period of six weeks.
Outcomes The primary outcome measures were the progression of cognitive 
 impairment and behavioural disturbance in elderly demented patients.
Notes This study was supported by grants from The Wellcome Trust, Bencard plc and ISFE, The International Foundation for Nutrition Education and Nutrition Research, Zurich, Switzerland. Vitamin tablets were produced by Orovite, Bencard plc, Brentford, Middlesex, UK).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Unclear risk The trial was described as blinded, but the method of blinding was not described, so that knowledge of allocation was possible during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

CARET 2004Low.

Methods The Beta‐Carotene and Retinol Efficacy Trial (CARET).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design in a pilot phase and then one‐by‐one.
Participants Country: United States of America. Number of participants randomised: 18314; 12025 males and 6289 females. Inclusion criteria: smokers, former smokers, and workers exposed to asbestos at high risk of developing lung cancer. A total of 4060 male workers, mean age 57 years, exposed to asbestos and 14254 heavy smokers (44% of whom were women), mean age 58 years, were randomised. The participants agreed to limit their supplemental intake of vitamin A to less than 5500 IU per day and to take no supplemental beta‐carotene.
Interventions CARET builds on the experience of two pilot studies performed in Seattle (1985‐1988). The first pilot study initiated a phase III trial of the safety and efficacy of the study vitamins in 816 asbestos‐exposed participants randomised to a daily combination of 15 mg 13‐carotene and 25,000 IU retinol or a placebo medication. Participants were eligible up to age 74 and were not required to have a history of cigarette smoking; otherwise, the eligibility criteria were the same as for the asbestos‐exposed population in CARET. The second pilot study was a phase II trial of the comparative safety of the study vitamins in heavy smokers. The eligibility criteria were identical to those for heavy smokers in CARET. Overall 539 men and 490 women were randomised to one of four intervention groups: group 1: a daily combination of 30 mg beta‐carotene and 25,000 IU retinol; group 2: 30 mg beta‐carotene only; group 3: 25,000 IU retinol only; group 4: placebo medication. All 1845 participants in the two pilot studies continue to be followed for outcomes in CARET, together with approximately 16,000 additional participants. Participants of CARET trial were randomly assigned to receive: group 1: combination of 30 mg beta‐carotene and 25,000 IU vitamin A (n = 9420); group 2: placebo, (n = 8894). Both formulations were given as capsules. Beta‐carotene beadlets were combined with retinyl palmitate in a single capsule and dispensed in bottles, which were weighed and their content checked. The design projected active intervention until late 1997. The CARET active intervention was stopped 21 months earlier because of clear evidence of no benefit and substantial evidence of possible harm. The average duration of follow‐up was 10.0 years
Outcomes The primary outcome measure was: the incidence of lung cancer. Other outcomes reported are: mortality rates, and incidence of other cancers.
Notes Compliance was assessed by weighing the returned bottles to estimate the number of capsules remaining (in 85% of the assessments), or by relying on the participants own estimates (in 15% of the assessments). Compliance with treatment was excellent. Among the active participants, the mean rate of capsule consumption was 93% through five years of follow‐up, with no significant differences between the treatment groups. Participants who stopped receiving study vitamins for any reason other than death were defined as inactive participants and were still followed for outcomes and counted in the analysis. As of December 15, 1995 ascertainment of vital status for more than 98% was complete.
The losses to follow‐up were less than 2% at the end of treatment.
Active agents and placebos were purchased from Hoffmann‐La Roche and formulated by Tischon Corporation.
Data were extracted from the primary publication, but additional information was received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Chandra 1992.

Methods Randomised, double‐blind, placebo‐controlled trial.
Generation of the allocation sequence: adequate, four blocks of 24 random numbers.
Allocation concealment: unclear, not reported.
Blinding: adequate, the supplement and the placebo appeared identical and were prepared specifically for this study.
Follow‐up: adequate. Five participants on placebo and 3 participants from the supplemented group withdrew from the trial for personal reasons.
Intention‐to‐treat analysis: yes.
Sample‐size calculation: no.
Participants Country: Canada.
Number of participants randomised: 96 independently living, healthy elderly individuals aged 66 to 86 years, mean age 74 years, 41 men and 55 women.
Inclusion criteria: independently living, healthy elderly individuals over 65 years of age.
Exclusion criteria: none stated.
Interventions Participants were randomly assigned to receive:
group 1: vitamin A 400 µg retinol equivalents, beta‐carotene 16 mg, thiamin 2.2 mg, riboflavin 1.5 mg, niacin 16 mg, vitamin B6 3.0 mg, folate 400 µg, vitamin B12 4.0 µg, vitamin C 80 mg, vitamin D 4 µg, vitamin E 44 mg, iron 16 mg, zinc 14 mg, copper 1 .4 mg, selenium 20 µg, iodine 0.2 mg, calcium 200 mg, and magnesium 100 mg (n = 48);
group 2: placebo, calcium 200 mg and magnesium 100 mg (n = 48)
daily for one year.
Influenza vaccine was given four weeks before the end of the study. Participants with infection were treated appropriately with antimicrobial agents and supportive measures.
Outcomes The primary outcome measures were: immunocompetence and occurrence of infection related illness.
Notes Compliance was verified by interview at fortnightly visits and counting of leftover medication.
There was no report about the compliance.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Unclear risk The trial was described as blind, but the method of blinding was not described, so that knowledge of allocation was possible during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias High risk There are other factors in the trial that could put it at risk of bias (for‐profit involvement).

CHAOS 1996Low.

Methods Cambridge Heart Antioxidant Study (CHAOS).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United Kingdom.
Number of participants randomised: 2002; 1690 men and 312 women, mean age 61.8 years.
Inclusion criteria: angiographically proven coronary arthrosclerosis.
Exclusion criteria: prior use of vitamin supplements containing vitamin E.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E 800 IU, and then 400 IU (free 2R,4'R,8'R‐alpha‐tocopherol from natural sources in soya oil) (n = 1035): 546 patients received 800 IU daily for a median of 731 days (range 3 to 981), and 489 newly recruited patients received 400 IU daily for a median of 366 days (8 to 961). These two groups are not distinguished in the analysis.
group 2: matching placebo (oil only), (n = 967).
Median follow‐up was 510 days (range 3 to 981).
Outcomes The primary outcome measures were: non‐fatal myocardial infarction alone and combination of non‐fatal myocardial infarction and cardiovascular death.
Notes Compliance with treatment was measured as the ratio of days that study medication was requested to per‐protocol days prescribed. 73.2% of all prescribed alpha‐tocopherol or placebo were requested as follow‐up medications. There was no difference between treatment groups in the proportion of participants who were 100% compliant with the trial medication (48% placebo, 49% alpha‐tocopherol).
Complete follow‐up data were available in 98% of participants. There were no differences between the groups in completeness of follow‐up (98% placebo, 97.8% active treatment). Overall 23 participants in the active and 19 participants in the placebo group were lost to follow‐up.
Study agents were supplied by Henkel Corporation (La Grange, Illinois, USA)
Data were extracted from Mitchinson MJ, Stephens NG, Parsons A, Bligh E, Schofield PM, Brown MJ. Mortality in the CHAOS trial. 
 Lancet 1999;353:381‐2.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Collins 2003Low.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America. Number of participants randomised: 52, mean age 67, 98% males.
Inclusion criteria: current diagnosis of peripheral arterial disease, a history of intermittent claudication, and an ankle‐brachial index < 0.95 at rest and/or < 0.85 after exercise.
Exclusion criteria: taking any of the following drugs, vitamin E, Coumadin, or pentoxifylline.
Interventions Participants were randomly assigned in four groups to receive:
group 1: PoleStriding exercise with vitamin E (n = 13);
group 2: PoleStriding exercise with placebo (n = 14);
group 3: vitamin E without PoleStriding exercise (n = 13);
group 4: placebo without exercise (n = 12).
The dose of vitamin E was 400 IU daily. Participants were supplemented 0.5 year, and followed 2.5 years.
PoleStriding is a form of walking that uses muscles of the upper and lower body in a continuous movement similar to cross country skiing.
Outcomes The primary outcome was: walking ability and perceived quality of life.
Notes Compliance with the study drug treatment was monitored in two ways: patient self‐report and measured vitamin E levels. Vitamin E levels were obtained at baseline and 3 and 6 months. Investigators did not receive the measured vitamin E levels until the trial ended.
For the first 3 months drug compliance was assessed biweekly by the study staff and monthly thereafter.
Six randomised participants did not complete the study; 1 participant from the exercise and vitamin E group, 3 participants from exercise plus placebo group, and 2 participants from the placebo group.
Vitamin E and placebo capsules were provided by the Henkel Corporation, La‐ Grange, IL).
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Correa 2000Low.

Methods Randomised, controlled, partially double‐blind, chemoprevention trial with two‐by‐two‐by‐two factorial design.
Participants Country: Colombia.
Number of participants randomised: 976; 46% males, aged 29 to 69 years, mean age 51.1 years.
Inclusion criteria: preliminary histologic diagnosis of multifocal atrophic gastritis with or without intestinal metaplasia and dysplasia, good health.
Exclusion criteria: normal histology, non‐atrophic gastritis, gastric cancer.
Before randomisation, participants were classified into one of three strata: atrophy (without metaplasia), intestinal metaplasia, or dysplasia, according to baseline histologic diagnosis.
Interventions Participants were assigned to a dietary supplement of beta‐carotene (30 mg once per day) and/or ascorbic acid (1 g twice a day) or their corresponding placebos, for a six‐year period. The prevalence of Helicobacter pylori infection among all gastric biopsy specimens was 97%. Anti‐Helicobacter pylori treatment consisting of amoxicillin (500 mg three times per day), metronidazole (375 mg three times per day), and bismuth subsalicylate (262 mg three times per day) was given for 14 days to half of the study participants assigned randomly. This treatment was not blinded or placebo controlled because an appropriate placebo was not available for bismuth subsalicylate.
Participants were divided in eight treatment groups to receive:
group 1: placebo (n = 117);
group 2: anti‐Helicobacter pylori treatment, which consisted of amoxicillin, metronidazole, and bismuth subsalicylate (n = 120);
group 3: beta‐carotene (n = 117);
group 4: ascorbic acid (n = 130);
group 5: Helicobacter pylori treatment, which consisted of amoxicillin, metronidazole, and bismuth subsalicylate, and additionally beta‐carotene (n = 126);
group 6: Helicobacter pylori treatment, which consisted of amoxicillin, metronidazole, and bismuth subsalicylate, and additionally ascorbic acid (n = 111);
group 7: beta‐carotene and ascorbic acid (n = 121);
group 8: Helicobacter pylori treatment, which consisted of amoxicillin, metronidazole, and bismuth subsalicylate, and additionally beta‐carotene and ascorbic acid (n = 134).
Gastric biopsy specimens taken at baseline were compared with those taken at 72 months.
Outcomes The primary outcome measures were: progression, no change or regression of gastric precancerous lesions (preliminary histologic diagnosis of multifocal atrophic gastritis with or without intestinal metaplasia and dysplasia). For our purposes we extracted data about the incidence of gastric cancer.
We have also extracted data on overall mortality for all antioxidants as well as for beta‐carotene and vitamin C.
Notes Compliance with treatment was constantly encouraged and monitored by a social worker who interviewed the participants and recorded pill counts every three months. In addition, blood levels of beta‐carotene and ascorbic acid were measured every three months in a 20% random sample of the participants.
Compliance with treatment among participants who completed the study was high for all intervention modalities (mean compliance for ascorbic acid, 91.8%; for beta‐carotene, 92.3%; and for anti‐Helicobacter pylori treatment, 99.1%).
The average rate of loss was 4.3% per year over the six‐year trial. Two hundred twenty‐one participants withdrew from the study before their 72‐month evaluation: 102 quitted treatment, 59 were lost to follow‐up, 34 dropped out of the study because of pregnancy and other medical conditions, 18 died of causes unrelated to gastric cancer, and eight developed cancer other than gastric cancer. In one participant, the 72‐month biopsy specimen was inadequate for histologic evaluation and determination of outcome. A total of 684 participants came to the 36‐month biopsy; of those, 92% (631) came for the 72‐month biopsy, there was a dropout rate of 2.6% per year for the last three years of the trial. Overall 24 participants from the placebo group, 25 from anti‐Helicobacter pylori (anti‐HP), 34 from the beta carotene (BC), 23 from the ascorbic acid (AA), 20 from the anti‐HP + BC, 23 from anti‐HP + AA, 17 from BC + AA, and 37 from anti‐HP + BC + AA were lost to follow‐up.
Active medication and placebos were provided like identical coded tablets by Hoffmann‐La Roche Inc. (Nutley, NJ).
Additional information received through personal communication with the authors.
Data were extracted from the article: Correa, et al. Re: Chemoprevention of gastric dysplasia: Randomized trial of antioxidant supplements and anti‐Helicobacter pylori therapy. Journal of the National Cancer Institute 2001; 93: 559.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

CTNS 2008.

Methods Clinical Trial of Nutritional Supplements and Age‐Related Cataract (CTNS).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Italy.
Number of participants randomised: 1020, aged 55 to 75 years, mean age 68 years, 45% women, with early or no cataract.
Inclusion criteria: early cataract or no cataract.
Exclusion criteria: late cataract, eye conditions that might interfere with the prospective evaluation of lens changes, current use of dietary supplements containing nutrients in the study medication, conditions or circumstances that might interfere with participant follow‐up, and refusal to sign the informed consent form.
Interventions Patients were randomly assigned to receive:
group 1: multivitamin tablet (vitamin A 5000 IU, vitamin C 60 mg, vitamin E 30 IU, selenium 25 μg) (n = 510);
group 2: placebo tablet(n = 510)
daily, for an average of nine years.
Outcomes The primary outcome measure was a prespecified increase from baseline in nuclear, cortical, or posterior subcapsular cataract opacity grades or cataract surgery. Secondary outcome measures included an increase in type‐specific opacity grades, cataract surgery, and visual acuity loss from baseline ≥15 letters.
Notes Overall, more than 90% of participants took more than 50% of their study tablets (data not shown). Another measure of compliance for each participant is the median of the annual compliance data for the participant. Using this measure, we can say that, on average, participants took 91% of the study tablets. Adherence to treatment regimen was balanced in the treatment and placebo groups.
One hundred fifty‐eight participants (15%) were nominally lost to follow‐up, including 45 who declined to participate in the study’s extension. However, closeout participant contacting procedures resulted in 48 participants returning for a final study visit, date and cause of death ascertained for another 26, cataract surgery information obtained for another 77, and only seven lost to all follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

DATATOP 2005Low.

Methods The Deprenyl and Tocopherol Antioxidative Therapy of Parkinsonism (DATATOP).
Randomised, double‐blind, placebo‐controlled secondary prevention trial with two‐by‐two factorial design.
Participants Country: United States and Canada.
Number of patients randomised: 800, 530 men and 270 women, 66% men, mean age 61 years.
Inclusion criteria: early Parkinsons disease not requiring levodopa without severe postural instability (Parkinsons Disease at Hoehn and Yahr stage 1 or 2) within a 5 years of symptom onset and not yet requiring symptomatic therapy.
Exclusion criteria: important comorbid illness, cognitive impairment (Mini Mental State examination score of < 23, or severe tremor (tremor score of >3 on the Unified Parkinsons Disease Rating Scale.
Interventions Patients were randomly assigned to receive: group 1: vitamin E (dl‐alpha‐tocopherol; all‐racemic) 1000 IU; group 2: deprenyl 5 mg; group 3: tocopherol and deprenyl; group 4: placebo; twice daily with morning and evening meals. 401 participants were assigned to tocopherol and 399 participants to deprenyl. Participants were instructed to take one tablet and one capsule twice daily with morning and evening meals. A standard multivitamin containing vitamin E (30 IU) was provided to all participants. Median duration of vitamin E exposure during the randomised phase was 2.6 years. Preliminary analysis in the fall of 1989, after an average 1.5 years of follow‐up, indicated unexpectedly striking effects of deprenyl in postponing PD disability as measured by the need for levodopa therapy. After this disclosure, all active trial participants, whether or not they required levodopa therapy, were placed on open‐label deprenyl, 10 mg/day, for about 3.5 years, from fall 1989 to spring 1993. Blinded tocopherol treatment assignments were maintained for about 3 years after the initial randomisation. Participants began taking levodopa (with carbidopa, a peripheral dopa decarboxylase inhibitor) in addition to their experimental treatments at any point in the trial when they were judged clinically to require therapy for emerging disability. Investigators adjusted levodopa dosage to achieve optimal clinical benefits and avoid dopaminergic adverse effects. Because of concerns about the sustained benefit of deprenyl, a second randomisation was undertaken in spring 1993. Consenting research participants who required levodopa were randomised, independently of their original randomisation, to continue deprenyl (50%) or to switch to deprenyl placebo (50%). Further adjustments of levodopa dosage were permitted after the second randomisation. This additional placebo‐controlled phase of deprenyl assignment was continued for 2 years until the last formal (face‐to‐face) clinical evaluation in spring 1995. The 800 participants have therefore been followed at least annually for an average of 8.2 years. Participants who underwent the second randomisation had a minimum of 3.2 years and a maximum of 7.3 years of exposure to active deprenyl. Participants were followed 13 years.
Outcomes The primary outcome in the trial occurred when, in the judgement of the enrolling investigator, a participant reached a level of functional disability sufficient to warrant the initiation of levo‐dopa therapy. Operationally, the primary response variable in the trial was defined as the time from randomisation to the end point. After the outcome, the experimental treatments were withdrawn in blinded fashion, and approximately 30 days later the participants received a final evaluation.
Notes Losses to follow‐up almost equal in each treatment group.
Monitoring of compliance was carried out in follow‐up evaluations in which unused doses return by the subject were counted, the serum tocopherol levels measured, and the urinary levels of amphetamine and methamphetamine metabolites of deprenyl determined. The results of the compliance monitoring were not shared with the participants or the investigators.
Compliance in taking experimental medications was excellent among all treatment groups. The overall compliance rate, as a percentage of the doses dispensed that were actually taken, ranged from 97.9 to 99.5 percent for both tocopherol and deprenyl.
Tablets of l‐deprenyl and matching placebos were provided by Someret Pharmaceuticals, Denville; N.J. Capsules of d‐l‐alpha tocopherol and matching placebos were provided by Hoffman‐LaRoche, Nutley, N.J. A standard vitamin (One‐A‐Day) by Miles Laboratories, Elkhart, Ind.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

DATOR 2004Low.

Methods Randomised, single‐blind, placebo‐controlled trial with parallel group design.
Participants Country: Belgium.
Number of participants randomised: 24, mean age 51, 86% males.
Inclusion criteria: type 1 diabetic patients attending the outpatient diabetes clinic having history of high serum cholesterol, (total cholesterol > 4.9 and LDL cholesterol > 3.0 mmol/L but Triglycerides < 4.5 mmol/L) and normal blood levels of thyroxin (9.7‐23.4 pmol/L) and TSH (0.47‐4.7µU/mL).
Exclusion criteria: none reported.
Interventions Participants were randomly assigned to receive:
group 1: Atorvastastin® 20 mg together with 750 IU (504 mg) d‐alpha‐tocopherol;
group 2: Atorvastastin® with placebo (280 mg soybean oil containing 0.25 mg tocopherol per capsule), daily.
Participants were supplemented 0.5 years and followed 4.5 years.
Outcomes The primary outcome measure was: impact on lipids and peroxidation during statin treatment.
Notes During the course of the trial, 1 participant from each group dropped out ‐ 1 participant due to thyroid dysfunction and 1 participant due to an accident.
Omega–Pharma NV is acknowledged for the supply of alpha‐tocopherol and placebo.
Additional information received through personal communication with the authors.
Due to the addition of 0.25 mg tocopherol to the control group, the 'placebo' control of the trial can be discussed.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

de la Maza 1995.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design.
Participants Country: Chile.
Number of participants randomised: 74, mean age 50, 85% males.
Inclusion criteria: clinical evidence of alcoholic liver disease at the time of enrolment (two or more of the following): jaundice, encephalopathy, ascites, oedema, spider nevi, marked collateral circulation, bleeding disorders, oesophageal varices on endoscopy; a history of > 5 years of heavy alcohol consumption (daily alcohol intake > 150 g); absence of hepatitis B surface antigen; absence of significant renal, pulmonary or cardiac disease, clinical diabetes or malignant tumours (including hepatoma).
Exclusion criteria: none mentioned.
Interventions Patients were randomly assigned to receive:
group 1: vitamin E 500 mg (in the form of alpha‐tocopheryl acetate, n = 37;
group 2: placebo, n = 37.
Participants were supplemented and followed 1 year.
Outcomes The primary outcome measure was: the liver function, mortality, and hospitalisation rates.
Notes Patients were seen once a month by a nurse practitioner at the liver disease clinic. Patients were asked about compliance to the treatment, which was assessed by counting the leftover tablets.
Seven randomised participants were removed from the trial due to lack of compliance.
Blood samples were obtained at the beginning of the study and every 3 months to measure serum levels of vitamin E.
Financing was provided by Roche and Saval Laboratories.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias High risk There are other factors in the trial that could put it at risk of bias (for‐profit involvement).

de Waart 2001.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: The Netherlands.
Number of patients randomised: 
 218 men, aged from 50 to 76 years, mean age 60 years.
Inclusion criteria: male cigarette smokers aged 50 to 76 years.
Exclusion criteria: diabetes patients, users of (multi)vitamin‐, vitamin E, vitamin C, beta‐carotene, garlic, or fish oil supplements, users of vitamin K antagonists (phenprocoumon, acenocoumarol), individuals with current illness interfering with participation, and unwillingness to participate.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (dl‐alpha‐tocopherol) 400 IU (n = 109);
group 2: placebo (n = 109)
for a period of 2 years. Mean follow‐up time was 1.8 years.
Outcomes The primary outcome measure was: progression of atherosclerosis in lifelong male smokers measured by 2‐year change of the common carotid intima media thickness as measured by B‐mode ultrasonography.
Notes Compliance with treatment is not reported.
Twenty‐nine participants (13 in active and 16 in placebo groups) were lost to follow‐up.
Vitamin E or placebo capsules were provided by F Hoffman La Roche Ltd, Basel.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Unclear risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes High risk The number or reasons for dropouts and withdrawals were not described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

Garbagnati 2009Low.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Nutristroke Trial
Participants Country: Italy.
Number of participants randomised: 72 stroke patients, mean age 65.3 years, 35% females, admitted for the rehabilitation of sequelae of 
 their first Ischaemic stroke.
Inclusion criteria: stroke survivors admitted for the rehabilitation of sequelae of their first Ischaemic stroke.
Exclusion criteria: onset‐admission interval > 60 days, hemorrhagic lesions and the presence of other chronic disabling pathologies and/or medical conditions that would contraindicate physical therapy, and inability or refusal to give consent.
Interventions Participants were randomly assigned to receive:
group 1: antioxidants (vitamin C 240 mg; vitamin E 700 IU; beta‐carotene (n = 19 mg) (n = 16);
group 2: n‐3 fatty acids 0.5 mg (n = 20);
group 3: antioxidants and n‐3 fatty acids (n = 18);
group 4: placebo (n = 18)
orally, daily, for the period of one year.
Outcomes The primary outcome measures were clinical and functional status of the patients.
Notes Sigma‐Tau Health Science, Rome, supplied n–3 dietary supplements free of charge.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using a random number table.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Gillilan 1977.

Methods Randomised, double‐blind, placebo‐controlled cross‐over trial.
Participants Country: United States.
Number of patients randomised: 
 52.
Inclusion criteria: typical, stable, effort‐related angina pectoris, and Q wave electrographic evidence of previous myocardial infarction and/or positive coronary arteriograms as defined by 75% obstruction at least one major coronary artery (31 patients).
Exclusion criteria: none stated.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (d‐alpha‐tocopherol succinate) 1600 IU (n = 26);
group 2: placebo (containing 2.5 mg of riboflavin), (n = 26);
three capsules of vitamin E (400 IU) or placebo daily for a period of six months, and then cross‐over.
The mean duration of double‐blind therapy was 189 days of vitamin E and 192 days of placebo.
Outcomes The primary outcome measure was: any improvement of angina pectoris.
Notes Drug adherence was followed by capsule count and a urine fluorescence test. Serum vitamin E levels were measured at baseline and at the end of the first and six months of each treatment phase.
The capsule count data shows a mean consumption of 88% of the prescribed capsules during vitamin E phase, and 84% consumption during placebo therapy. The percent of urine specimens with fluorescence indicate 78% of taking placebo prescribed medication. The serum tocopherol levels were significantly higher during the supplementation.
Forty‐eight participants (of 52) completed the trial. There were no losses to follow‐up.
Vitamin E and placebo capsules supplied by Wilson and Wolfer Pharmaceutical Manufacturers and Distributors, Detroit, Michigen.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Unclear risk Not all pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

Girodon 1997.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design.
Participants Country: France.
Number of participants randomised: 81 elderly participants, 20 males and 61 females, aged 65 to 102 years, with an average age of 84 years.
Inclusion criteria: at least 65 years of age, only age related diseases (osteoarthritis, hypertension, residual stroke etc.), that required chronic care.
Exclusion criteria: history of cancer, taking medication that might interfere with nutritional status, immunocompetence, or vitamin or mineral supplements.
Interventions Participants were randomly assigned to receive:
group 1: placebo (n = 20);
group 2: trace elements (zinc 20 mg in a form of zinc sulphate; selenium 100 µg in a form of selenite) (n = 20);
group 3: vitamins (vitamin C 120 mg; beta‐carotene 6 mg; vitamin E 15 mg) (n = 20);
group 4: combination of trace elements and vitamins at equal doses (n = 21);
daily (one capsule a day) for a period of 2 years.
Mean duration of follow‐up was 730 days.
Outcomes The primary outcome measure was impact of a trace element and vitamin supplementation on infectious morbidity.
Notes Compliance with treatment was checked by measuring the plasma vitamin levels and counting of the returned capsules.
Compliance was good. After 6 months of supplementation, a significant increase in vitamin and trace element serum levels was obtained in the corresponding treatment groups: a plateau was then observed for the whole study. No changes appeared in the placebo group.
There were no losses to follow‐up.
Study agents were provided by Produits Roche SA.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

GISSI 1999.

Methods GISSI‐Prevenzione trial (Gruppo Italiano per lo Studio della Sopravvivenza nell'Infarto miocardico).
Randomised clinical secondary prevention trial with two‐by‐two factorial design.
Participants Country: Italy.
Number of participants randomised: 
 11324; 9659 males and 1665 females, mean age 59 years.
Inclusion criteria: recent (< 3 months) myocardial infarction, informed written consent. Age limits were not defined.
Exclusion criteria: contraindications to the dietary supplements (ie, known allergy to n‐3 PUFA or alpha‐tocopherol, or known congenital defects of coagulation), unfavourable short‐term outlook (eg, overt congestive heart failure, cancers, etc).
Interventions Patients were randomly assigned to four treatment groups to receive:
group 1: n‐3 PUFA alone (as 1 gelatin capsule containing 850 mg to 882 mg eicosapentaenoic acid and docosahexaenoic acid as ethyl esters in the average ratio of 1:2), 1 g daily (n = 2836);
group 2: vitamin E alone, 300 mg daily (n = 2830);
group 3: n‐3 PUFA and vitamin E combined (n = 2830);
group 4: no supplement (n = 2828);
for 3.5 years.
Outcomes The primary outcome measures were: cumulative rate of all‐cause death, non‐fatal myocardial infarction, and non‐fatal stroke; and the cumulative rate of cardiovascular death, non‐fatal myocardial infarction, and nonfatal stroke.
Notes Compliance with treatment was measured by refilling drug supplies every three months. Compliance with assigned treatment was excellent. At year one and at the end of the study, 11.6% and 28.5% of patients receiving n‐3 PUFA and 7.3% and 26.2% of those receiving vitamin E, respectively, had permanently stopped taking the study drug. Conversely, during the whole course of the study, only two patients not assigned vitamin E and 26 patients not assigned n‐3 PUFA were receiving these drugs.
Information on the vital status of patients at the end of the trial was available for 99.9% of the population.
The trial was supported by grants from Bristol‐Myers Squibb, Pharmacia‐Upjohn, Società Prodotti Antibiotici, and Pfizer. Pharmacia‐Upjohn and Società Prodotti Antibiotici supplied marketed capsules containing 850 to 882 mg EPA/DHA ethyl esters. Vitamin E (acetyl d, l‐a‐ tocopherol) was supplied by Bracco.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) High risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes High risk The trial was not blinded, so that the allocation was known during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias High risk There are other factors in the trial that could put it at risk of bias (for‐profit involvement).

Graat 2002Low.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: The Netherlands.
Number of participants randomised: 652, 325 men and 327 women, older than 60 years.
Inclusion criteria: noninstitutionalized elderly persons older than 60 years.
Exclusion criteria: used immunosuppressive treatment, anticoagulants interfering with vitamin K metabolism, or dietary supplements in the previous 2 months or if they had a history of cancer, liver disease, or fat malabsorption during the 5 years before randomisation.
Interventions Participants were randomly assigned to receive:
group 1: multivitamins and minerals (n = 163);
group 2: vitamin E 272 IU (n = 164);
group 3: multivitamins and minerals plus vitamin E (n = 172);
group 4: placebo (n = 153).
The multivitamin‐mineral capsule contained: retinol (600 µg), beta‐carotene (1.2 mg), ascorbic acid (60 mg), vitamin E (10 mg), cholecalciferol (5 µg), vitamin K (30 µg), thiamin mononitrate (1.4 mg), riboflavin (1.6 mg), niacin (18 mg), pantothenic acid (6 mg), pyridoxine (2.0 mg), biotin (150 µg), folic acid (200 µg), cyanocobalamin (1 µg), zinc (10 mg), selenium (25 µg), iron (4.0 mg), magnesium (30 mg), copper (1.0 mg), iodine (100 µg), calcium (74 mg), phosphor (49 mg), manganese (1.0 mg), chromium (25 µg), molybdenum (25 µg), and silicium (2 µg). The vitamin E capsule contained 200 mg/dL of alpha‐tocopheryl. Placebo capsules contained soybean oil. Quality control of the capsules after treatment showed no decrease in the original contents.
Each participant received 2 capsules per day to be ingested with dinner for a maximum of 15 months.
Outcomes The primary outcomes were: incidence and severity of acute respiratory tract infections.
Notes Compliance with treatment was checked by measuring the plasma vitamin levels and counting of the returned capsules. Baseline plasma samples were collected for determination of alpha‐tocopherol, ascorbic acid, retinol, and carotenoids. To monitor compliance, these assessments were repeated in a postintervention sample of a subset (n = 300). Returned capsules were counted for all participants. After treatment, ascorbic acid, total carotenoids, alpha‐tocopherol, and cholesterol‐adjusted alpha‐tocopherol levels increased significantly in the multivitamin‐mineral and multivitamin‐mineral plus vitamin E group, while gamma‐tocopherol decreased significantly. In the vitamin E group, alpha‐tocopherol and cholesterol‐adjusted alpha‐tocopherol levels increased significantly, while gamma‐tocopherol levels decreased significantly. In the placebo group, none of the measured vitamins changed significantly. 94% of the participants met the compliance criteria of 80% capsule intake.
In total, 16% of the participants discontinued the treatment. Overall, 26 participants assigned to receive multivitamin‐mineral, 25 to vitamin E, 26 to multivitamin‐mineral plus vitamin E, and 20 to placebo were lost to follow‐up.
Trial agents were provided by Roche Vitamins, Europe, Basel, Switzerland.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Graf 2005Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design.
Participants Country: Germany.
Number of participants randomised: 160, 104 males and 56 females, mean age 58 years with probable or definite amyotrophic lateral sclerosis (ALS).
Inclusion criteria: patients with probable or definite amyotrophic lateral sclerosis (ALS) treated with riluzole and disease duration of less than 5 years.
Exclusion criteria: none stated.
Interventions Patients were randomly assigned to receive:
group 1: vitamin E (alpha‐tocopherol), 5000 mg (five times daily one capsule of 1000 mg) (n = 83);
group 2: placebo (n = 77);
for a period of 18 months.
Outcomes Primary outcome measure was: survival, calculating time to death, tracheostomy, or permanent assisted ventilation.
Secondary outcome measures were: the rate of deterioration of function assessed by the modified Norris limb and bulbar scales, manual muscle testing, spasticity scale, ventilatory function and the Sickness Impact Profile.
Additional information obtained through personal communication with authors.
Notes Compliance with treatment was checked by measuring the plasma vitamin E levels. Vitamin plasma levels were, as expected, significantly higher in the high dose vitamin E group than in the placebo group.
There were no losses to follow‐up.
Trial agents were provided by Schwarzhaupt, Cologne, Germany.
Additional information obtained with personal communication with authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Grieger 2009Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Australia.
Number of participants randomised: 115 aged care residents, 52% females.
Inclusion criteria:
Interventions Participants were randomly assigned to receive:
group 1: multivitamin tablet (beta‐carotene 3 mg, vitamin C 75 mg, vitamin E 10 IU) (n = 58);
group 2: placebo tablet (n = 57)
orally, daily, for a period of 5 months.
Outcomes The primary outcome measures were nutritional status, bone 
 quality and muscle strength.
Notes Multivitamin and placebo tablets were administered by the nursing staff during the morning medication round (08:00 hours) with some participants in low care facilities choosing to self‐medicate.
Participants consumed on average 82% of the tablets (range: 79?100%) in the placebo group and 91% (range 76?100%) in the multivitamin group.
Initially, it was intended to be a two‐by‐two factorial design, utilizing fortified milk vs usual milk, where subjects were stratified by mobility for milk allocation; and multivitamins vs placebo supplementation. However, due to problems with the distribution of the milk to subjects within the facility, the use of fortified milk ceased completely at week 16 of the six‐month study.
This study was funded by Murray Goulbourn Co‐operative Co. Ltd, Sigma Australia Pharmaceuticals Company Pty and the Australian Research Council 
 (Linkage‐Projects), but had no input into the results presented in this report.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

HATS 2001Low.

Methods The HDL‐Atherosclerosis Treatment Study (HATS).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States and Canada.
Number of participants randomised: 160, mean age 53 years, 13 % female.
Inclusion criteria: men (younger than 63 years of age) and women (younger than 70 years of age) with clinical coronary disease (defined as previous myocardial infarction, coronary interventions, or confirmed angina) and with at least three stenoses of at least 30 percent of the luminal diameter or one stenosis of at least 50 percent. All had low levels of HDL cholesterol (35 mg per decilitre [0.91 mmol per litre] or lower in men and 40 mg per deciliter [1.03 mmol per litre] in women), LDL cholesterol levels of 145 mg per deciliter (3.75 mmol per litre) or lower, and triglyceride levels below 400 mg per decilitre (4.52 mmol/L).
Exclusion criteria: lipid levels outside of the specified ranges, coronary bypass surgery, severe hypertension, recent gout, or liver, thyroid, or kidney disease, or uncontrolled diabetes.
Interventions Patients were randomly assigned to receive:
group 1: simvastatin (10 mg to 20 mg) plus niacin (500 mg to 4 g), (n = 33);
group 2: antioxidant vitamins, 800 IU of vitamin E (as d‐alpha‐tocopherol), 1000 mg of vitamin C, 25 mg of natural beta‐carotene, and 100 µg of selenium;
group 3: simvastatin plus niacin plus antioxidants (n = 40).
group 4: all placebos (n = 34);
for three years.
Simvastatin therapy began at 10 mg per day for patients with an LDL cholesterol level of 110 mg per decilitre (2.84 mmol per litre) or lower on screening and 20 mg per day for those with an LDL cholesterol level higher than 110 mg per decilitre. The dose was increased by 10 mg per day in patients whose LDL cholesterol level was higher than 90 mg per decilitre (2.33 mmol per litre) in any sample during the first year of the study and was reduced by 10 mg per day if the LDL cholesterol level fell below 40 mg per decilitre (1.03 mmol per litre) at any time during the study. During treatment, patients receiving the matching placebo were given 10 mg of simvastatin if their LDL cholesterol level was 140 mg per decilitre (3.62 mmol per litre) or higher; the target level was 130 mg per decilitre (3.37 mmol per litre) or lower. The dose of slow‐release niacin was increased linearly from 250 mg twice daily to 1000 mg twice daily at four weeks. Patients whose HDL cholesterol levels had not increased by at least 5 mg per decilitre (0.13 mmol per litre) at 3 months, at least 8 mg per decilitre (0.21 mmol per litre) at 8 months, and at least 10 mg per decilitre at 12 months were switched to crystalline niacin the dose of which was gradually increased to 3 g per day or, at most, 4 g per day in order to meet the target levels. Niacin "placebo" tablets (taken at a dose of 50 mg twice daily) were active, provoking flushing without affecting lipid levels.
Outcomes The primary outcome measures were: arteriographic evidence of a change in coronary stenosis and the occurrence of a first cardiovascular event (death, myocardial infarction, stroke, or revascularisation).
Notes Compliance with the trail regimens, measured by means of pill counts, ranged between 80 percent and 95 percent. The mean doses of simvastatin and niacin taken by patients were 13 ± 6 mg per day and 2.4 ± 2.0 g per day, respectively. Plasma vitamin concentrations increased significantly in 75 patients who received active vitamin therapy.
Vital status was ascertained at 38 months for all 160 patients enrolled. Follow‐up information for 159 patients was complete, including records of events from the patient's physicians.
The active agents and placebos were provided by: Simvastatin (Zocor, Merck, West Point, Pa.) slow‐release niacin (Slo‐Niacin, Upsher‐Smith, Minneapolis) crystalline niacin (Niacor, Upsher‐Smith).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Hogarth 1996.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United Kingdom.
Number of participants randomised: 106, 64 (44%) men and 42 (56%) women, mean age 82.65 years.
Inclusion criteria: elderly medical in‐patients.
Exclusion criteria: already taking nutritional supplements, had diabetes mellitus, dysphagia, or a body mass index > 25 or < 15 kg/m2.
Interventions Participants were randomly assigned to receive:
group 1: active energy and active vitamin supplementation (n = 31);
group 2: active energy and placebo vitamin supplementation (n = 24);
group 3: placebo energy and active vitamin supplementation (n = 23);
group 4: placebo energy and placebo vitamin supplementation (n = 28)
for one month period.
Supplementation was provided as 750 ml glucose drink (2317,5 kJ, 540 kcal) (Lucozade), and capsules containing vitamins A 8000 IU, B1 15 mg, B2 15 mg, B3 50 mg, B6 10 mg, C 500 mg. Matching placebos were Nutrasweet drinks or capsules of maize, starch and lactose.
Outcomes The primary outcome measures were: weight, serum albumin levels, and activities of daily living, cognitive functioning and length of stay.
Notes Compliance with the energy supplement (active or placebo) was monitored by measuring unconsumed fluid each day during admission. Following discharge, patients or carers were asked to complete a form estimating the volume of fluid (in quarters) remaining in each bottle each day. Vitamin compliance was monitored by tablet count at the end of the 1‐month period at the final assessments.
Compliance was poor with the liquid energy supplement with only one‐third of patients consuming > 50% of offered drinks during the study period (17/55 patients, active group; 16/51 patients, placebo group). Vitamin capsule compliance was higher with approximately 90% of patients taking more than 50% of the capsules provided (48/52 patients, active group; 49/54 patients, placebo group).
Eighty‐seven patients completed the trial (13 died, 6 withdrew). Three participants in supplemented group and 3 participants in placebo group withdrew.
The energy supplement (Lucozade) and placebo preparation were provided by SmithKline and Beecham.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Unclear risk The trial was described as blind, but the method of blinding was not described, so that knowledge of allocation was possible during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes High risk The number or reasons for dropouts and withdrawals were not described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

HOPE TOO 2005Low.

Methods The Heart Outcomes Prevention Evaluation Study (HOPE).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: International, North America, South America, Europe (19 countries) Number of participants randomised: 9541; 6996 males and 2545 females, 55 years old or older, mean age 66 years. Inclusion criteria: 55 years or older, had a history of CV disease (coronary artery disease, stroke or peripheral arterial disease) or diabetes in the presence of at least one additional CV risk factor (total cholesterol > 5.2 mmol/l, HDL cholesterol =0.9 mmol/l, hypertension, defined as use of medications to treat high blood pressure, or blood pressure at time of recruitment > 160 mmHg systolic or > 90 mmHg diastolic, known microalbuminuria, or current smoking). Exclusion criteria: Dipstick‐positive proteinuria, diabetic nephropathy, serum creatinine > 200 mmol/l, history of congestive heart failure, or known left ventricular ejection fraction (< 40%), hyperkalaemia, uncontrolled hypertension, myocardial infarction, unstable angina or stroke within 1 month before study enrolment, and use of or intolerance to vitamin E or angionetsin‐converting‐enzyme (ACE) inhibitors.
Interventions Patients were randomly assigned to receive either group 1: 400 IU of vitamin E (RRR‐a‐tocopheryl acetate) daily from natural sources (n = 4761); or group 2: matching placebo (n = 4780); or group 3: an angiotensin‐converting–enzyme inhibitor (ramipril 10 mg) (n = 4645); or group 4: matching placebo (n = 4652), once a day for a four to six years, mean 4.5 years. The Heart Outcomes Prevention Evaluation [HOPE] trial is conducted between December 21, 1993, and April 15, 1999. The Heart Outcomes Prevention was extended (HOPE‐The Ongoing Outcomes [HOPE‐TOO]) between April 16, 1999, and May 26, 2003. Of the initial 267 HOPE centres that had enrolled 9541 patients, 174 centres participated in the HOPE‐TOO trial. Of 7030 patients enrolled at these centres, 916 were deceased at the beginning of the extension of the trial, 1382 refused participation, 3994 continued to take the study intervention, and 738 agreed to passive follow‐up. The mean follow‐up period was 7 years.
Outcomes The primary outcome measures were: cancer incidence, cancer deaths, major cardiovascular events (myocardial infarction, stroke, and cardiovascular death). The secondary outcomes were: unstable angina, congestive heart failure, revascularisation or amputation, death from any cause, complications of diabetes, and cancer.
Notes Compliance with treatment was checked by measuring the plasma vitamin E levels in randomly selected patients. The rate of compliance with the assigned regimen was high throughout the study. The percentages of patients who were taking vitamin E in the vitamin E and placebo groups, respectively, were 94.2% and 1.0% at 1 year, 93.3% and 1.7% at 2 years, 91.3% and 2.0% at 3 years, 90.2% and 2.7% at 4 years, and 89.2% and 3.4% at the final visit. There was no significant interaction between the study treatments (ramipril and vitamin E) for the primary, secondary, and other study outcomes.
At the end of the initial HOPE trial, vital status was ascertained for 9535 (99.9%) of 9541 randomized patients. At the end of the HOPE‐TOO trial, vital status was ascertained for 4724 (99.8%) of 4732 patients who participated in the extension trial.
Funded by the Medical Research Council of Canada, Natural Source Vitamin E Association, Negma, Hoechst‐Marion Roussel, AstraZeneca, King Pharmaceuticals, and the Heart and Stroke Foundation of Ontario.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

HPS 2002Low.

Methods Heart Protection Study (HPS).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United Kingdom.
Number of participants randomised: 20,536; 15,454 males and 5082 females at an age 40 to 80 years.
Inclusion criteria: adults with coronary disease, other occlusive arterial disease, or diabetes, and non‐fasting blood total cholesterol concentrations of at least 3.5 mmol/L.
Exclusion criteria: other life‐threatening conditions, such as chronic liver disease, severe renal disease, severe heart failure, severe chronic airways disease, or diagnosed cancer (other than non‐melanoma skin cancer). In addition, anyone already taking high‐dose vitamin E supplements, or in whom such supplements were considered indicated, was not to be randomised.
Interventions Participants were randomly assigned to receive:
group 1: 600 mg vitamin E, 250 mg vitamin C, and 20 mg beta‐carotene daily (n = 10,269);
group 2: matching placebo capsules (n = 10,267), daily
during the scheduled five‐year treatment period.
Outcomes The primary outcome measures were: major coronary events (for overall analyses) and fatal or non‐fatal vascular events (for subcategory analyses), with subsidiary assessments of cancer and of other major morbidity.
Notes Compliance with treatment was assessed at each follow‐up by reviewing the calendar packed tablets remaining and, for those who had stopped, the reasons for doing so were sought. An average of 83% of participants in each treatment group remained compliant during the scheduled five‐year treatment period. To assess the effects of the treatment allocation on blood concentrations of the vitamins being studied, assays were performed in non‐fasting samples collected from about 5% of participants at the initial screening visit and at an average of about three years of follow‐up (the approximate mid‐point of the study).
There were 99.7% of the participants were with complete follow‐up for average of 5 years in vitamins allocated group and 99.6% in placebo group. Overall, 25 participants allocated to vitamins group and 35 participants to placebo group were lost to follow‐up.
Vitamins were provided by Roche.
Data were extracted from the primary publication, but additional information was received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ICARE 2008Low.

Methods ICARE trial.
Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Israel.
Number of participants randomised: 1434, 55 years of age or older, mean age 69 years, 52% females, with the Haptoglobin 2‐2 genotype.
Inclusion criteria: type II diabetes mellitus and 55 years of age or older.
Exclusion criteria: uncontrolled hypertension, myocardial infarction or stroke within one month before enrolment, unwillingness to stop antioxidant supplements, known allergy to vitamin E.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (d‐alpha tocopherol) 500 IU (n = 726);
group 2: placebo (n = 708).
Participants were supplemented daily and followed 1.5 year.
Outcomes The primary composite outcome was myocardial infarction, stroke, and cardiovascular death. Secondary outcome measures were: total mortality, hospitalisation for congestive heart failure, and coronary revascularization.
Notes Assessment of compliance was based on telephone interviews.
Two participants were lost to follow up (1 in each group). Seven individuals discontinued intervention because of advice from a physician (5 in vitamin E group, 2 in placebo). Eleven individuals discontinued the study because of perceived side effects (5 in vitamin E and 6 in placebo). Fifty‐five participants taking vitamin E and 61 participants taking placebo were noncompliant with taking the respective pills based on telephone interviews.
This work was supported by grants from the United States‐Israel Binational Science Foundation, Israel Science Foundation, Juvenile Diabetes Research Foundation, the Kennedy Leigh Charitable Trust, NIH, US Agency for Healthcare Research and Quality, and the Kaiser Permanente Center for Health Research.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Jacobson 2000Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 121, mean age 42, 58% males.
Inclusion criteria: adults 18 years of age and older who smoked one or more packs of cigarettes per day and were not currently taking the study vitamins.
Exclusion criteria: nondetectable polycyclic aromatic hydrocarbon PAH‐DNA adduct levels in mononuclear cells and plasma vitamin levels higher than 1.0 mg/dl for vitamin C, 15 mg/dl for beta‐carotene, and 1.2 mg/dl for alpha‐tocopherol at the first study visit.
Interventions Participants were randomly assigned to receive:
group 1: vitamin C 500 mg, alpha‐tocopherol 400 IU, beta‐carotene 6 mg (n = 60);
group 2: placebo (n = 61).
Participants were supplemented and followed 0.5 years.
Outcomes The primary outcome measure was: DNA damage.
Notes Compliance with treatment was not reported.
Seventy‐three participants completed the trial. Drop‐out rates were high in both groups, but were higher in the placebo (53%) than in the treatment (35%) group.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

LAST 2004Low.

Methods Lutein Antioxidant Supplementation Trial (LAST).
Randomised, double‐blind placebo‐controlled trial with parallel group design (three intervention groups).
Participants Country: United States of America.
Number of participants randomised: 90, 86 men and 4 women, mean age 74.7 years.
Inclusion criteria: diagnosis of atrophic age‐related macular degeneration (ARMD) by stereo bio‐ophtalmoscopy and at least one vision‐degrading visual‐psychophysical abnormality associated with ARMD in one or both eyes; clear non‐lenticular ocular media (cornea, aqueous and vitreous), free of advanced glaucoma and diabetes or any other ocular or systemic disease that could affect central or parafoveal macular visual function.
Exclusion criteria: recent (within 6 months) cataract or retinal surgery; taking photosenzing drugs (such as phenotiazines and chloroquine).
Interventions Patients were randomly assigned to receive:
group 1: lutein 10 mg (n = 29);
group 2: lutein 10 mg, and antioxidants/vitamins and minerals broad spectrum supplementation formula (n = 30);
group 3: placebo (maltodextrin) (n = 31);
taken as three capsules twice per day with food, over a period of 12 months.
The OcuPower supplement consists of: lutein 10 mg; vitamin A 2,500 IU; natural beta‐carotene 15,000 IU (Betatene R); vitamin C 1500 mg (as calcium ascorbate‐Ester C R); vitamin D3 400 IU; natural vitamin E (d‐alpha tocopherol succinate) 500 IU; vitamin B1 50 mg; vitamin B2 10 mg; vitamin B3 70 mg; vitamin B5 50 mg; vitamin B6 50 mg; vitamin B12 500 µg; folic acid 800 µg; biotin 300 mcg; calcium 500 mg; magnesium 300 mg; iodine 75 µg; zinc (as zinc L‐methionine‐L‐Optizinc R) 25 mg; copper 1 mg; manganese 2 mg; selenium 200 µg; chromium 200 µg; molibdenum 75 µg; lycopene 600 µg; bilberry extract (standardized to 25% anthocyanosides); alpha lipoic acid 150 mg; N‐acetyl cysteine 200 mg; quercetin 100 mg; rutin 100 mg; citrus bioflavonoids 250 mg; plant enzymes 50 mg; black pepper extract (Bioperine R) 5 mg; malic acid 325 mg; taurine 900 mg; L‐glycine 100 mg; L‐glutathione 10 mg; boron 2 mg.
Outcomes The primary outcome measures were: visual function and symptoms in atrophic age‐related macular degeneration (ARMD).
Notes Compliance was assessed by telephone at one week, two weeks, four weeks, six weeks, three months, and 12 months. Compliance with treatment was good. During one‐year study, 96% of the participants took approximately 92% of their assigned capsules. There was no difference in compliance among the three groups.
During the one year clinical trial, one, two, and one participant was lost to follow‐up from each group.
Lutein (Floraglo R) was provided by Kemin Foods International, Des Moines, Iowa); lutein in combination with additional antioxidants and nutrients (OcuPower R) and placebo were provided by Nutraceutical Sciences Institute, Boynton Beach, Florida.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Limburg 2005Low.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: China.
Number of participants randomised: 360, mean age 47, 42% males.
Inclusion criteria: at least 1 grossly visible oesophageal lesion with biopsy‐proven mild or moderate squamous dysplasia, according to histological interpretation by a single pathologist.
Exclusion criteria: history of cancer (except nonmelanoma skin cancer), symptoms suggestive of an upper gastrointestinal tract malignancy, recently treated peptic ulcer disease, or contraindications to the intervention agent(s) or study‐related procedures. Subjects were also excluded if a grossly visible lesion could not be confirmed during the baseline EGD or if the worst histological diagnosis at baseline was less than mild or greater than moderate dysplasia.
Interventions Participants were randomly assigned into four groups to receive:
group 1: selenium 200 µg and celecoxib 400 mg, (n = 90);
group 2: celecoxib 400 mg, (n = 90);
group 3: selenium 200 µg, (n = 90);
group 4: placebo, (n = 90).
Participants were supplemented and followed 10 months.
Outcomes The primary outcome measure was: change in histological grade of squamous dysplasia (determined by comparing the most advanced histological diagnosis for each subject at the baseline and end‐of‐trial evaluations) and was categorised as regression, stable, or progression.
Notes Compliance was assessed by pill counts and by direct observation by the village doctors who watched all participants take 2 morning pills each day throughout the intervention period. Compliance was further assessed biochemically by comparing baseline and end‐of‐trial serum selenium concentrations. Compliance with the single daily dose of selenomethionine (or placebo) and 1 of the 2 daily doses of celecoxib (or placebo) was in excess of 99% by both direct observation and pill counts.
Vital status was ascertained at the trial end in all patients in the vitamin E group and in 99.9% in the placebo group.
Pfizer, Incorporated, provided active and placebo celecoxib agent supplies.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Liu 2007Low.

Methods Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Canada.
Number of participants randomised: 763 elderly institutionalised people, 70% females, aged 65 to 103 years, mean age 85 years.
Inclusion criteria: elderly institutionalised people.
Exclusion criteria: younger than 65, expected to survive less than 6 months, unable to take whole or crushed tablets, receiving immunosuppressive drugs, already receiving multivitamin supplementation, renal failure (serum creatinine > 200 mmol/L), active malignancy, presence of an indwelling Foley catheter, or presence of severe protein‐energy malnutrition (body mass index < 16).
Interventions Participants were randomly assigned to receive:
group 1: multivitamin and multimineral tablet (vitamin A 1333 IU, beta‐carotene 16 mg, vitamin D 160 IU, vitamin E 74 IU, vitamin C 80 mg, thiamin 2.2 mg, riboflavin 1.5 mg, niacin 16 mg, vitamin B6 3 mg, vitamin B12 4 mg, folate 400 mg, calcium 200 mg, magnesium 100 mg, iron 16 mg, iodine 200 mg, zinc 14 mg, copper 1.4 mg, and selenium 20 μg) (n = 379);
group 2: matched placebo tablet (n = 384),
daily for 19 months.
Outcomes The primary outcome was number of infections per participant. Secondary outcomes were antibiotic use and hospitalisation rates.
Notes Pill counts were performed every 3 months for each subject.
During the 18‐month study surveillance period, similar numbers of subjects died or withdrew from the intervention and placebo arms.
Arrow Pharmaceuticals provided the supplement used in this study.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

MAVET 2006Low.

Methods Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
The Melbourne Atherosclerosis Vitamin E Trial (MAVET).
Participants Country: Australia.
Number of participants randomised: 409 male and female smokers aged 55 years and over, mean age 63.5 years, 55% females.
Inclusion criteria: 55 years of age or over and regularly smoke over five cigarettes per day.
Exclusion criteria: life‐threatening illness, previous carotid artery surgery or existing carotid stenosis warranting surgery, known sensitivity or intolerance to vitamin E, treatment with anticoagulant drugs, myocardial infarction or stroke within the previous six months or uncontrolled hypertension.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (d‐α tocopherol) 500 IU (n = 205);
group 2: matched placebo capsule (n = 204),
orally, daily for a period of four years.
Outcomes The primary outcome measure was progression of carotid atherosclerosis determined by intima–media thickness of the right common carotid artery. Secondary outcomes were change in systemic arterial compliance and low density lipoprotein (LDL) oxidative susceptibility over time.
Notes Following randomisation, participants were telephoned three‐monthly to encourage compliance. Compliance was determined by counting capsules in the returned medication bottles.
Overall 75.0 and 73.6% of the vitamin E and placebo groups, respectively, consumed 80% or more of their capsules. After 4 years of follow‐up, 83.4% of the vitamin E group and 79.4% of the placebo group remained on their assigned medication.
Vitamin E capsules were produced by Henkel Australia, New South Wales, Pty Ltd Australia.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

MAVIS 2005 Low.

Methods Mineral And Vitamin Intervention Study (MAVIS)
Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Scotland
Number of participants randomised: 910, 479 men and 431 women, aged 65 or over who did not take vitamins or minerals.
Inclusion criteria: all people aged 65 or over who were registered with the practices were eligible, irrespective of chronic illness, unless their doctors considered them too unwell.
Exclusion criteria: use of vitamin, mineral, or fish oil supplements in the previous three months (one month in the case of water soluble vitamins) or vitamin B12 injection in the past three months.
Interventions Participants were randomly assigned to receive:
group 1: multivitamin and multimineral supplement (800 µg vitamin A (acetate), 60 mg vitamin C, 5 µg vitamin D3, 10 mg vitamin E (D, L alpha‐tocopheryl acetate), 1.4 mg thiamin (mononitrate), 1.6 mg riboflavin, 18 mg niacin (nicotinamide), 6 mg pantothenic acid (calcium D‐pantothenate), 2 mg pyridoxine (hydrochloride), 1µg vitamin B12, 200 µg folic acid, 14 mg iron (fumurate), 150 µg iodine (potassium iodide), 0.75 mg copper (gluconate), 15 mg zinc (oxide), and 1 mg manganese (sulphate); or
group 2: matched sorbitol placebo,
one tablet daily for one year.
Tablets were purchased from a commercial supplier.
Outcomes The primary outcome measures were: number of contacts with primary care (doctor and other primary care workers, in person or by phone) for infection, number of self reported days of infection, and health related quality of life measured by the EuroQol and SF‐12.
The secondary outcome measures were: number of antibiotic prescriptions in primary care, number of days that antibiotics were prescribed, number of hospital admissions (including those related to infection), number of days in hospital with infection, number of infection related and all outpatient visits, adverse events reported by participants, and compliance with trial drugs (from diaries submitted monthly in all participants and tablet count at six and 12 months in a random sample of 10% of participants).
Notes Compliance with treatment was assessed by self‐report and was consistent with tablet counting.
There were no differences between the groups for compliance with drug taking. Compliance in participants still taking tablets and returning information in diaries was over 91% throughout the trial.
Only 13% (n = 121) of the participants were lost to follow‐up or reported stopping taking tablets. At least 1 diary was provided by 99% (901) of participants, 6 diaries by 93% (846), and 12 diaries by 89% (808). Losses to follow‐up was equal in the active and the placebo (n = 22) groups. Fourteen participants in the active group and 18 participants in the placebo group withdrew.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

McKeown‐Eyssen 1988.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Canada.
Number of participants randomised: 200. Fifteen participants had to be excluded after initial randomisation because none of the polyps was adenomatous. Of the 185 participants, 121 were males and 64 females. Mean age was 58 years.
Inclusion criteria: At least one polyp in the colon or rectum identified by colonoscopy and removed at two Toronto hospitals between 1979 and 1984. Patients who used supplements of vitamin C or E agreed to discontinue their use for the duration of the trial.
Of the 185 eligible participants 137 completed the study with a second colonoscopic examination.
Interventions Participants were randomly assigned to receive:
group 1. vitamin C 400 mg; vitamin E 400 mg (n = 96).
group 2: lactose placebos (n = 89);
over a period of two years.
Second colonoscopic examination was performed approximately two years after the initial examination, but could be performed earlier if judged clinically necessary. The physician assessed the presence and location of polyps, and any observed were removed.
Outcomes The primary outcome measure was: recurrence of colorectal polyps.
Notes Compliance with treatment was assessed by random urine samples collected at each visit from which urinary vitamin C levels were assessed, using a dipstick, as an index of compliance. The compliance to the vitamin supplements appears to be good. Of the 185 eligible participants, 137 (75%) completed the study with second colonoscopic examination conducted when most participants (81,5% of those on vitamins and 82,3% of those on placebos) had been receiving supplements for 12 to 30 months.
The losses to follow‐up were 14.1% in the vitamins group and 11.9% in the placebo group. Seventeen participants of 96 assigned to receive vitamins withdrew, and five were lost to follow‐up. Of 89 participants assigned to placebo, 15 withdrew and 4 were lost to follow‐up.
Trial agents were supplied by H. Newmark of Roche, New Jersey and Roche, Canada.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Meydani 2004Low.

Methods Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States.
Number of participants randomised: 617, 169 men and 448 women, mean age 84 years.
Inclusion criteria: aged 65 years or older; life expectancy greater than 6 months; no anticipated discharge within 3 months; not room‐bound for the past 3 months; absence of active neoplastic disease; no tube feeding, no kidney dialysis; no intravenous or urethral catheters for the last 30 days; no tracheostomy or chronic ventilator; antibiotic‐free for more than 2 weeks; no long‐term steroid treatment greater than 10 mg/d, no use of immunosuppressive drugs, or greater than the recommended daily allowance (RDA) level of supplements of vitamins E, C, or B6, selenium, zinc, beta‐carotene, or fish oil; body mass index of at least 18; serum albumin at least 3.0 g/dL; able to swallow pills; willing to receive influenza vaccine; and willing to provide informed consent (for patients with dementia, family members provided informed consent).
Interventions Participants were randomly assigned to receive:
group 1: 200 IU of vitamin E (dl‐alpha‐tocopherol) (n = 311);
group 2: placebo 4 IU of vitamin E (n = 306); both in soybean oil, one capsule daily for a period of one year.
All participants received a capsule containing half the recommended daily allowance of essential vitamins and minerals.
All participants received influenza vaccine.
Outcomes Primary outcomes of the trial were: incidence of, number of persons with, and number of days with respiratory tract infections (upper and lower), and number of new antibiotic prescriptions for respiratory tract infection. Secondary outcomes included emergency department visits, hospitalisation, and death. A post hoc subgroup analysis was performed to determine the effect of vitamin E on common colds.
Notes Adherence to trial protocol was verified by nursing home medication records, returned pill count, and quarterly measurement of plasma vitamin E levels. Ninety‐eight percent of those completing the trial consumed the capsules for at least 330 days (> 90% of the 1‐year supplementation period). The number of missed supplements did not differ statistically between the vitamin E and placebo groups.
Of the 617 randomised persons, 37% in the vitamin E and 36% in the placebo groups, respectively, completed the 1‐year trial period. Forty‐one participants in the vitamin group and 42 participants in the placebo group were lost to follow‐up. The losses to follow‐up were equal in both groups.
Capsules were manufactured by Tishcon Corporation (Westbury, NY). The capsules were packed by Pharmasource Healthcare Inc (Marlboro, Mass).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Mezey 2004Low.

Methods Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States and Spain.
Number of participants randomised: 51, 34 men and 17 women, mean age 48 years,
Inclusion criteria: age (18 years to 70 years), recent history of heavy alcohol ingestion and clinical and laboratory characteristics adopted by the International Informatics Hepatology Group for the diagnosis of alcoholic hepatitis. These criteria included moderate elevation of the serum aspartate aminotransferase AST (< 10 times above normal), an AST/alanine aminotransferase (ALT) ratio greater than 1.0 and no evidence of liver disease due to viral hepatitis, autoimmune disease, haemochromatosis, Wilsons disease or drug‐induced hepatitis.
Exclusion criteria: pregnancy, breast feeding, cardiovascular, pulmonary, kidney disease, pancreatitis, type I diabetes, recent (within 1 month) gastrointestinal bleeding, peptic ulcer disease, concurrent infection, history of thrombophlebitis, HIV positivity and history of ingestion of more than 100 IU vitamin E for the prior month.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (dl‐alpha‐tocopheryl acetate) 1000 IU ( n = 25)
group 2: placebo (n = 26); one capsule daily 3 months.
The patients were followed for 1 year after entry into the trial.
Outcomes The primary outcome measures was: clinical and laboratory parameters of liver function and on markers of fibrogenesis.
Notes Compliance with treatment was checked by serum assessments. Plasma alpha‐tocopherol levels increased in patients on vitamin E.
During the initial three‐month period of therapy, one patient in the treatment group withdrew from the trial. Four patients, 2 in each group, died during the initial three months. Five patients in the treatment group and 4 patients in the placebo group were lost between three and 12 months to follow up.
The authors published results of shorter (three months) and longer (one‐year) follow‐up period.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

MINVITAOX 1999Low.

Methods MIN.VIT.AOX geriatric network.
Randomised, double‐blind placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: France.
Number of participants randomised: 725 institutionalised elderly patients from 25 geriatric centres, 185 men and 540 women, aged 65 to 103 years, mean age 83.9 years.
Inclusion criteria: no acute illnesses and at least 65 years of age. Age‐related diseases were allowed.
Exclusion criteria: patients with a history of cancer or those taking medication that might interfere with nutritional status, immunocompetence, or vitamin or mineral supplementation.
Interventions Participants were randomly assigned to receive:
group 1: trace elements zinc sulfate and selenium sulfide (providing 20 mg of zinc and 100 µg of selenium), (n = 182);
group 2: vitamin group ‐ ascorbic acid (120 mg), beta‐carotene (6 mg), and alpha‐tocopherol (15 mg) (n = 180);
group 3: trace element and vitamin supplements (n = 181);
group 4: placebo capsules (n = 182).
Participants received one capsule daily, with their breakfast for two years.
Outcomes The primary outcome measures were: delayed‐type hypersensitivity skin response, humoral response to influenza vaccine, and infectious morbidity and mortality.
Notes Compliance with treatment was assessed first by the nursing teams that administered the pills every morning and then at the end of each six months by counting the remaining capsules in the pillboxes, and by random serum assessments. High compliance (> 85%) was observed.
Losses to follow‐up were approximately 2.2% in each treatment group. Three participants from the vitamin group, three participants from the vitamin and trace elements group, four participants from the trace elements group, and four participants from the placebo group withdrew before the end of the trial.
The supplements and placebo were provided by Produits Roche SA, Fontenay‐aux‐Roses, France.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Mooney 2005Low.

Methods Randomised, double‐blind placebo‐controlled parallel group trial.
Participants Country: United States.
Number of participants randomised: 284, 55% men and 45% women were aged 18 or older, mean age 36.8 years.
Inclusion criteria: men and women ages > 18 years who smoked at least 10 CPD, did not take vitamin supplements or use a nicotine patch in the 3 months before enrolment, had no prior history of cancer or liver disease, lived at a permanent address, owned a home telephone, were willing to comply with the 2‐year protocol, and completed a 1‐month placebo run‐in.
Exclusion criteria: none stated.
Interventions Participants were randomly assigned to receive:
group 1: vitamin C 500 mg and vitamin E 400 IU (n = 142);
group 2: placebo (n = 142);
for a period of 1.25 years.
Outcomes The primary outcome measure was: level of benzo(a)pyrene [B(a)P]‐DNA adducts as an intermediate cancer risk marker.
Notes Treatment compliance was assessed by serum vitamin measurements and pill counts. Compliance with treatment did not differ by gender measured by blood levels of {alpha}‐tocopherol at 15 months of follow‐up or by pill counts. At all time points after randomisation, in all participants, the treatment group had significantly higher levels of vitamin E than the placebo group. However, in women, the blood levels of vitamin E did not plateau until the 9‐month time point.
Eighty‐three of 142 (58%) in the treatment group and 93 of 142 (66%) in the placebo group completed 15 months of treatment.
Trial agents were provided by Hoffman‐LaRoche.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Murphy 1992Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design.
Participants Country: United States of America.
Number of participants randomised: 109, 37 men and 72 women, mean age 73 years.
Inclusion criteria: patients on the chronic medical and skilled/intermediate care wards of an academically affiliated nursing home.
Exclusion criteria: patients in the rehabilitation unit.
Interventions Participants were randomly assigned to receive:
group 1: vitamin A 60,000 µg retinol equivalent (200,000 IU) (n = 53);
group 2: placebo (vitamin A 300 retinol equivalents (1000 IU) as retinyl palmitate in arachis oil (n = 56).
All capsules contained 40 IU of vitamin E as an antioxidant.
Patients receiving multivitamin preparations at the onset of the trial continued to receive them. No patient was commenced on vitamin A‐containing supplements during the follow‐up period.
A content of a single capsule was given to participants by research assistant. Participants were followed‐up for 90 days.
Outcomes The primary outcome measure was: incidence of antibiotic treated bacterial infections among elderly nursing‐home residents.
Notes Participants were given a content of a single capsule by a research assistant.
Due to the administration of vitamin A in the control group, the 'no intervention' in that group can be discussed.
Two patients from vitamin A group were lost to follow‐up.
Trial capsules were provided by Roche, Basel, Switzerland.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

NIT1 1993.

Methods Nutrition Intervention Trial (NIT); The General Population Trial, in Linxian, China.
Randomised, placebo‐controlled trial with one‐half replicate of a two‐by‐two‐by‐two‐by‐two factorial design.
Participants Country: China, Henan Province of north central China, Linxian County.
Number of participants randomised: 29584; 45% males, aged 40 to 69 years.
Inclusion criteria: residents willing to take part in a multi‐year, daily pill‐taking regimen.
Exclusion criteria: debilitating disease or prior oesophageal or stomach cancer.
Interventions Participants were randomly assigned to receive one of eight vitamin/mineral supplement combinations in the form of individual oral tablets. The eight intervention groups (each with 3677 to 3709 participants) were derived from a one‐half replicate of a two‐by‐two‐by‐two‐by‐two factorial design which allowed to asses four factors (ie, nutrient combinations) in a single experiment. The four factors designated by the letters A, B, C, D were:
A ‐ retinol (as palmitate) 5000 IU, zinc (as zinc oxide) 22.5 mg;
B ‐ riboflavin (vitamin B2) 3.2 mg and niacin (vitamin B3) 40 mg;
C ‐ ascorbic acid 120 mg and molybdenum (as molybdenum yeast complex) 30 µg;
D ‐ beta carotene 15 mg, selenium (as selenium yeast) 50 µg, and alpha‐tocopherol 30 mg.
Doses of each nutrient varied from one to two times US Recommended Daily Allowances (RDAs).
The eight intervention groups were defined by the following combinations of supplements; AB, AC, AD, BC, BD, CD, ABCD, or placebo and packed in coded bottles containing a one‐month supply. Bottles were distributed monthly beginning in March 1986 and continuing through May 1991, average 5.25 years.
Outcomes The primary outcome measures were: cancer incidence, cancer mortality, and overall mortality.
Notes Compliance with study treatment was assessed by monthly pill counts and biochemical measures.
Compliance was excellent throughout the study. The overall pill disappearance rate was 93% for all participants, with no difference by treatment group (range 92% to 93%) and little change during the trial.
Losses to follow‐up not reported.
All vitamin/mineral supplements and placebos were provided by Hoffmann‐La Roche, Basel, Switzerland and Lederle Laboratories, Inc.
Data were extracted from the primary publication.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Squence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes High risk The number or reasons for dropouts and withdrawals were not described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

NIT2 1993Low.

Methods The Dysplasia Trial.
Randomised, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: China, Henan Province of north central China, Linxian County.
Number of participants randomised: 3318; 1461 males and 1857 females at age 40 to 69 years, mean age 54 years.
Inclusion criteria: place of living in one of the three northern Linxian communes (Yaocun, Rencun, or Donggang), provided consent, diagnosis of oesophageal dysplasia on a balloon cytology examination.
Exclusion criteria: taking vitamins of any type regularly, or antitumour B (a traditional Chinese drug consisting of a mixture of six medical herbs), history of malignancy or other debilitating disease.
Interventions Participants were randomly assigned to receive:
group 1: 14 vitamins and 12 minerals (vitamin A (acetate) 10000 IU; vitamin E (dl‐alpha tocopherol acetate) 60 IU, vitamin C (ascorbic acid) 180 mg, vitamin B1 5 mg, vitamin B2 5.2 mg, vitamin B6 6 mg, vitamin B12 18 µg, vitamin D 800 IU; beta‐carotene 15 mg, folic acid 800 µg, niacinamide 40 mg, biotin 90 µg, pantothenic acid 20 mg, calcium 324 mg, phosphorus 250 mg, iodine 300 µg, iron 54 mg, magnesium 200 mg, copper 6 mg, manganese 15 mg, potassium 15.4 mg, chloride 14 mg, chromium 30 µg, molybdenum 30 µg, selenium (sodium selenate) 50 µg, and zinc (n = 1657);
group 2: placebo (n = 1661);
for a period of 6 years.
The doses were typically two to three times the US Recommended Daily Allowances (RDAs), but ranged from 0.26 to seven times the RDA depending on the vitamin or mineral. Each participant was given three pills daily, including one capsule beta‐carotene or placebos and two tablets of vitamin/mineral supplement, or placebos.
Outcomes The primary outcome measures were: cancer incidence, cancer mortality, and overall mortality.
Notes Compliance with treatment was assessed by counting unused pills for all trial participants and by assessing nutrient levels in blood collected from samples of individuals randomly selected without replacement every three months throughout the trial.
Compliance with treatment was excellent. The overall pill disappearance rate was 94% in both groups with slight decline (from 96% in year 1 to 92% in year 6 in both groups) over the duration of the trial.
The morbidity and mortality follow‐up rates were 99%.
Data were extracted from the primary publication.
Active medications and placebos were provided: beta‐carotene as Solatane by Hoffmann‐La Roche, Inc., Nutley, N.Y., and vitamin/mineral supplement as Centrum Lederle Laboratories, Inc., Pearl River, N.Y.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

NPCT 1996Low.

Methods Nutritional Prevention of Cancer Trial (NPCT).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 1312; 75% males, aged 18 to 80 years, mean age 63 years.
Inclusion criteria: history of two or more basal cell skin cancers (BCC) or one squamous cell skin cancer (SCC), with one of this occurring within the year prior the randomisation, life expectancy of at least five years and no internal malignancies treated within the previous five years.
Exclusion criteria: history of significant liver or kidney disorders.
Recruitment began on September 15, 1983 and continued each year through 1991.
Interventions Patients were randomly assigned to receive:
group 1: 200 µg of selenium supplied in a 0.5 g high‐selenium bakers yeast tablet (n = 653);
group 2: placebo (n = 659);
The end of a blinded period of treatment was on February 1, 1996. Mean length of treatment was 4.5 years and follow‐up 7.4 years.
Outcomes The primary outcome measures were: incidences of basal cell and squamous cell carcinoma of the skin.
In 1990 secondary outcome measures were identified, which included: total mortality and cancer mortality, as well as the incidence of the lung, colorectal, and prostate cancers.
Notes Compliance with treatment: excellent, 79.3% of the participants (80.3% in the placebo group and 78.4% in the selenium group) missed taking a pill less than twice a month.
At the end of the blinded period of treatment no participants were lost to vital follow‐up, and only 7 participants (3 in the selenium group and 4 in the placebo group) declined to provide additional information about the illness.
Trial medications were provided by Nutrition 21 (La Jolla, CA), through 1995 and by Cypress Systems (Fresno, CA) thereafter.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

NSCPT 1999Low.

Methods Nambour Skin Cancer Prevention Trial (NSCPT).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: Australia.
Number of participants randomised: 
 1621, 708 men and 913 women, aged between 20 and 69 years, mean age 48.8 years.
Inclusion criteria: examination by a dermatologist with removal of all diagnosed skin cancers, written informed consent.
Exclusion criteria: taking vitamin supplements containing beta‐carotene and those who reported that they were already applying sunscreen on a strict daily basis.
Interventions Patients were randomly assigned to receive:
group 1: sunscreen and beta‐carotene (30 mg) (n = 404);
group 2: sunscreen and oral placebo (n = 408);
group 3: beta‐carotene (n = 416);
group 4: oral placebo (n = 393);
one tablet daily for a period of 4.5 years.
Use of a placebo skin cream is considered unethical from two points of view: an oil‐in‐water emulsion with no active chemicals may enhance ultraviolet damage after evaporation of the water component; and people in the trial may use the placebo skin cream rather than a protective sunscreen in situations that lead to sunburn. The treatment protocol involves the self‐application of an adequate layer of sunscreen to all exposed sites on the head, neck, and upper limbs every morning, after heavy sweating or bathing.
Outcomes The primary outcomes were: incidence of basal‐cell and squamous‐cell carcinomas both in terms of people treated for newly diagnosed disease and in terms of the numbers of tumours that occurred. Analysis of the effect of sunscreen was based only on skin cancers that developed on sites of daily application.
Notes Compliance is assessed on a 3‐monthly basis when supplies of sunscreens and tablets are replenished comparing the weight of sunscreen used to an empirical standard usage rate, by counting the remaining tablets, and by determining serum beta‐carotene in a random sample of participants at 12 and 54 months.
At the end of the trial, 75% of participants who were assigned daily sunscreen use were applying sunscreen to their head, neck, arms, and hands at least 3 or 4 days per week. Of those people not assigned to the sunscreen group, 74% were applying sunscreen to head, neck, and arms either not at all or no more than 1 or 2 days per week. Self‐reported frequency of sunscreen application was well correlated with estimated daily weight of sunscreen used (averaged across the entire intervention period) in the sunscreen group.
On the basis of counts of returned tablets, 72% of the beta‐carotene group and 70% of the placebo group took at least 80% of the prescribed tablet intake over the entire intervention period. The beta‐carotene group had significantly greater mean skin reflectance on the palm at follow‐up than at baseline and their follow‐up values were greater than those of the placebo group.
At the end of the trial after five years, 15% participants had withdrawn without a complete skin examination by a dermatologist in the follow‐up period. Fifty participants, assigned to sunscreen and beta carotene, 70 assigned to sunscreen and placebo, 59 assigned to no sunscreen and beta carotene, and 59 assigned no sunscreen and placebo were lost to follow‐up.
Study agents were supplied by Hoffman‐La Roche, Nutley, NJ (beta‐carotene), and broad‐spectrum, sun protection factor 15+ sunscreen supplied by Woolworths Limited, Sydney, Australia, under the brand Auscreen Ultrablock Lotion SPE 15+ Ross Cosmetics, Melbourne, Australia.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Penn 1991.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United Kingdom.
Number of participants randomised: 30; 20% males, mean age 83.7 years.
Inclusion criteria: patients who had been in hospital for more than 3 months, requiring nursing care as a consequence of stroke disease, but without active medical problem.
Exclusion criteria: patients who were catheterised, or who had pressure sores, or who were receiving medication known to affect immune function.
Interventions Participants were randomly assigned to receive:
group 1: vitamin A 8000 IU, vitamin C 100 mg, and vitamin E 50 IU (n = 15);
group 2: placebo (n = 15);
for 28 days.
Outcomes The primary outcome measures were: nutritional status and cell‐mediated immune function.
Notes Compliance with treatment is not described.
One patient from each group was lost to follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Selective reporting (reporting bias) Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

PHS 1996Low.

Methods Physicians Health Study (PHS).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America.
Number of participants randomised: 
 22071 US male physicians at age 40 to 84 years, mean age 53 years.
Inclusion criteria: US male physicians willing to take part in this trial.
Exclusion criteria: chronic liver disease or evidence of abnormal liver function, severe renal disease or evidence of impaired renal function, inflammatory muscle disease or evidence of muscle problems (creatine kinase > 750 IU/L); concurrent treatment with cyclosporin, fibrates, or high‐dose niacin; child‐bearing potential; severe heart failure; some life‐threatening condition other than vascular disease or diabetes (eg, severe chronic airways disease or any cancer other than non‐melanoma skin cancer); or conditions that might limit long‐term compliance (eg, severely disabling stroke, dementia, or psychiatric disorder).
Interventions Physicians were randomly assigned to one of the four groups including:
group 1: active aspirin 325 mg on alternate days plus beta‐carotene placebo;
group 2: active beta‐carotene 50 mg on alternate days plus aspirin placebo;
group 3: both active agents; or
group 4: both placebos.
The randomised aspirin component of the study was terminated early, on 25 January 1988. The beta‐ carotene component continued uninterrupted until its scheduled end in December 1995.
A total of 11036 physicians were assigned at random to receive beta‐carotene and 11035 to receive beta‐carotene placebo.
Time from randomisation to the end of study averaged 12 years, and time of follow‐up 12.9 years.
Outcomes The primary outcome measures were: overall and within subgroups, incidence of malignant neoplasms (except non melanoma skin cancer), incidence of cardiovascular disease, and overall mortality.
Notes Compliance with treatment was checked by random serum assessments obtained at unannounced visits to trial participants. Compliance with treatment excellent, the average per cent of pills taken was 97% in both the active and placebo groups. There was 85% compliance with beta‐carotene treatment after five years and 78% after 12 years. The use of vitamin A supplements was reported by only 6% of the placebo group even by the end of trial.
By December 31, 1995, the scheduled end of the trial, less than 1% of the participants were lost to follow up.
Active trial packs and matching placebos were provided by: aspirin (Bufferin) by Bristol Meyers; beta‐carotene (Lurotin), BASF corporation.
Additional information received through personal communication with the authors.
Data were extracted from the article: Cook et al. Effects of beta‐carotene supplementation on cancer incidence by baseline characteristics in the Physicians' Health Study (United States). Cancer Causes and Control 2000; 11: 617‐26, with extended follow‐up of 12.9 years.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

PHS 2008Low.

Methods Physicians Health Study II (PHS II).
Randomised, double‐blind, placebo‐controlled trial using two‐by‐two‐by‐two‐by‐two factorial design.
Participants Country: United States of America.
Number of participants randomised: 14 641 US male physicians, aged 50 years or older, mean age 64.3 years.
Inclusion criteria: US male physicians willing to take part in this trial, and willing to forgo during the course of PHS II any current use of multivitamins or individual supplements containing more than 100% of the recommended daily 
 allowance of vitamin E, vitamin C, beta carotene, or vitamin A.
Exclusion criteria: a history of cirrhosis, active liver disease, were taking anticoagulants, or reported a serious illness that might preclude participation.
Interventions Physicians were randomly assigned to receive:
group 1: active vitamin E (synthetic alpha tocopherol) 400 IU every other day, ;
group 2: placebo vitamin E 400 IU every other day;
group 3: active vitamin C 500 mg (synthetic ascorbic acid) daily ;
group 4: placebo vitamin C 500 mg daily ;
group 5: active beta‐carotene 50 mg every other day;
group 6: active beta‐carotene 50 mg every other day;
group 7: multivitamin daily (Centrum silver);
group 8: placebo multivitamin daily
for a mean period of 8 years, median 7.6 years.
Outcomes The primary outcome measures were cardiovascular diseases, cancer, and mortality. Secondary outcome measure was adverse events.
Notes Participants were sent monthly calendar packs, containing vitamin E or placebo (taken every other day), and vitamin C or placebo (taken daily), every 6 months for the first year and annually thereafter.
Participants also were sent annual questionnaires asking about adherence, potential adverse events, the occurrence of new end points, and updated risk factors. Treatment and follow‐ up continued in a blinded fashion through August 31, 2007, the scheduled end of the vitamin E and C components of PHS II.
Beta‐carotene component (50 mg Lurotin or placebo on alternate days; BASF Corporation), was terminated on schedule in March 2003.The multivitamin component is still ongoing.
Morbidity and mortality follow‐up were extremely high at 95.3% and 97.7%, respectively.
Adherence was defined from participant self‐reports as taking at least two thirds of the study agents. For active vitamin E and its placebo, adherence among participants at 4 years was 78% and 77%, respectively (P = 0.12), and at the end of follow‐up (mean of 8 years), 72% and 70% (P = 0.004). For active vitamin C and its placebo, adherence among participants at 4 years was 78% and 78%, respectively (P = 0.99), and at the end of follow‐up, 71% and 71% (P = 0.54).
The trial was supported by grant from BASF Corporation (Florham Park, New Jersey). Study agents and packaging were provided by BASF Corporation, Wyeth Pharmaceuticals (Madison, New Jersey), and DSM Nutritional Products Inc (formerly Roche Vitamins) (Parsippany, New Jersey).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias.

Pike 1995Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 47; 13 men and 34 women at age 61 to 79, mean age 69 years.
Inclusion criteria: healthy, noninstitutionalised elderly participants with no known chronic or serious medical illness (eg, cancer, end stage renal disease, chronic liver disease).
Exclusion criteria: taking nutritional supplements 3 months before the trial start.
Interventions Participants were randomly assigned to receive:
group 1: micronutrient supplement (Multivitol R) containing vitamin A (retinol acetate) 800 retinol equivalents (RE); vitamin D2 (ergocalciferol) 5.0 µg; vitamin E (alpha‐tocopherol acetate) 45 mg; vitamin B1 (thiamin mononitrate) 2.18 mg; vitamin B2 (riboflavin) 2.6 mg; vitamin B6 (pyridoxin hydrochloride) 3.65 mg; vitamin B12 (cyanocobalamin) 9 µg; nicotinamide 30 mg; folic acid 0.4 mg; vitamin C (ascorbic acid) 90 mg; calcium (calcium hydrogen phosphate 2H2O 695 mg) 162 mg; magnesium (magnesium oxyde 165.78 mg) 100 mg; iron (iron (II) fumarate 82.14 mg) 27 mg; copper (copper (II) oxyde 1.87 mg) 1.5 mg; zinc (zinc oxide 28 mg) 22.5 mg; iodine (potassium iodide 0.294 mg) 0.225 mg); (n = 24);
group 2: placebo: (n = 23);
one tablet daily for a period of one year.
Outcomes The primary outcome measure was: immune indices in healthy elderly.
Notes Compliance was verified by interview with the patient during three‐monthly visits to the centre and through morbidity forms, phone calls, and checking supplement containers when brought back to the centre.
Five participants in the placebo group and seven in the supplemented group did not complete the trial.
Trial agents were provided by Hermes Arzneimittel GmbH.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Plummer 2007Low.

Methods Randomized, double‐blind, placebo‐controlled, primary‐prevention trial with parallel group design.
Participants Country: Venezuela
Number of participants randomised: 1980, aged 35 to 69 years, 52.7% females.
Inclusion criteria: population at high risk for stomach cancer in general good health, and permanent residents of Tachira State.
Exclusion criteria: serious illness, including any type of cancer, those whose mental status made long‐term adherence to the treatment regimen unlikely and pregnant women.
Interventions Participants were randomly assigned to receive:
group 1: vitamin C (750 mg), vitamin E (600 mg), and beta‐carotene (18 mg) (n = 990);
group 2: placebo (n = 990);
daily for 3 years.
The treatment was taken in the form of three capsules per day, one with each of the three main meals. Each capsule contained 250 mg vitamin C, 200 mg vitamin E, and 6 mg of beta‐carotene, for a daily dose of 750 mg of vitamin C (12.5 times the recommended daily allowance), 600 mg vitamin E (20 times the recommended daily allowance), and 18 mg beta‐carotene (considered the maximum dose if carotenoderma is to be avoided).
Outcomes The primary outcome of the trial was the progression and regression of precancerous lesions of the stomach, as determined by histologic findings.
Notes Compliance for the intervention group was confirmed by the pill counts and measuring the biochemical markers of supplementation. Excellent compliance was indicated by pill counts when participants returned for their vitamin pills: 91% of all containers were returned with less than 10% of pills. There were clear increases in beta‐carotene and vitamin E levels in the treated group beyond the levels observed at baseline. In the placebo group, by contrast, no changes were observed. Participants who did not return for their supply of capsules were contacted first by telephone, then visited at home by social workers who enquired about the reasons for nonattendance, encouraged continuing participation, and provided the next month's supply of capsules.
Overall 302 participants from active and 278 participants from placebo group dropped‐out during the trial. The number of participants who dropped out was slightly higher in the vitamin group than in the placebo group, but the difference was not statistically significant (P = 0.14, for difference of two proportions).
Both vitamin capsules and placebo were supplied by Hoffman‐La Roche.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

PPP 2001.

Methods The Primary Prevention Project (PPP)
Randomised controlled clinical trial with two‐by‐two factorial design.
Participants Country: Italy.
Number of participants randomised: 4495; 1912 males and 2583 females, mean age 64.4 years.
Inclusion criteria: old age (> 65 years); hypertension (systolic blood pressure > 160 mmHg or diastolic blood pressure > 95 mm Hg on at least three separate occasions); hypercholesterolaemia (total blood cholesterol > 6.4 mmol/L on at least two separate occasions); diabetes mellitus (fasting venous plasma glucose concentration > 7.8 mmol/L on at least two separate occasions (chronic drug treatment for any of the three latter conditions was also a criterion for inclusion); obesity (body mass index > 30 kg/m2); and family history of myocardial infarction before 55 years of age in at least one parent or sibling.
Exclusion criteria: treatment with antiplatelet drugs (history of vascular events or diseases); chronic use of anti‐inflammatory agents or anticoagulants; contraindications to aspirin; diseases with predictable poor short‐term prognosis; and predictable psychological or logistical difficulties affecting compliance with the trial requirements.
Interventions Patients were randomly assigned to receive:
group 1: aspirin, 100 mg (enteric‐coated aspirin a day) (n = 2226); or
group 2: no aspirin (n = 2269); and
group 3: vitamin E (one capsule of 300 mg synthetic alpha‐tocopherol a day), (n = 2231) or;
group 4: no vitamin E (n = 2264),
following a two‐by‐two factorial design.
The mean follow‐up was 3.6 years.
Outcomes The primary outcome measure was: the cumulative rate of cardiovascular death, non‐fatal myocardial infarction, and non‐fatal stroke. Predefined analyses included cardiovascular deaths, total deaths, total cardiovascular events (cardiovascular death, nonfatal myocardial infarction, non‐fatal stroke, angina pectoris, transient ischaemic attacks, peripheral artery disease, and revascularization procedures).
Notes At the beginning, and repeatedly during the trial, all patients received advice on compliance with background treatments. Compliance with treatments: at year 1 and at the end of the study 19.2% and 19.3% of the patients randomised to aspirin and 13.1% and 13.6% of those randomised to vitamin E had stopped taking the treatment. Side effects were the reason for discontinuation for 7.9% of the patients in the aspirin group and 1.1% in the vitamin E group. At the end of the trial, 7.2% of the patients not randomised to aspirin were taking aspirin and 0.2% of those not randomised to vitamin E were taking vitamin E.
At the end of the trial, 4150 (92.3%) patients had clinical follow‐up. For 314 (7.0%) participants, information on vital status was obtained through census offices. Overall, vital status information was obtained for 99,3% of the population enrolled. Fourteen participants assigned to vitamin group and 17 assigned to control group were lost to follow‐up.
Bayer supplied the aspirin preparation, and vitamin E capsules were provided by Bracco SpA.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes High risk The trial was not blinded, so that the allocation was known during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias.

PPS 1994Low.

Methods The Polyp Prevention Study (PPS).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America.
Number of participants randomised: 864; 751 participants completed the study, 592 males and 159 females, mean age 61 years.
Inclusion criteria: at least one adenoma diagnosed within the previous three months, patients have undergone colonoscopy with the entire large bowel seen and judged to be free of further polyps, good health, age less than 80 years.
Exclusion criteria: familial polyposis, a history of invasive colorectal cancer, malabsorption syndromes, or any condition (such as a history of renal calculi or thrombophlebitis) that might be worsened by dietary supplementation with vitamin C or E. 
 Participants agreed not to take supplemental vitamin C or E or beta carotene outside the trial.
The trial protocol called for two follow‐up colonoscopic examinations, the first approximately one year after the colonoscopy that qualified the patient for study (year 1), and second 36 months after the first (year 4). A colonoscopy was considered to be satisfactory for study purposes if cecum was reached, the entire mucosa was seen, and all polyps were removed. The endoscopist recorded the size and location of all raised mucosal lesions.
Interventions Patients were randomly assigned to receive:
group 1: beta carotene 25 mg, vitamin C 1000 mg, vitamin E (dl‐alpha‐tocopherol) 400 mg (n = 208);
group 2: vitamin C 1000 mg, vitamin E 400 mg, and placebo (n = 225);
group 3: beta carotene 25 mg plus placebo (n = 217);
group 4: placebo (n = 214);
daily for four years.
The study agents were provided in the form of soft gelatine capsules (containing placebo, beta carotene alone, vitamin E alone, or beta carotene plus vitamin E) and tablets (containing placebo or vitamin C) packaged in calendar packs, with each day's blister containing one capsule and one pill.
Outcomes The primary trial outcome was: the occurrence of new adenomas between the colonoscopic examinations conducted at year 1 and year 4.
Notes Compliance was checked by random serum assessments. Compliance with treatment was good, 82% of all patients reported taking the study agents at least six days per week, and further 5% took them three to five days per week. Only five patients stopped taking the medications because of their presumed toxicity.
Of 864 patients randomised, 751 (87%) underwent follow‐up colonoscopic examinations and provided all subsequent data. Overall, 69 participants were lost to follow‐up; 56 in three active intervention groups and 13 in placebo group.
Trial agents were provided at no cost by BASF of of Wiandotte, Michigan.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Prince 2003Low.

Methods Randomised, double‐blind, placebo‐controlled cross‐over trial (two intervention groups).
Participants Country: United Kingdom.
Number of participants randomised: 
 61 patient with primary biliary cirrhosis, 92% women, mean age 58 years.
Inclusion criteria: primary biliary cirrhosis and self‐reported fatigue.
Exclusion criteria: change in disease (or symptom) altering medication in the 3 months prior to randomisation (e.g. ursodeoxycholic acid, colestyramine, rifampicin), current use or use within the last 3 months of nutritional supplements containing antioxidants, inability to complete symptom severity assessment documents, life‐threatening intercurrent disease; presence of other uncontrolled disease with fatigue forming part of its clinical spectrum (eg, hypothyroidism, anaemia, renal failure, depression); drug dependency or addiction; women of child‐bearing potential who were not practising effective contraception.
Interventions Participants received 12 weeks each of placebo and antioxidant supplementation (vitamins A, C and E, selenium, methionine and ubiquinone) in random order, separated by a four‐week washout period.
Active medication consisted of four gelatine‐covered capsules daily containing selenium (l‐selenomethionine) 75 µg, beta‐carotene 3 mg, vitamin E (d‐alpha‐tocopherol acetate) 50 mg, vitamin C 150 mg, l‐methionine 375 mg, and ubiquinone (coenzyme Q10) 25 mg.
Placebo consisted of identical‐looking capsules containing inactive carrier.
Participants were requested not to start any other nutrient supplements or complementary therapies during the trial.
Forty‐three (70%) patients were co‐prescribed ursodeoxycholic acid. The median dose (interquartile range) prescribed was 9.6 mg/kg (8.5–11.3 mg/kg). Fifteen (25%) patients were taking thyroxine for pre‐existing hypothyroidism. All patients had normal thyroid stimulating hormone levels and had been on stable thyroxine doses for at least 3 months prior to enrolment. Two patients were taking long‐term beta‐adrenergic blocking medication (one each propranolol and sotalol). Three patients were long‐term users of benzodiazepines.
Outcomes The primary outcome measure for this study was: the change in patient fatigue. Fatigue was assessed using the Fisk fatigue severity score (FFSS). The FFSS assesses the impact of fatigue‐associated impairment in three domains (physical, cognitive and psychosocial) that can be summed to give a total score. Higher scores relate to increased fatigue severity.
Notes Forty‐four (72%) patients completed the trial per protocol. One patient died from previously undiagnosed Ischaemic heart disease during the first (active) treatment period. Eight further patients (5 from the active group) withdrew from the trial during the first treatment period and 8 (4 from the active group) withdrew during the second treatment period.
Trial medications were provided by Bioquantox, Pharma Nord, Morpeth, UK.
Additional information received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

REACT 2002Low.

Methods The Roche European American Cataract Trial (REACT).
Randomised, double blind, placebo‐controlled, trial with parallel groups design (two intervention groups).
Participants Country: United States and United Kingdom
Number of patients randomised: 297, mean age 68 years, 59 % females.
Inclusion criteria: at least one eye met the following ocular criteria: cataract extraction unlikely within two years, immature idiopathic 'senile' cataract present in one or both eyes, (U.S. patients) presence of minimal cataract by Lens Opacities Classification System (U.K. patients) presence of cataract of minimal Oxford grade: cortical and posterior subcapsular grades: grade I; nuclear brunescence: grade II; and white nuclear scatter: grade II. If both eyes met the inclusion criteria, and cataracts were of different types, the cataract type in the eye with the worse visual acuity determined the group for randomisation. If an eye had more than one type of cataract, the morphological type that in the clinicians opinion was the more destructive to visual acuity determined the group for randomisation, no visually significant fundus pathology, no clinical signs of glaucoma and intraocular pressure, no history of amblyopia, eye surgery, argon or YAG laser eye treatment, or major eye trauma, no history of iritis, retinal crystalline deposits, or optic nerve disease, no extended (daily for >3 months) use of ocular corticosteroid or glaucoma therapy, no participation in another clinical trial investigating an anticataract formulation within the last year.
Exclusion criteria: pregnancy, insulin dependent diabetes mellitus, severe renal failure or kidney stones, fat malabsorption syndrome, history of major intestinal surgery, chronic diarrhoea, alcoholism, extended use (daily for > 3 months) of systemic corticosteroid treatment, use of anticoagulants, or regular use of any vitamin supplement.
Interventions Patients were randomly assigned to receive:
group 1: 600 mg vitamin E (all‐rac alpha‐tocopherol acetate), vitamin C 750 mg, and beta‐carotene 18 mg (n = 149);
group 2: placebo (n = 148);
The actual supplementation period ranged from 2 to 51 months, 231 patients were followed for at least two years, 158 patients for at least three years and 36 patients for at least four years. Patients remained in the study for 34 months.
Outcomes The primary outcome was: the measure of area, 'increase % pixels opaque' cataract severity documented with serial digital retroillumination imagery of the lens; progression was quantified by image analysis assessing increased area of opacity.
Notes Compliance with treatment was checked by serum assessments. The plasma concentrations of vitamin C, vitamin E, and beta‐carotene in the treated and placebo groups were maintained at consistent levels throughout the trial indicating excellent compliance with instructions about the use of the trial medication. There appeared to be little if any supplementation of placebo with other vitamins or failure to take the vitamin capsules.
The pattern of drop‐outs was similar in the groups, and drop‐outs created no imbalances between the placebo and treatment groups. There were no differences noted between the vitamin and placebo groups regardless of the length of follow‐up. Overall 59 participants in the vitamin group and 68 participants in the placebo group were lost to follow‐up.
The work was supported by grants from F. Hoffmann‐La Roche, Ltd., Basel, Switzerland, and Roche Vitamins, Inc., Parsippany, NJ.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Sasazuki 2003.

Methods Randomised, double‐blind placebo‐controlled trial with three‐by‐three, then two‐by‐two factorial design and then parallel group design.
Participants Country: Japan.
Number of participants randomised: 
 439, 35% men and 65% women, aged 40 to 69 years, mean age 57 years.
Inclusion criteria: men and women living in four municipalities (3 towns and one village of Yokote Public Health Centre District in Akita prefecture, participated in annual screening programmes for circulatory diseases with chronic atrophic gastritis (determined by serum pepsinogen (PG) levels (PG I < 70 ng/ml and PG I/PG II ration < 3.0).
Exclusion criteria: past history of gastric cancer or surgery, liver cancer or cirrhosis, and other cancers within 5 years; abnormal liver function (AST > 100 IU/L, ALT > 100 IU/L, or ALP > 800 IU/L), use of supplements containing beta‐carotene or vitamin C, unable to follow‐up for at least one year.
Interventions Participants were randomly assigned to receive:
group 1: vitamin C 50 mg and beta‐carotene placebo;
group 2: vitamin C 500 mg and beta‐carotene placebo;
group 3: vitamin C 50 mg and beta‐carotene 15 mg;
group 4: vitamin C 500 mg and beta‐carotene 15 mg.
217 participants (low‐dose group) were assigned to receive 50 mg of vitamin C and 0/15 mg of beta‐carotene;
222 participants were assigned to receive 500 mg of vitamin C and 0/15 mg of beta‐carotene.
daily for 5 years.
Out of 439 persons initially participating in the study, 134 participants dropped before and on modification of the study protocol based on a National Cancer Institute report that indicated that 2 beta‐carotene trials had shown no benefit or potential harm from the supplement. Of the 305 remaining participants, 244 completed this study.
Participants were supplemented with beta‐carotene from September 1995 to March 1996, (three to six months). After that study was continued with parallel group design.
Participants were randomly assigned to receive:
group 1: vitamin C 50 mg (n = 144);
group 2: vitamin C 500 mg (n = 161);
for five years.
Outcomes The primary outcome measure was: the 10‐year cumulative incidence of gastric cancer.
The secondary outcome measure was: 5‐year change in serum levels of pepsinogens.
After the modification of the protocol the primary outcome measure was 5‐year change in serum levels of pepsinogens and other biomarkers.
Notes Compliance with treatment was constantly encouraged and monitored by nurses, who interviewed the participants and recorded pill counts every 3 months (compliance rate, 80%). Compliance with treatment was checked by serum assessments. Blood samples were drawn and stored three times (at baseline, and after the first, and the fifth year) in order to measure serum level of ascorbic acid. Compliance in taking the vitamin capsules was 92.9% in men and 95.4% in women.
Losses to follow‐up were high due to modification of the protocol. Out of 439 participants randomised, 134 has dropped out before the trial was altered. Of the 397 remaining participants, 305 (77%) consented to take part in a modified trial and 244 completed the trial. Overall, 98 participants, assigned to receive 500 mg vitamin C and 97 participants, assigned to receive 50 mg vitamin C were lost to follow‐up.
Additional information about all‐cause mortality obtained through personal communication with authors. These data, which are extremely positive, are not published.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit.
Blinding (performance bias and detection bias) 
 All outcomes Unclear risk The trial was described as blind, but the method of blinding was not described, so that knowledge of allocation was possible during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes High risk The number or reasons for dropouts and withdrawals were not described.
Selective reporting (reporting bias) High risk One or more clinically relevant and reasonably expected outcomes were not reported on; data on these outcomes were likely to have been recorded.
Other bias High risk There are other factors in the trial that could put it at risk of bias (attrition bias).

SCPS 1990Low.

Methods Skin Cancer Prevention Study (SCPS).
Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 1805, 1251 (70%) men and 554 (30%) women.
Inclusion criteria: < 85 years old, at least one biopsy‐proved basal‐cell or squamous‐cell carcinoma, could not become pregnant, agreement not to take vitamin supplements containing vitamin A or beta‐carotene, not a vegetarian (one who eats no animal products, including milk or eggs).
Exclusion criteria: xeroderma pigmentosum, basal‐cell nevus syndrome, an active non‐skin cancer, known exposure to arsenic, or any other major medical problem that would limit their ability to participate in the planned five years of study.
Interventions Patients were randomly assigned to receive:
group 1: beta‐carotene 50 mg (n = 913);
group 2: placebo (n = 892);
one capsule daily for five years.
Duration of follow‐up was five years.
Outcomes The primary outcome measures were: the first occurrence of basal‐cell or squamous‐cell skin cancer.
Notes Compliance with the study medication was determined by interviewing the patients and by measurement of plasma beta‐carotene levels. At four months interval patients were asked to complete questionnaires concerning compliance in taking the capsules. Reported adherence to treatment did not differ appreciably between the placebo and beta‐carotene groups, and during each of the first four years at least 80% of the patients reported taking half or more of their capsules. Plasma beta‐carotene levels showed more than eight‐fold increase in the group that received beta‐carotene and almost no‐change in the placebo group.
Of the 1805 patients who have been randomised, 89% completed at least three years of observation, 79% four years, and 46% five years. The patterns of follow‐up were similar in the two treatment groups.
The trial agents were provided by BASF, Wyendotte, Michigen.
Though a study with a longer follow‐up on this same trial was published in JAMA (Mortality associated with low plasma concentration of beta carotene and the effect of oral supplementation. JAMA 1996;275(9):699‐703.), we could not use the mortality data from the latter because it does not report on the 85 excluded patients. However, replacing the number of deaths from Greenberg 1990 with the mortality data given in the JAMA publication (146 vs 139 (placebo) does not change the result noticeably.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

SELECT 2009Low.

Methods The Selenium and Vitamin E Cancer Prevention Trial (SELECT).
Multicentre randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America, Canada, and Puerto Rico
Number of participants randomised: 35,533 healthy men, median 62 years of age.
Inclusion criteria: age 50 years or older for African American men and 55 years or older for all other men, no prior prostate cancer diagnosis, 4 ng/mL or less of prostate‐specific antigen (PSA) in serum, a digital rectal examination (DRE) not suspicious for cancer, no current use of anticoagulant therapy other than 175 mg/d or less of acetylsalicylic acid or 81 mg/d or less of acetylsalicylic acid with clopidogrel bisulfate, no history of hemorrhagic stroke, and normal blood pressure.
Exclusion criteria: none stated.
Interventions Participants were randomly assigned to receive:
group 1: oral selenium (200 μg/d from L‐selenomethionine) and matched vitamin E placebo (8910);
group 2: vitamin E (400 IU/d of all rac‐tocopheryl acetate) and matched 
 selenium placebo (n = 8904);
group 3: oral selenium (200 μg/d from L‐selenomethionine) plus vitamin E (400 IU/d of all rac‐tocopheryl acetate) (n = 8703);
group 4: selenium vitamin E, or placebo placebo (n = 8856)
orally, daily, for a period of 7 to 12 years (median 5.46 years (4.17 to 7.33 years).
Outcomes The primary outcome measure was prostate cancer 
 incidence. Secondary outcome measures were incidence of other cancers and overall mortality.
Notes "The trial was activated in July 2001 and follow‐up blinded to the trial results ended on October 23, 2008. Adherence and adverse events were monitored every 6 months and a limited physical examination including assessments of blood pressure, weight, and smoking status was conducted annually."
Adherence and adverse events were monitored every 6 months and a limited physical examination including assessments of blood pressure, weight, and smoking status was conducted annually. Adherence to both study agents as determined by pill count was similar across all study groups, and averaged 83% at year 1 and 65% at year 5. Adherence to at least 1 of the 2 agents was 87% at year 1 and 72% at year 5 (the design estimated adherence rates were 90% at year 1 and 68% at year 5).
Almost equal number of participants lost to follow‐up in each intervention group.
Study agents and packaging were provided by Perrigo Company (Allegan, Michigan), Sabinsa Corporation (Piscataway, NewJersey), Tishcon Corporation (Westbury, New York), and DSM Nutritional Products Inc (Parsipanny, New Jersey).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

SIT 2006.

Methods Shandong Intervention Trial (SIT)
Randomised, double‐blind, placebo controlled, primary prevention trial with stratified, factorial design 2x2x2 versus 2x2.
Participants Country: China (Linqu County, Shandong Province).
Number of participants randomised: 3411, 1753 men and 1658 women aged 35 to 64 years.
Inclusion criteria: participants aged 35 to 64 years willing to participate in 42‐month study, baseline gastroscopy with biopsies, known Helicobacter pylori status.
Exclusion criteria: illness, bleeding disorders, cancers (except nonmelanoma skin cancer), heart failure, emphysema, renal or liver diseases, other life‐threatening illnesses, allergy to penicillin or related antibiotics.
Interventions Participants were first divided on the basis of whether they showed serologic evidence of Helicobacter pylori infection at baseline (2285) or not (1126). Participants with serologic evidence of Helicobacter Pylori at baseline were eligible to receive amoxicillin (1 g twice a day) and omeprazole (20 mg twice a day) in three capsules (two 500 mg amoxicillin and one 20 mg omeprazole) to be taken twice daily (before breakfast and dinner) for 2 weeks. Look‐alike placebo capsules containing lactose and starch for amoxicillin and sucrose and starch for omeprazole were given to serologically positive controls and to all seronegative participants. Approximately 3 months after initial treatment for Helicobacter Pylori, supplementation with 100 IU alpha‐tocopherol, 250 mg vitamin C, and 37.5 µg selenium twice a day began its 39‐month course. Participants receive this mixture in one capsule, to be taken twice daily before or after breakfast and dinner. From December 1995 to May 1996, this mixture also contained beta‐carotene (7.5 mg twice a day). Look‐alike placebo capsules contained cellulose, lactose, and magnesium stearate.
In the garlic group, participants take two capsules twice a day before or after breakfast and dinner. Each capsule contains 200 mg Kyolic aged garlic extract and 1 mg steam‐distilled garlic oil. To prepare the extract, the manufacturer slices garlic cloves and soaks them in aqueous ethanol (about 20%) for over 18 months at room temperature. The extract is then filtered, concentrated, and dried. The look‐alike placebo capsules contain cellulose, granulated sugar, caramel, and magnesium stearate. Bottles holding placebo capsules contained minute quantities of garlic oil so they would smell like garlic.
HP‐seropositive at baseline (2258) entered 2x2x2 factorial of antibiotics, vitamins, and garlic. HP‐seronegative at baseline (1126) entered 2x2 factorial trial of vitamins, and garlic.
Participants were randomised in 12 groups:
group 1: amoxicillin and omeprazole, garlic, vitamin and selenium (n=286);
group 2: amoxicillin and omeprazole, garlic, vitamin and selenium placebo (n=285);
group 3: amoxicillin and omeprazole, garlic placebo, vitamin and selenium (n=286);
group 4: amoxicillin and omeprazole, garlic placebo, vitamin and selenium placebo (n=285);
group 5: amoxicillin and omeprazole placebo, garlic, vitamin and selenium (n=285);
group 6: amoxicillin and omeprazole placebo, garlic, vitamin and selenium placebo (n=286);
group 7: amoxicillin and omeprazole placebo, garlic placebo, vitamin and selenium (n=286);
group 8: amoxicillin and omeprazole placebo, garlic placebo, vitamin and selenium placebo (n=286);
group 9: amoxicillin and omeprazole placebo, garlic; vitamin and selenium (n=282);
group 10: amoxicillin and omeprazole placebo, garlic, vitamin and selenium placebo (n=281);
group 11: amoxicillin and omeprazole placebo, garlic placebo, vitamin and selenium (n=281);
group 12: amoxicillin and omeprazole placebo, garlic placebo, vitamin and selenium placebo (n=282);
Outcomes The primary outcome measures were: prevalence of dysplasia or gastric cancer, prevalence of severe chronic atrophic gastritis, intestinal metaplasia, dysplasia or gastric cancer, and average severity score.
Secondary outcome measures were: rates of transition from baseline to final histopathologic states and the effects of treatments on these rates of transition; evidence of the effectiveness of amoxicillin and omeprazole in eradicating Helicobacter pylori, based on 13C‐urea breath tests 3 months following treatment, on annual serology, and on a final pathologic examination of biopsies to look for Helicobacter pylori; and blood pressure at the time of the final examination.
Notes Compliance with treatment was checked by measuring the plasma vitamin levels in randomly selected participants every 3 months and counting of the pills. Compliance with treatment was good. The average monthly proportion of participants taking all pills was 92.3%. Serum samples obtained from randomly selected participants demonstrate higher levels of vitamins C and E in participants assigned to vitamins 2 and higher levels of S‐allylcysteine in those assigned to garlic preparation.
Overall 15 participants from placebo and 19 participants from active intervention group were lost to follow‐up.
(Wakunaga of America, Co., Ltd, Mission Viejo, CA) provided the garlic preparation, Astra (East Asia Region) provided amoxicillin and omeprazole; and Sino‐American Shanghai‐Squibb Pharmaceuticals, Ltd. provided vitamin and mineral supplement.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Unclear risk The reasons for dropouts and withdrawals in all intervention groups were not described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

SKICAP AK 1997Low.

Methods Skin Cancer Prevention Study ‐ actinic keratoses (SKICAP‐AK).
Randomised, double‐blind placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States.
Number of participants randomised: 2297, 679 (30%) women and 1618 (70%) men, aged 21 to 84 years, median age 63 years, with a history of more than 10 actinic keratoses and at most 2 squamous cell carcinoma (SCC) or basal cell carcinoma (BCC) skin cancers.
Inclusion criteria: free living participants aged 21 to 84 years, ambulatory and capable of self care, with no diagnosis of life threatening diseases, an intended continual resident of Arizona for at least five years, willing to return during the five years for semi‐annual follow‐up clinic visits, and willing to limit non‐study vitamin A supplementation to no more than 10,000 IU per day, clinical laboratory values within the 95% normal range for total cholesterol, liver function (AST and ALT), WBC count, haemoglobin and platelet count, and history of more than 10 actinic keratoses and at most 2 squamous cell carcinoma (SCC) or basal cell carcinoma (BCC) skin cancers.
Exclusion criteria: cancer diagnosis or treatment within the year preceding the trial other than BCC or SCC, history of xeroderma pigmentosum or basal‐cell nevus syndrome.
Interventions Patients were randomly assigned to receive:
group 1: vitamin A (retinol) 25,000 IU (n = 1157);
group 2: placebo (n = 1140);
one capsule daily for a period of 5 years (median follow‐up time of 3.8 years.
Outcomes The primary outcome measures were: the time to first new occurrence of SCC and time to first new occurrence of BCC pathologically confirmed by the study pathologist.
Notes Compliance with the study medication was determined by counting capsules in the returned medication bottles and by measurement of plasma vitamin A levels. Participants were scheduled for a return clinic visit one month after randomisation and then every six months. They were interviewed to evaluate adherence, motivated to adhere and provided with a six‐month supply of capsules. Participants were telephoned and mailed postcards between clinic visits for symptom assessment and adherence monitoring and motivation. Vitamin intake was reported by 73% of the participants, and 30% of participants reported dietary intake near or below the recommended. Calculated adherence to the intervention was almost identical between the placebo and retinol groups. During the five‐year intervention period, at least 85% of participants reported taking at least three‐quarters of their capsules, and at least 95% reported taking at least half of their capsules. The results show very similar baseline retinyl palmitate levels and approximately an eight‐fold increase in the median serum retynil palmitate level in the group assigned to receive retinol.
Overall, 99 participants from the intervention group and 88 from the placebo group were lost to follow‐up.
Hoffmann‐LaRoche provided intervention capsules.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

SPACE 2000Low.

Methods Secondary prevention with antioxidants of cardiovascular disease in end stage renal disease (SPACE).
Randomised, double blind, placebo‐controlled, secondary prevention intervention trial with parallel group design (two intervention groups).
Participants Country: Israel.
Number of patients randomised: 196; 135 males and 61 females, aged 40 to 75 years, mean age 64.6 years.
Inclusion criteria: stable haemodialysis patients between the ages of 40 and 75 years inclusive at baseline with a documented medical history of cardiovascular disease (including hospital records, appropriate electrocardiographic and biochemical supporting indices).
Exclusion criteria: anticoagulant therapy with warfarin sodium; known history of malignant disease (except non‐melanoma skin cancer); active liver disease; treatment with hypolipaemic agents for less than eight weeks before the study started; pregnant or planning to become pregnant during duration of the study; any condition the treating physician deemed to preclude the patient on grounds of safety or study evaluation.
Interventions Patients were randomly assigned to receive:
group 1: vitamin E 800 IU/day (n = 97);
group 2: matching placebo (n = 99);
Vitamin E was provided as two capsules of 400 IU each. Patients were instructed to take two capsules nightly.
Median follow‐up time was 519 (range 10 to 763) days.
Outcomes The primary outcome measure was: a composite variable consisting of: acute myocardial infarction (fatal and nonfatal); ischaemic stroke; peripheral vascular disease (excluding the arterio‐venous fistula) in a limb not previously affected; and unstable angina.
Secondary outcome measures were: fatal and non‐fatal myocardial infarction, cardiovascular disease mortality (fatal myocardial infarction, ischaemic stroke or sudden death), total mortality, ischaemic stroke, peripheral vascular disease, and unstable angina.
Notes Compliance with treatment was evaluated by measuring serum vitamin E concentrations. Throughout the study, patients continued to receive regular monthly follow‐up by their unit dieticians, who instructed them to comply with dietary recommendations for maintenance haemodialysis patients. Additional vitamin supplementation was similar in the two treatment conditions. Folate (5 to 10 mg/day), vitamin B6 (10 to 250 mg/day), and vitamin B12 (250 µg/day) were prescribed to 57 (57.5%) patients in the placebo group and 55 (56.7%) patients in the vitamin E group. Only one patient (in the vitamin E group) received vitamin B12 as a monthly intramuscular injection. Vitamin C (100 to 500 mg/day) was prescribed to 42 (42.5%) of the placebo group and 42 (43.3%) of the vitamin E group.
There were no losses to follow‐up.
Vitamin E and placebos were provided by Solgar, Inc, New York, USA, during the first year and Henkel Corp, La Grange, IL, USA, during the second year.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Stevic 2001.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Yugoslavia.
Number of participants randomised: 28, 75% men and 25% women, aged 20 to 70 years, mean age 57 years.
Inclusion criteria: probable or definite ALS by El Escorial criteria, age 20 to 70 years, disease duration < 3 years, ambulatory.
Exclusion criteria: significant compromise of bulbar or respiratory function, conduction block, M protein, significant imaging abnormality, dementia, and concurrent systemic disease.
Interventions Participants were randomly assigned to receive:
group 1: alsemet‐L‐methionine (2 g), vitamin E (400 IU), selenium (3 x 10‐5g) three times daily (n = 16);
group 2: placebo (n = 12);
for a period of one year.
Outcomes The primary outcome measures were: survival and rate of disease progression as expressed by decline in limb‐function, bulbar‐function and muscle‐testing scores.
Secondary outcome measures were: activity of antioxidative components, and level of vitamin E in blood.
Notes Compliance with treatment is not reported.
There were no losses to follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes High risk The number or reasons for dropouts and withdrawals were not described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

SUVIMAX 2010Low.

Methods The SUpplementation en VItamines et Mine´raux AntioXydants (SU.VI.MAX) Study (SU.VI.MAX).
Randomised, double‐blind, placebo‐controlled, primary‐prevention trial with parallel group design (two intervention groups).
Participants Country: France.
Number of participants randomised: 13017 French adults, 5141 men and 7876 women, aged from 35 to 60 years, mean age 48.95 years.
Inclusion criteria: lack of disease likely to hinder active participation or threatened 5‐year survival; acceptance of possibility to be given placebo and acceptance of the constraints of participation; lack of previous regular supplementation with any of the vitamins and minerals in the supplement provided and absence of extreme beliefs or behaviour regarding diet.
Exclusion criteria: none stated.
Interventions Participants were randomly assigned to receive:
group 1: beta carotene 6 mg; vitamin C 120 mg; vitamin E 30 mg; selenium 100 µg; zinc 20 mg (n = 6481);
group 2: placebo (n = 6536).
All participants took a single daily capsule. Median follow‐up time was 7.5 years.
Postintervention follow‐up assessment of total cancer incidence, Ischaemic cardiovascular disease incidence and total mortality was carried out for five years (September 1, 2002, to September 1, 2007).
Outcomes The primary outcome measures were: major fatal and nonfatal Ischaemic cardiovascular events and cancer of any kind, except for the basal cell carcinoma of the skin.
The secondary outcome measure was: all cause mortality.
Notes Primary analyses were performed on events validated on September 1, 2002, and published in 2004. The median follow‐up time was 7.54 years, ranging from two days to 7.89 years. Participants still alive at the end of the supplementation period (September 1, 2002), excepted subjects who dropped out or were lost to follow‐up during the supplementation period (n = 11,054), were asked to participate to the post supplementation follow‐up and to fill every 6 months a postal questionnaire thereafter to collect self‐reported health information.
Compliance for the intervention group was confirmed by measuring the biochemical markers of supplementation after 2 years and after 7 years for beta‐carotene, vitamin C and selenium. At the end of follow‐up, 74% of participants reported having taken at least two thirds of the capsules. There were no differences between the groups mean percentage of capsules taken, ie, 79% in each group).
Losses to follow‐up were 5.4 % in the intervention group and 6.2% in the placebo group. Overall, 739 participants in the active and 828 participants in the placebo group were lost to follow‐up.
Sponsors of the trial: Fruit d'Or Recherche, Candia, Lipton, Kellogg's, Centre d'Information sur Canderel, Orangina, Este e Lauder, Cereal, Grands Moulins de Paris, CERIN, L'Ore al, Peugeot, Jet Service, RP Scherer, Sodexho, France Telecom, Santogen, Becton Dickinson, Fould Springer, Boehringer Diagnostic, Seppic Givaudan Lavirotte, Le Grand Canal.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Capsule boxes were labelled with the participant's number, using partitioned organisation to ensure total security of the blind study.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Takagi 2003.

Methods Randomised, clinical trial with parallel group design (two intervention groups).
Participants Country: Japan.
Number of patients randomised: 93, 45% males and 55% females, mean age 62.5 years.
Inclusion criteria: liver cirrhosis caused by hepatitis C infection.
Exclusion criteria: none stated.
Interventions Patients were randomly assigned to receive:
group 1: vitamin E (600 mg) (n = 51);
group 2: no treatment (n = 42);
for a period of 5 years.
Outcomes The primary outcome measures were: tumour‐free survival and cumulative survival rate.
Notes Compliance was not reported.
Seven patients in the vitamin E group and 3 patients from the control group dropped out from the trial.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes High risk The trial was not blinded, so that the allocation was known during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

Takamatsu 1995.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Japan.
Number of participants randomised: 161, 64 men and 97 women, aged 39 to 56 years.
Inclusion criteria: healthy Japanese adults free of acute and chronic illness, including hypertension.
Exclusion criteria: taking oral contraceptives, vitamins or mineral supplements, pregnant or lactating women.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (d‐alpha‐tocopheryl acetate) 100 mg (n = 82);
group 2: placebo (vitamin E 3 mg), (n = 79);
for a period of six years.
Outcomes The primary outcome measure was any illness.
Notes Medication compliance during the trial period was 89.6% in vitamin E group and 91.3% in the placebo group.
Losses to follow‐up were four participants (6.75%) in the active treatment group and 10 participants (13.69%) in the placebo group during the trial.
Vitamin E (d‐alpha tocopheryl acetate capsules) were provided by Eisai Co. Ltd (Tokyo, Japan).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Unclear risk The trial is described as randomised, but the method of sequence generation was not specified.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Unclear risk The trial may or may not be free of other components that could put it at risk of bias.

Tam 2005Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Hong Kong.
Number of patient's randomised: 39 females, mean age 46.
Inclusion criteria: female patients with systemic lupus erythematosus.
Exclusion criteria: flare of systemic lupus erythematosus requiring increase in immunosuppressive agents.
Interventions Patients were randomly assigned to receive:
group 1: vitamin C 500 mg, vitamin E (D‐alpha tocopheryl succinate) 800 IU n=20;
group 2: placebo, n=19; daily, 12 weeks.
Patients were followed 2.67 years.
Outcomes The primary outcome measures were: effects on markers of oxidative stress, antioxidant defence, and endothelial function.
Notes Compliance was assessed by tablet counting, and patients with less than 70% compliance were excluded from the analyses. Overall compliance by pill count was 95%.
There were no losses to follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ter Riet 1995.

Methods Randomised, double‐blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: The Netherlands.
Number of participants randomised: 88.
Inclusion criteria: patients with pressure ulcers (partial thickness skin loss or worse).
Exclusion criteria: difficulties with swallowing or frequent vomiting, osteomyelitis in the ulcer area, idiopathic haemochromatosis, thalassaemia major, sideroblastic anaemia, Cushing's syndrome or disease, pregnancy, radiotherapy in the ulcer area, and the use of antineoplastic agents or systemic glucocorticosteroids, high probability to drop out within the 12 week follow‐up period (terminally ill patients, patients for whom surgical treatment of the ulcer‐other than debridement), taking vitamin C supplements in excess of 50 mg/day.
Interventions Participants were randomly assigned in four groups to receive:
group 1: vitamin C 1000 mg and ultrasound;
group 2: vitamin C 1000 mg and sham ultrasound;
group 3: vitamin C 20 mg and ultrasound;
group 4: vitamin C 20 mg and sham ultrasound;
daily for 12 weeks. Overall 43 participants were supplemented with 1000 mg of vitamin C, while 45 participants were in 'placebo' group supplemented with 20 mg of vitamin C.
Outcomes The primary outcome measures were: wound survival, healing rates of wound surfaces, and clinimetric changes.
Notes Compliance with prevention is not reported. The trial used 20 mg vitamin C in the placebo pills. Removing this trial from our analyses does not noticeably change our results.
During the course of the study three participants withdrew. Overall one participant withdrew from the intervention group and two withdrew from the control group.
Hoffmann‐La Roche & Co., Ltd., Basel supplied vitamin C tablets.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Unclear risk The trial was described as randomised but the method used to conceal the allocation was not described, so that intervention allocations may have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias.

UK PRECISE 2006Low.

Methods Prevention of Cancer by Intervention with Selenium Pilot Study (PRECISEp).
Randomised, double‐blind, placebo‐controlled trial with parallel group design (four intervention groups).
Participants Country: United Kingdom.
Number of participants randomised: 501, 53% men at age 60 to 74 years, mean age 67 years.
Inclusion criteria: volunteers from four general practices.
Exclusion criteria: incapable of carrying out light housework or office work, active liver or kidney disease, prior diagnosis of cancer (excluding nonmelanoma skin cancer), diagnosed HIV infection, immunosuppressive therapy, diminished mental capacity, taking > 50 µg/day of selenium supplements in the previous six months (by patient report).
Interventions Participants were randomly assigned to receive:
group 1: placebo (n = 121);
group 2: selenium 100 µg (n = 127);
group 3: selenium 200 (n = 127);
group 4: selenium 300 µg (n = 126);
in the form of high‐selenium yeast, Seleno PreciseTM per day for two years.
Outcomes The primary outcome measures were: mood, quality of life, and plasma selenium level.
Notes Compliance with randomised treatment was determined by pill count, with participants considered compliant if they took at least 80% of their allocated tablets. Reasons for participant withdrawal were noted. Four hundred fifty three of the 467 participants (97%) who completed six months were compliant according to pill count.
Thirty‐four participants (7%) withdrew from treatment within the first six months. There was no significant difference in treatment withdrawals between groups (7, 10, 5, and 12 in the placebo and 100, 200, and 300 µg groups respectively.
Trial agents were provided by Pharma Nord, Vejle, Denmark.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

VEAPS 2002Low.

Methods The Vitamin E Atherosclerosis Prevention Study (VEAPS).
Randomised, double‐blind, placebo‐controlled, primary‐prevention trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 353 men and women, aged from 40 to 82 years, mean age 56 years.
Inclusion criteria: age > 40 years, with LDL cholesterol (LDL‐C) > 3.37 mmol/L (130 mg/dL) and no clinical signs or symptoms of cardiovascular disease (CVD).
Exclusion criteria: fasting triglycerides > 5.64 mmol/L, diabetes mellitus or fasting serum glucose > 3.62 mmol/L, regular vitamin E supplement intake > 1 year, lipid standardised plasma vitamin E > 35 µmol/L, diastolic blood pressure > 100 mm Hg, untreated thyroid disease, serum creatinine > 0.065 mmol/L, life‐threatening disease with prognosis < 5 years, or alcohol intake > 5 drinks daily.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (DL‐alpha‐tocopherol) 400 IU (n = 177);
group 2: placebo (n = 176);
daily for a period of three years.
Participants were instructed to take trial pills with their greatest fat‐containing meal of the day.
The initial trial design called for a 2‐year treatment period. Based on evolving null results from other antioxidant clinical trials, the External Data and Safety Monitoring Board recommended after 2 years of initiation of the study that the treatment period be extended to 3 years. Participants were offered the opportunity to continue another year of their randomised and blinded treatment assignment. The 81% of 2‐year completers, 73% of randomised elected to continue for a 3‐year treatment period.
Outcomes The primary trial outcome was: rate of change in the right distal common carotid artery intima‐media thickness in computer image‐processed B‐mode ultrasonograms.
Notes Compliance with treatment was assessed by counting unused pills and measuring plasma vitamin levels. Mean pill compliance was 92% in the placebo‐treated group and 91% in the vitamin E group. Pill compliance for the placebo‐treated versus the active vitamin E participants was maintained throughout the trial, as follows: 89% versus 87%, 90% versus 89%, 92% versus 92%, 91% versus 93%, 94% versus 91%, and 93% versus 93% at 6, 12, 24, 30, and 36 months, respectively. There was an appropriate rise in the mean plasma vitamin E level in the active vitamin E group from a baseline level.
Ninety percent of the randomised participants completed the two‐year treatment. Overall, 15 participants in the active group and six participants in the placebo group were lost to follow‐up.
The trial was supported by Hoffmann‐La Roche, Inc.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

VECAT 2004Low.

Methods Vitamin E, Cataract and Age‐Related Maculopathy Trial (VECAT).
Randomised, double blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Australia.
Number of participants randomised: 1193, 44% men and 56% women, aged 55 to 80, mean age 65.7 years.
Inclusion criteria: good general health, early or no cataract.
Exclusion criteria: prior cataract surgery, advanced cataract in both eyes, glaucoma, known sensitivity to vitamin E, and long‐term treatment with steroids or anticoagulants.
Interventions Participants were randomly assigned to receive:
group 1: vitamin E (natural vitamin E in soybean oil) 500 IU (n = 595).
group 2: placebo (n = 598);
for four years.
Outcomes The primary outcome measures were: major age‐related types of cataract: nuclear, cortical cuneiform, and posterior subcapsular.
Notes Compliance with the trial medication was determined by counting capsules in the returned medication bottles and by measurement of plasma vitamin E levels in a random sample of participants. Overall, 77% of the actively treated group and 79% of those participants randomised to placebo were estimated to have consumed 80% or more of their capsules. After 4 years of follow‐up, 74% of the vitamin E group and 76% of the placebo group remained on their assigned medication and participated in the annual reviews. Among the remaining 25% of the participants, 12% in each group ceased taking the assigned medication but continued participating to have their eye examined.
Overall, 60 participants from the placebo group and 58 participants from the vitamin group withdrew from the trial.
The trial was funded by Smith and Nphew, Australia, Henkel, Australia.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

WACS 2007Low.

Methods Women's Antioxidant Cardiovascular Study (WACS).
Randomised, double‐blind, placebo‐controlled trial using two‐by‐two‐by‐two factorial design and than two‐by‐two‐by‐two‐by‐two design.
Participants Country: United States of America.
Number of participants randomised: 8171 female health professionals aged 40 years or older, mean age 60.6 years, with a history of cardiovascular disease or three or more cardiovascular risk factors.
Inclusion criteria: 40 years or older, postmenopausal, or had no intention of becoming pregnant, had a self reported history of cardiovascular disease, or had at least three cardiac risk factors (self reported diagnosis of hypertension, high cholesterol level, or diabetes mellitus); parental history of premature myocardial infarction (MI) (before age 60 years); obesity (body mass index (BMI) ≥ 30), current cigarette smoking; and inconsistent report of prior cardiovascular disease.
Exclusion criteria: self‐reported history of cancer (excluding nonmelanoma skin cancer) within the past 10 years, any serious non‐cardiovascular illness, or were currently using warfarin sodium or other anticoagulants.
Interventions Participants were randomly assigned to receive:
group 1: active vitamin C (synthetic vitamin C) 500 mg daily, active vitamin E (d‐alpha tocopherol acetate) 600 IU every other day, and beta‐carotene (Lurotin) 50 mg, every other day (n = 1020);
group 2: active vitamin C (synthetic vitamin C) 500 mg daily, active vitamin E (d‐alpha tocopherol acetate) 600 IU every other day, and placebo beta‐carotene (Lurotin) 50 mg, every other day (n = 1021);
group 3: active vitamin C (synthetic vitamin C) 500 mg daily, placebo vitamin E every other day, and beta‐carotene (Lurotin) 50 mg, every other day (n = 1023);
group 4: active vitamin C (synthetic vitamin C) 500 mg daily, placebo vitamin E every other day, and placebo beta‐carotene, every other day (n = 1023);
group 5: placebo vitamin C daily, active vitamin E (d‐alpha tocopherol acetate) 600 IU every other day, and beta‐carotene (Lurotin) 50 mg, every other day (n = 1021);
group 6: placebo vitamin C daily, active vitamin E (d‐alpha tocopherol acetate) 600 IU every other day, and placebo beta‐carotene every other day (n = 1021);
group 7: placebo vitamin C daily, placebo vitamin E every other day, and beta‐carotene (Lurotin) 50 mg, every other day (n = 1020);
group 8: placebo vitamin C daily, placebo vitamin E every other day, and placebo beta‐carotene every other day (n = 1022);
for a mean period of 9.4 years (range, 8.3 to 10.1 years).
In 1998, approximately 2 to 3 years following randomisation to the antioxidant arms, a folic acid ‐ vitamin B6/B12 component was added to the trial, expanding it to a two‐by‐two‐by‐two‐by‐two factorial trial.
Outcomes The primary outcome was a combined end point of CVD morbidity and mortality, including incident myocardial infarction, stroke, coronary revascularization procedures (coronary artery bypass grafting or percutaneous transluminal coronary angioplasty), and cardiovascular mortality. Secondary outcome measures were myocardial infarction, stroke, coronary revascularisation, and cardiovascular death. Information on transient ischaemic attack and total mortality was also collected.
Notes Between June 1995 and October 1996, a total of 8171 women were randomly assigned according to a two‐by‐two‐by‐two factorial design. Study treatment and endpoint ascertainment were continued in a blinded fashion through January 31, 2005, the scheduled end of the trial.
Compliance was assessed through self‐report and defined as taking at least two‐thirds of study pills. Reported compliance was, on average, 76% at four years and 68% at eight years of follow‐up for each antioxidant, with no significant difference between active and placebo groups at these times except for ascorbic acid at eight years (70% versus 67% in the active versus placebo group). Mean compliance over follow‐up was approximately 73% for all active and placebo agents.
Overall, vital status was known for 93.3% of randomised participants.
Vitamin C and beta carotene were supplied by BASF Corp (Wyandotte, MI) and vitamin E was supplied by (Cognis Corp, LaGrange, Il).
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

WAVE 2002Low.

Methods Women's Angiographic Vitamin and Estrogen Trial (WAVE).
Randomised, double blind, placebo‐controlled trial with two‐by‐two factorial design.
Participants Country: United States of America and Canada.
Number of participants randomised: 423 women mean age 65 years.
Inclusion criteria: postmenopausal women as defined by any one of the following criteria: (bilateral oophorectomy at any age or age 45 to 55 with FSH 40 mIU/ml or older than 55 years. Protocol angiogram within four months performed while haemodynamically stable demonstrating at least one vessel segment free of intervention, with 15 to 75% stenosis. If the angiogram was performed within two weeks of a myocardial infarction, the qualifying segment may not be the infarct segment.
Exclusion criteria: oestrogen replacement therapy within the past three months. Estrogen vaginal cream permitted if used no more than 25% of the time. Concurrent use of vitamins C and E exceeding the recommended dietary allowance, history of breast cancer or mammogram suggestive of cancer without subsequent negative biopsy, history of endometrial carcinoma without subsequent hysterectomy, any abnormal uterine bleeding or endometrial hyperplasia at baseline, pap smear with dysplasia of cervical intraepithelial neoplasia grade I or greater, uncontrolled diabetes or hypertension, myocardial infarction less than four weeks prior to randomisation, planned or prior coronary artery bypass grafting, fasting triglycerides 500 mg/dl within four months of randomisation, creatinine 2.0 mg/dl, symptomatic gallstones, New York Heart association class IV congestive heart failure or known ejection fraction 25%, history of haemorrhagic stroke or bleeding diathesis, history of pulmonary embolism or idiopathic deep venous thrombosis, history of osteoporosis unless treated with nonhormonal therapy, anticipated survival three years, concurrent participation in other masked clinical trial, participation in an interventional device trial or short‐term postangioplasty antithrombotic trial was permitted so long as follow‐up angiography was not a requirement of that trial.
Interventions The participants were randomly assigned to receive:
group 1: vitamins (vitamin E 400 IU and vitamin C 500 mg) and hormone replacement therapy (HRT) placebo (n = 105);
group 2: HRT (women with a prior hysterectomy took one tablet containing conjugated equine estrogens (0.625 mg of Premarin, while the women who had not had a hysterectomy took one tablet containing conjugated equine estrogens and medroxyprogesterone acetate (0.625 mg/2.5 mg of Prempro) and vitamins placebo daily (n = 103);
group 3: vitamins C and E and HRT (n = 107);
group 4: vitamin placebo and HRT placebo (n = 108);
twice daily for a median of three years.
Outcomes The primary outcome measure was: annualised mean change in minimum lumen diameter from baseline to concluding angiogram of all qualifying coronary lesions averaged for each patient. Patients with intercurrent death or myocardial infarction were imputed the worst rank of angiographic outcome.
Notes Compliance with treatment was checked by serum assessments. Among the women with angiographic follow‐up, those assigned to HRT took 67% of their prescribed medication according to pill counts, and those assigned to HRT placebo took 70%. The corresponding figures were both 84% for antioxidant vitamins and vitamin placebo. Nine women assigned to placebo oestrogen crossed over to open‐label oestrogen, and one woman assigned to placebo vitamin supplements crossed over to open‐label vitamins.
Twenty‐three participants from the placebo group and 29 from the HRT group, 38 from the vitamins, and 27 from the HRT and vitamin group withdrew from the trial.
Hormone replacement drugs supplied by Wyeth Pharmaceuticals, Collegeville, Pa.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

White 2002Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United Kingdom. Number of participants randomised: 100, mean age 63, 58% males.
Inclusion criteria: patients with Barrett's oesophagus on long‐term (> 12 months) proton pump inhibitors (PPI) treatment attending for surveillance endoscopy.
Exclusion criteria: pregnancy or lactation, previous gastric surgery, serious cardiovascular, respiratory, renal or neurological diseases, history of alcohol or drug abuse or use of non‐steroidal antiinflammatory drugs.
Interventions The participants were randomly assigned to receive:
group 1: vitamin C 100 mg, vitamin E 200 mg, n = 50;
group 2: placebo, n = 50.
Participants were supplemented and followed 12 weeks.
Outcomes The primary outcome measure was: changes in putative markers of DNA damage in gastric tissue following supplementation with vitamins C and E.
Notes Plasma vitamin C and E were measured to assess patient compliance.
Seventeen participants failed to attend for endoscopy and 11 participants were not compliant with the trial medication, leaving 72 participants for final analyses. Overall, 14 participants in each group were lost to follow‐up.
Additional information about all‐cause mortality obtained from the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

WHS 2005Low.

Methods Women's Health Study (WHS).
Randomised, double‐blind, placebo‐controlled trial with two‐by‐two‐by‐two factorial design in the beginning and than two‐by‐two.
Participants Country: United States of America. Number of participants randomised: 39876 females aged 45 years or older, mean age 54.6 years. Inclusion criteria: female health professionals willing to take part in the trial. Age 45 years or older; no previous history of coronary heart disease, cerebrovascular disease, cancer (except nonmelanoma skin cancer), or other major chronic illnesses; no history of adverse effects from aspirin; no use of aspirin or nonsteroidal anti‐inflammatory drugs (NSAIDs) more than once a week, or willingness to forgo their use; no use of anticoagulants or corticosteroids; and no use of individual supplements of vitamin A, E, or beta carotene for more than once a week. Exclusion criteria: history of cancer (except non‐melanoma skin cancer), coronary heart disease, or cerebrovascular disease.
Interventions Participants were randomly assigned to one of the eight treatment groups. The active agents were 100 mg of aspirin, given on alternate days; 600 IU of vitamin E, given on alternate days; and 50 mg of beta‐carotene, given on alternate days. group 1: aspirin 100 mg, beta carotene 50 mg, vitamin E 600 IU; group 2: aspirin 100 mg, beta carotene 50 mg, vitamin E placebo; group 3: aspirin 100 mg, beta carotene 50 mg placebo, vitamin E 600 IU; group 4: aspirin 100 mg, beta carotene placebo, vitamin E placebo; group 5: aspirin placebo, beta carotene 50 mg, vitamin E 600 IU; group 6: aspirin placebo, beta carotene 50 mg, vitamin E placebo; group 7: aspirin placebo, beta carotene placebo, vitamin E 600 IU; group 8: aspirin placebo, beta carotene placebo, vitamin E placebo; A total of 19939 women were assigned at random to receive beta‐carotene and 19937 to receive placebo in the beginning of April 1993. A total of 19937 women were assigned at random to receive vitamin E and 19939 to receive placebo. The beta‐carotene component of the trial was terminated early, on January 18, 1996. The aspirin and vitamin E components of the trial continued uninterrupted. The time from randomisation to the end of beta‐carotene component of the study averaged 2.1 years. Authors published results of the beta‐carotene component of the trial on February 6, 1998, after a median total follow‐up of 4.1 years (2.1 years treatment plus 2.0 years follow‐up). From that time trials proceeded as two‐arm (vitamin E and placebo). Follow‐up and validation of reported end points were completed in February 2005. The average duration of follow‐up from randomization to the end of the trial was 10.1 years (range, 8.2‐10.9 years).
Outcomes The primary outcome measures were: incidence of invasive cancer (except non‐melanoma skin cancer), myocardial infarction, and stroke. The secondary outcome measures were: non‐fatal myocardial infarction, non‐fatal stroke, death from cardiovascular causes, and death from any cause.
Notes Compliance with treatment was checked by random serum assessments. Compliance with treatment was excellent. At the time of termination of the beta‐carotene component, 87% of the active group have taken at least two thirds of the study capsules, while 9.9% of the women in the placebo group have taken beta‐carotene or vitamin A supplements outside the trial.
The active agents were provided as follows: aspirin by Bayer AG, Leverkusen, Germany; vitamin E by Natural Source Vitamin E Association, Washington DC; and beta‐carotene by Lurotin, BASF Corporation, Wiandotte, MI. Data were extracted from the primary publication, but additional information was received through personal communication with the authors.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Centrally by computer in batches of blocks of size 16.
Allocation concealment (selection bias) Low risk The randomisation allocation is coded. Shipping department sends out calendar packs (which are identical whether active or placebo) to individual participants depending on this code. All of the calendar packs are in coded boxes, supplied by the drug manufacturer, so that the shippers do not know which drug they are shipping.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Witte 2005Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: United States of America.
Number of participants randomised: 32, aged > 70 years.
Inclusion criteria: stable chronic heart failure due to ischaemic heart disease.
Exclusion criteria: neurological or inflammatory conditions or other significant chronic morbidity affecting quality of life (eg, severe rheumatoid arthritis) or requiring long‐term systemic steroid or non‐steroidal anti‐inflammatory drugs therapy (except low‐dose aspirin). Patients in persistent atrial fibrillation were also excluded in order to optimise the reproducibility of the estimation of left ventricular function.
Interventions Participants were randomly assigned to receive:
group 1: calcium 250 mg; magnesium 150 mg; zinc 15 mg; copper 1.2 mg; selenium 50 µg; vitamin A 800 mg; thiamine 200 mg; riboflavin 2 mg; vitamin B6; 200 mg; folate 5 mg; vitamin B12 200 µg; vitamin C 500 mg; vitamin E 400 mg; vitamin D 10 µg; Co‐enzyme Q10 150 mg, (n = 16);
group 2: placebo (n = 16) (cellulose);
four capsules per day for a period of nine months. Patients were followed for an average of 295 days.
Patients were on otherwise optimal therapy including diuretics, angiotensin‐converting enzyme inhibitors, and beta‐blockers if tolerated.
Outcomes The primary outcome measures were: left ventricular function, levels of pro‐inflammatory cytokines, and quality‐of‐life in elderly patients with chronic heart failure.
Notes Authors did not measure blood levels of all the micronutrients to assess compliance, although the changes in the ferritin, vitamin B12, and folate levels in the patients, randomised to the micronutrient combination suggest that they took them.
One patient in each arm was lost to follow‐up.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

Wluka 2002Low.

Methods Randomised, double‐blind, placebo‐controlled trial with parallel group design (two intervention groups).
Participants Country: Australia.
Number of patients randomised: 136, 44,5% men and 55.5% women, mean age 64 years.
Inclusion criteria: men and women aged 40 years and more fulfilling American College of Rheumatology clinical and radiographic criteria for osteoarthritis knee (all had osteophytes), have pain more than half the days of the previous month and at least one pain dimension of the Western Ontario and McMaster University osteoarthritis index (WOMAC) pain score above 20%. Pain that was at least mild in severity (no compromise of daily activities, frequent but tolerable pain that is worsened by unusual activity and patient may take a pain reliever occasionally).
Exclusion criteria: known sensitivity to vitamin E, current anticoagulation therapy, previous stroke or history of poorly controlled hypertension, major morbidities such as a cancer or life threatening illnesses, inability to co‐operate with study requirements and give informed consent, dementia, other forms of arthritis, inability to walk 50 feet without the use of assistive devices, hemiparesis of either lower limb, those awaiting knee replacement, grade IV knee osteoarthritis, and any contraindication to magnetic resonance imaging (MRI) (eg, pacemaker, cerebral aneurism clip, cochlear implant, presence of shrapnel/metal in strategic locations such as in the orbit, and claustrophobia).
Interventions Patients were randomly assigned to receive:
group 1: vitamin E 500 IU (n = 67);
group 2: placebo (soybean);
for a period of two years.
Outcomes The primary outcome measure was: change in cartilage volume in patients with knee osteoarthritis.
Notes Compliance was assessed by returned pill counts at each visit.
Average compliance, assessed on the basis of residual capsule counts, was similar in the two groups; 95.7% in the vitamin E group and 97.0% in the placebo group. No side effects were attributed to vitamin E.
Losses to follow‐up were five patients in the active and four in the placebo group.
Risk of bias
Bias Authors' judgement Support for judgement
Random sequence generation (selection bias) Low risk Sequence generation was achieved using computer random number generation.
Allocation concealment (selection bias) Low risk Allocation was controlled by a central and independent randomisation unit, so that intervention allocations could not have been foreseen in advance of, or during, enrolment.
Blinding (performance bias and detection bias) 
 All outcomes Low risk The trial was described as blinded, the parties that were blinded, and the method of blinding was described, so that knowledge of allocation was adequately prevented during the trial.
Incomplete outcome data (attrition bias) 
 All outcomes Low risk The numbers and reasons for dropouts and withdrawals in all intervention groups were described.
Selective reporting (reporting bias) Low risk Pre‐defined, or clinically relevant and reasonably expected outcomes are reported on.
Other bias Low risk The trial appears to be free of other components that could put it at risk of bias. 

ADCS: Alzheimer`s Disease Cooperative Study 
 AMDS: Age Related Macular Degeneration Study 
 AREDS: Age Related Eye Disease Study 
 ASAP: The Antioxidant Supplementation in Atherosclerosis Prevention Study 
 ATBC: Alpha‐Tocopherol, Beta‐Carotene Cancer Prevention Study 
 CARET: The Beta‐Carotene and Retinol Efficacy Trial 
 CHAOS: Cambridge Heart Antioxidant Study 
 DATATOP: The Deprenyl and Tocopherol Antioxidative Therapy of Parkinsonism 
 DATOR: D Alpha Tocopherol atORvastatin 
 GISSI: Gruppo Italiano per lo Studio della Sopravvivenza nell'Infarto miocardico 
 HATS: The HDL‐Atherosclerosis Treatment Study 
 HOPE: The Heart Outcomes Prevention Evaluation Study 
 HOPE TOO: The Heart Outcomes Prevention Evaluation Study The Ongoing Outcomes 
 HPS: Heart Protection Study 
 LAST: Lutein Antioxidant Supplementation Trial 
 MAVIS: Mineral And Vitamin Intervention Study 
 MINVITAOX: The Geriatrie/MINéraux, VITamines, et AntiOXydants Network 
 NIT: Nutrition Intervention Trial 
 NPCT: Nutritional Prevention of Cancer Trial 
 NSCPT: Nambour Skin Cancer Prevention Trial 
 PHS: Physicians Health Study 
 PPP: The Primary Prevention Project 
 PPS: The Polyp Prevention Study 
 REACT: The Roche European American Cataract Trial 
 SCPS: Skin Cancer Prevention Study 
 SKICAP‐AK: Skin Cancer Prevention Study ‐ Actinic Keratoses 
 SPACE: Secondary Prevention with Antioxidants of Cardiovascular disease in End stage renal disease 
 SUVIMAX: The SUpplementation en VItamines et Mine´raux AntioXydants 
 VEAPS: The Vitamin E Atherosclerosis Prevention Study 
 VECAT: Vitamin E, Cataract and Age‐Related Maculopathy Trial 
 WAVE: Women's Angiographic Vitamin and Estrogen Trial 
 WHS: Women's Health Study 
 BCC: basal cell skin cancers 
 SCC: squamous cell skin cancer 
 RDA: recommended daily allowance 
 HBsAg: hepatitis B surface antigen 
 AFP: alpha fetoprotein 
 US: United States 
 AST: aspartate aminotransferase 
 ALT: alanine aminotransferase 
 WBC: white blood cells 
 QoL: quality of life 
 ARMD: age‐related macular degeneration

Characteristics of excluded studies [ordered by study ID]

Study Reason for exclusion
Abbey 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Adler 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Afkhami‐Ardekani 2007 This is not a randomised controlled trial.
Aghdassi 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Aghdassi 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Aguiló 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Aguiló 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Akova 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Al‐Taie 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Albanes 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Alberts 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Allard 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Allard 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
America 2008 Randomised clinical trial. This trial included participants younger than 18 years.
Anah 1980 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Anderson 1974 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Anderson 1975 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Anderson 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Anderson 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Andreone 2001 Randomised, double‐blind, placebo‐controlled trial to evaluate vitamin E supplementation as therapy for chronic hepatitis B in a pilot study including 32 patients. Patients were randomly allocated to receive vitamin E at the dose of 300 mg twice daily for 3 months (15 patients) or no treatment (17 patients). This trial did not meat our inclusion criteria.
Angstwurm 1999 Randomised open‐label pilot trial comparing patients with and without selenium replacement. The aim was to determine the effect of selenium replacement on morbidity and mortality in patients with systemic inflammatory response syndrome. Three patients in active treatment group and two patients in placebo group had cancer at the time of randomisation. This trial did not meet our inclusion criteria.
Arad 2005 A double‐blind, placebo‐controlled randomised clinical trial of atorvastatin 20 mg daily, vitamin C 1 g daily, and vitamin E (alpha‐tocopherol) 1,000 U daily versus matching placebos in 1,005 asymptomatic, apparently healthy men and women age 50 to 70 years with coronary calcium scores at or above the 80th percentile for age and gender. All trial participants also received aspirin 81 mg daily. Mean duration of treatment was 4.3 years. The aim of the trial was to determine whether lipid‐lowering therapy and antioxidants retard the progression of coronary calcification and prevent atherosclerotic cardiovascular disease (ASCVD) events. One patient died, but it is not known in which arm. Authors did not respond to our request for further information.
Argyriou 2006 Randomised clinical trial. This trial included patients with cancer.
Arvilommi 1983 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Aryaeian 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Astley 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Avery 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bacic Vrca 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Backman 1990 This is not a randomised trial.
Bailey 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Baillie 2009 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Baines 1988 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Barany 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Barbagallo 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Barbarich 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Barker 2009 Randomised clinical trial. This trial included patients undergoing surgery.
Barringer 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Bartlett 2008 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Basnayake 1983 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bassenge 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bates 1998 This is not a randomised clinical trial.
Beaton 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Beckman 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Behndig 2009 Randomised clinical trial. All participants completed the follow‐up period.
Benton 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Berger 1998 This randomised, placebo‐controlled trial studied clinical and immune effects of trace element supplements. Twenty patients, aged 40 +/‐ 16 y (mean +/‐ SD), burned on 48 +/‐ 17% of their body surfaces, were studied for 30 d after injury. They consumed standard trace element intakes plus supplements (40.4 micromol Cu, 2.9 micromol Se, and 406 micromol Zn; group TE) or standard trace element intakes plus placebo (20 micromol Cu, 0.4 micromol Se, and 100 micromol Zn; group C) for 8 days. Demographic data were similar for both groups. This trial did not fulfil our inclusion criteria.
Bernard 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Berson 1993 Randomised, clinical, double‐masked trial with 2 x 2 factorial design and duration of 4 to 6 years to determine whether supplements of vitamin A or vitamin E alone or in combination affect the course of retinitis pigmentosa. Six‐hundred‐and‐one patients aged 18 through 49 years with retinitis pigmentosa met preset eligibility criteria. Ninety‐five per cent of the patients completed the trial (29/601). Four of 29 patients died and 25 decline to continue participation, most after the fourth year. We were not able to extract relevant data about the mortality in each arm from the published article. Authors were not able to provide these data too.
Bespalov 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bhardwaj 2009 Randomised clinical trial. There were no deaths during the follow‐up period.
Bierenbaum 1985 This is not a randomised clinical trial.
Biesalski 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bjorneboe 1988 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Blackhall 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Block 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bloomer 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Boardley 2000 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Bogden 1990 This is not randomised clinical trial. The objective of this study was to determine the effects of a year of Zn supplementation on Zn concentrations in circulating cells and on cellular immune functions in the elderly. Participants, aged 60 to 89, were given a placebo, 15 mg Zn, or 100 mg Zn daily for 12 months. All participants also received a multivitamin/mineral supplement that contained no additional Zn. Blood samples were drawn and immune functions assessed prior to and at 3, 6, 12, and 16 months after beginning Zn supplementation. Participant diets were also assessed at each visit.
Bogden 1994 A placebo‐controlled double‐blind trial of the effects of daily micronutrient supplements on circulating vitamin and trace metal concentrations and delayed‐hypersensitivity skin test. Participants aged 59 to 85 years, were randomly assigned to placebo (n = 27) or micronutrient (n = 29) treatment groups. Delayed‐hypersensitivity skin test and circulating concentrations of nine micronutrients were measured before and after 6 and 12 months of micronutrient ingestion. Authors reported that one patient died after 1 month of enrolling the study, but did not report in which arm. Letter to the author was sent. Answer was not received.
Booth 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Boshtam 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Boshtam 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bostom 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Brand 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Broome 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Brouwers 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Brown 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Brown 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Brown 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Brude 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Bucca 1989 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Buchman 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bugianesi 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Bukin 1993 This is not a randomised trial. Increase of ornithine decarboxylase (ODC) activity is known to be associated with cell proliferation and, very likely, with tumour promotion. This prompted us to study the activity of ODC in gastric mucosa of patients with chronic atrophic gastritis that has been considered as a precursor of stomach cancer.
Bukin 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bukin 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bunker 1994 This is not a randomised clinical trial. In this study a balanced nutritional supplement consisting of several macro‐ and micro‐nutrients was administered daily to 27 housebound elderly (aged 70 to 85 years) for 12 weeks. Thirty‐one matched participants served as a control group.
Bunout 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Bursell 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Bussey 1982 Randomised, double‐blind trial of 49 patients with polyposis coli. Among the 49 patients, 36 were evaluable. During the trial two participants died, but authors did not report in which group they were. Letter to the author was sent.
Butcher 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Cadenas 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Cafolla 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Calabrese 1987 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Calzada 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Candan 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Carpenter 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Carty 2000 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Cases 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ceriello 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chan 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chandra 2001 Randomised, double‐blind, placebo‐controlled trial to assess cognitive function in apparently healthy, elderly participants. Author reported two deaths during the trial, but not in which arm of the study. The publication on the trial is retracted in 2005. Conflict of interest: Chandra failed to declare that he holds a patent on the tested supplement formula and has a financial stake in it because the supplement was licensed to Javaan Corporation, a company founded by his daughter, that sells the supplement. Letter to the author was sent.
Chandra 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chavance 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chesney 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Cheung 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chuang 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Chuin 2009 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Clarke 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Clarke 2009 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Clausen 1989 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Colette 1988 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Connolly 2006 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Corridan 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Cox 1975 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Crary 1987 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Crimi 2004 Randomised, double‐blind placebo‐controlled trial to examine the effect of antioxidant supplementation in enteral feeding in critically ill patients. Baseline characteristics of patients were reason for exclusion. Eighteen patients in antioxidant group and 16 patients in placebo group had malignancy. This trial did not meet our inclusion criteria.
Crogan 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dabiri 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Daga 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dakhale 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Darko 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Davison 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Davison 2006 Randomised clinical trial. All participants completed the follow‐up period.
Davison 2007 Randomised clinical trial. All participants completed the follow‐up period.
Dawson 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
De las Heras 2000 This is not a randomised clinical trial. The purpose of this report is to analyse the results of a 1‐year clinical study of antioxidant therapy in the treatment of pain and recurrent inflammatory episodes in patients with chronic and acute recurrent pancreatitis, using a prospective, descriptive, pre‐post, open design. The studied patients were with acute recurrent or chronic pancreatitis who had suffered from pain or acute inflammatory episodes the year before the beginning of treatment with a complex containing L‐methionine, beta‐carotene, vitamin C, vitamin E and organic selenium.
de Sanjose 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
de Vet 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
de Waart 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
DeCosse 1989 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dell'Anna 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
DeMaio 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Desideri 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Desideri 2002a Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Devaraj 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Devaraj 2007 Randomised clinical trial. The authors reported deaths during the follow‐up period but not the number of deaths.
Devaraj 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dieber‐Roth 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Diepeveen 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dietrich 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Dietrich 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
DK PRECISE Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dorfman‐Etrog 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Duffy 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Duffy 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dunstan 2007 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Duthie 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Dziaman 2009 Randomised clinical trial. This trial included cancer patients.
Earnest 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Economides 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Edmonds 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Egan 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Eiselt 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
El‐Bayoumy 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Elkashef 1990 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ernster 1985 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Everett 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fairley 1996 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Fairris 1989 Since reduced concentrations of selenium in whole blood, plasma and white cells had previously been observed in psoriasis, 69 patients were supplemented daily with either 600 micrograms of selenium‐enriched yeast, 600 micrograms of selenium‐enriched yeast plus 600 IU of vitamin E, or a placebo for 12 weeks. Before supplementation, the patients' mean concentrations of selenium in whole blood and plasma were reduced compared with those of matched healthy controls, but their red cell glutathione peroxidase (GSH‐Px) activity was normal. During the study, 4 patients were excluded. Authors did not report any deaths during the trial.
Falsini 2003 Non‐randomised, comparative clinical study to evaluate the influence of short‐term antioxidant supplementation on retinal function in age‐related maculopathy (ARM) patients by recording focal electroretinograms.
Fang 2002 Randomised, double‐blind, placebo‐controlled trial to examine the effect of vitamins C and E on progression of transplant‐associated arteriosclerosis. This trial did not meet our inclusion criteria.
Farvid 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Faure 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fawzi 2004 Randomised, double‐blind, placebo‐controlled trial in Dar es Salaam, Tanzania, to examine the effects of daily supplements of vitamin A (preformed vitamin A and beta carotene), multivitamins (vitamins B, C, and E), or both on progression of HIV disease, using survival models. Authors reported that A total of 343 women died during follow‐up. Of these deaths, 243 were deemed to be due or related to AIDS: 82 were due to AIDS, 61 to pulmonary tuberculosis, 3 to extrapulmonary tuberculosis, 10 to anemia, 14 to meningitis, 5 to stroke, 23 to pneumonia, 21 to diarrhea, and 24 to fever. Only data about the mortality related to AIDS (243 women) were reported. This trial did not meet our inclusion criteria.
Fenech 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Finley 1998 This is not a randomised clinical trial. Thirty healthy young men were fed diets that provided either 32.6 or 226.5 mg of selenium (Se)/day for 105 days.
Fischer 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Florencio 1981 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fogarty 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Fortes 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fotherby 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Frank 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fuchs 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fuller 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fuller 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Fumeron 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gaede 2001 Randomised clinical trial. All participants completed the trial.
Gaeini 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gal 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Galley 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Garewal 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gariballa 2007 Randomised clinical trial. This trial included acutely ill patients.
Garmyn 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gartner 2001 Mini review about the effect of a selenium supplementation on the outcome of patients with severe systemic inflammation, burn and trauma.
Gartner 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gauche 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gazis 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gertz 1990 A study of vitamin E in patients with primary systemic amyloidosis. This is not a randomised clinical trial.
Gesch 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Ghatak 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ghosh 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gianduzzo 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gokce 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Goldblum 2009 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Goldfarb 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Goldfarb 2005a Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Goldfarb 2007 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Gollnick 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gomez‐Perez 1996 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Goodman 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gosney 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Goudev 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Green 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Greul 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Griesinger 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Grievink 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Grievink 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gritz 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Guarnieri 2008 Two randomised clinical trials. The authors did not report any deaths during the follow‐up period.
Gueguen 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Gupta 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hajjar 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hamilton 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Harman 1986 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Harrison 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Haskell 2010 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Hasselmark 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hata 1992 This study was designed to prevent the patients already with cerebral infarction from recurrence of cerebrovascular accidents by administration of alpha‐tocopheryl nicotinate. In this article authors described rationales for trial. No results of this study have been published later on.
Hawkes 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hawkes 2009 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Heinle 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Heinrich 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Heitzer 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hernandez 2005 Randomised clinical trial. The Prevention Research Veteran Affairs E‐vitamin Nutrition Trial is a randomised, double‐blind, placebo controlled trial designed to assess the effects of vitamin E supplementation on biomarkers associated with prostate cancer risk in peripheral blood and prostate tissue. A total of 44 patients with increased prostate specific antigen (PSA) and/or abnormal digital rectal examination on initial evaluation were randomised to receive 400 IU vitamin E (22) versus placebo (22). Serum vitamin E, PSA, dehydroepiandrosterone, testosterone, and insulin‐like growth factor‐1 (IGF‐1) were measured in the 2 groups at baseline and then at 3‐month intervals. Results are reported in 28 patients (placebo in 14 and vitamin E in 14) who completed the treatment as specified by the protocol. Three of the 44 randomised patients had biopsy proven prostate cancer at the time of enrollment. This trial did not meet our inclusion criteria.
Herraiz 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Herrick 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hillert 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hininger 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hodis 1995 A subgroup analysis of the on‐trial antioxidant vitamin intake database acquired in the Cholesterol Lowering Atherosclerosis Study, a randomised, placebo‐controlled, serial angiographic clinical trial evaluating the risk and benefit of colestipol‐niacin on coronary artery disease progression.
Hoffman 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hofstad 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hornig 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Huang 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hughes 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Huijuan 1989 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Hurst 2010 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Iino 1977 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Inagaki 1978 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Itoh 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Iwanier 1995 This is not a randomised clinical trial. The objective of this study was to evaluate the effect of selenium (Se) supplementation on Se concentration and glutathione peroxidase (GSH‐Px) activity in blood components and seminal fluid and on spermatozoal quality characteristics in subfertile men. Thirty‐three men were supplemented for 12 weeks with 200 micrograms Se/day in the form of yeast‐rich Se (group I, n = 16) or sodium selenite (group II, n = 17). Blood samples and sperm were collected at the start of the study and after 2, 4, 8, and 12 weeks following Se supplementation.
Jacob 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jacques 1995 Randomised clinical trial.The authors did not report any deaths during the follow‐up period.
Jain 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jantti 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jaswal 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jeng 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jensen 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Jessup 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jialal 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jialal 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jialal 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Johnson 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Jourkesh 2007 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Kahler 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kaikkonen 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kaikkonen 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kaiser 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kanter 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Karanikas 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Karlowski 1975 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kawahara 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Kayan 2009 This is not randomised clinical trial.
Keefe 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Keith 1982 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Keith 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Keskes‐Ammar 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kessopoulou 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Khajehdehi 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Khajehdehi 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kharaeva 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Khassaf 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Kim 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
King 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kinlay 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Kiremidjian‐S 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Kitagawa 1989 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Koh 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Konen 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Korpela 1989 Randomised clinical trial. The effect of selenium supplementation was evaluated in 81 patients with acute myocardial infarction in a double‐blind, placebo‐controlled trial. Patients were randomised into two treatment groups receiving either selenium‐rich yeast (100 micrograms/day) or placebo in addition to conventional drug therapy for a 6‐month period. During treatment the mean serum selenium concentration increased from 82 micrograms/l to 122 micrograms/l (P less than 0.001) in the selenium supplemented group and remained unaltered in the placebo group (83 micrograms/l). The trial did not meat our inclusion criteria. This is a tertiary prevention trial.
Kugiyama 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
la Ruche 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Lagowska‐Lenard 2010 This is not randomised controlled trial.
LAST II 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Leonard 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Li 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Li 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Li 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
London 1983 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
London 1985 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
London 1987 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
London 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Loots 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Louis 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Lovat 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Lykkesfeldt 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mackerras 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
MacLennan 1995 Randomised partially double‐blinded, placebo‐controlled factorial trial. The aim was to assess the effects on the incidence of adenomas of reducing dietary fat to 25% of total calories and supplementing the diet with 25 g of wheat bran daily and a capsule of beta carotene (20 mg daily). Half the patients were assigned to each intervention, resulting in seven intervention groups and one control group. Eligibility criteria included histologic confirmation of at least one colorectal adenoma and confidence expressed by the colonoscopist that all polyps had been removed. Dietary changes were individually initiated and monitored by dietitians and research nurses. At surveillance colonoscopy, the size and location of all polyps were recorded, and their histology was later centrally reviewed. Among 424 patients who were randomly assigned in the trial, 13 were found to be ineligible upon histologic review. Among the remaining 411, complete outcome data were collected from 390 at 24 months and from 306 at 48 months. Eight participants died during the trial, but authors did not report in which arm they were. Letter to the author sent. We did not receive the answer.
MacPherson 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mader 1988 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mahalingam 2011 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Major 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Makinson 1948 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Malo 1986 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mann 1987 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Manzella 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Margaritis 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Marotta 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Martinez‐Abun 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Massey 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Mastaloudis 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Mathews‐Roth 1972 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mazokopakis 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
McAnulty 2010 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
McAuliffe 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
McDowell 1994 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
McGavin 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
McKay 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Meagher 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Meijer 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Meltzer 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Meydani 1990 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Meydani 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Meydani 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Meyer 1990 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Micheletta 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Micozzi 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Miller 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
MIVIT 2005 A randomised, double‐blind, placebo‐controlled trial to test the effects of antioxidant vitamins C and E on the clinical outcome of patients with acute myocardial infarction. Eight‐hundred patients (mean age 62) were randomly allocated to receive, on top of routine medication, one of two treatments: vitamin C (1000 mg/12 h infusion) followed by 1200 mg/24 h orally and vitamin E (600 mg/24 h) or matching placebo for 30 days. The trial did not meet our inclusion criteria.
Mohsenin 1987 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Moller 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mosca 1997 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Mottram 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Mudway 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mulholland 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mulholland 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Mullan 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Munoz 1985 A randomised double‐blind intervention trial was carried out in Huixian, Henan Province, People's Republic of China, to determine whether combined treatment with retinol, riboflavine, and zinc could lower the prevalence of precancerous lesions of the oesophagus. Six‐hundred and ten participants in the age group 35 to 64 were randomised to receive once a week the active treatment (15 mg [50 000 IU] retinol, 200 mg riboflavine, and 50 mg zinc) or placebo. Both at entry to the study and at the end of the treatment, 13.5 months later, the participants were examined, with an emphasis on signs of vitamin A and riboflavine deficiencies, and riboflavine, retinol, beta‐carotene, and zinc levels were measured. Compliance was excellent. The final examination, on 567 (93%) participants, included oesophagoscopy and at least two biopsies. During the study, one person died, but the authors did not report in which arm did this happen, and additional information was not received.
Munoz 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Mustad 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nadeem 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nakhostin‐Roohi 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nelson 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nenseter 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Neunteufl 2000 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Nielsen 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nieman 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nieman 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Nimmagadda 1998 This is an open‐label study to assess the effect of short‐term beta‐carotene administration (180 mg/d with meals for 4 weeks) on the plasma human immunodeficiency virus (HIV) RNA levels and CD4+ lymphocyte counts in 21 HIV‐infected patients. The trial did not meet our inclusion criteria.
Nyyssonen 1994 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
O'Byrne 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Olmedilla 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Olmedilla 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ono 1985 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Onofrj 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Orndahl 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Osilesi 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paganelli 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pallast 1999 A phase II, randomised, double‐blind, placebo‐controlled trial to study the effects of 6 month supplementation with 50 and 100 mg vitamin E on cellular immune responsiveness. The authors did not report any deaths during the trial.
Paolisso 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paolisso 1993a Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paolisso 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paolisso 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paolisso 1995a Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Paolisso 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pardiso Galatioto 2008 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Parisi 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Park 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Pasantes‐Morale 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Patrignani 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pearson 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pellegrini 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pemp 2010 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Peretz 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Peters 1993 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Peters 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Petersen 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Peyser 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Pfeiffer 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Pinkney 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Plantinga 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ponz‐de‐Leon 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Porkkala‐S 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Preziosi 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Prieme 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Princen 1992 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Proteggente 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Racek 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Radhakrishnan 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Raitakari 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ramos 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Range 2003 Randomised clinical trial of zinc and multi‐micronutrient (MMN) supplementation in pulmonary TB patients in Tanzania. A total of 499 pulmonary TB patients were included in the trial after being confirmed sputum‐positive by microscopy or culture. This study did not meet our inclusion criteria.
Rasool 2003 Randomised clinical trial aiming at establishing whether vitamin E improves arterial stiffness in postmenopausal women after 10 weeks of supplementation. Of 20 women, 3 withdrew before the end of trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Ravn‐Haren 2008 Randomised clinical trial with cross‐over design. All participants were supplemented with selenium.
Rayment 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Reaven 1994 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Reaven 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Remans 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Richards 1990 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Rinzler 1950 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Roberts 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Robinson 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Robson 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Rokitzki 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Rokitzki 1994a Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Romieu 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Romney 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Roncucci 1993 Randomised clinical trial evaluating the effect of antioxidant vitamins or lactulose on the recurrence rate of adenomatous polyps. After polypectomy, 255 individuals were randomised into three groups. Group 1 was given vitamin A (30,000 IU/day), vitamin C (1 g/day), and vitamin E (70 mg/day); group 2 was given lactulose (20 g/day); group 3 received no treatment. Forty‐six participants had to be excluded because the histologic diagnosis was not consistent with adenoma. The remaining 209 individuals were included in the analysis according to the 'intention to treat' criterion, though 34 did not adhere to the scheduled treatment, or were lost during the follow‐up. The participants were followed at regular intervals for an average of 18 months. Polyps recurring before one year from index colonoscopy were considered missed by the endoscopist. Two people died during the trial, but the authors did not report in which arm they were.
Roodenburg 2000 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Rossig 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ruiz‐Ramos 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Rytter 2010a Randomised clinical trial. All participants completed the trial.
Rytter 2010b Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sacheck 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Safarinejad 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Salonen 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Samet 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Samman 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sampson 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sankaranarayana 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Schachter 1982 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Schlebusch 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Schneider 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Schorah 1981 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Schroder 2001 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Schutte 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Scott 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Scott 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
SECURE 2001 Randomised clinical trial with a prospective, double‐blind, 3x2 factorial design trial. The Study to Evaluate Carotid Ultrasound changes in patients treated with ramipril and vitamin E (SECURE) was a substudy of the Heart Outcomes Prevention Evaluation (HOPE) trial, that evaluated the effects of long‐term treatment with the angiotensin‐converting enzyme inhibitor ramipril and vitamin E on atherosclerosis progression in high‐risk patients.
Seppanen 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Serwin 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Serwin 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
SETCAP 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Shafat 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Shahar 2004 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Shriqui 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Silva 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Simon‐Schnass 1990 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Simone 2002 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Simons 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Simons 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Singh 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Singh 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Siriwardena 2007 Randomised clinical trial. This trial included critically ill patients.
Sisto 1995 Randomised clinical trial. Eighty‐one patients with coronary artery disease were randomised into four study groups: group 1 (n = 20) patients had stable disease and received oral vitamin E for 4 weeks, and vitamin C and allopurinol 2 days before and 1 day after coronary artery bypass grafting. Group 2 (n = 25) consisted of their controls. Group 3 patients (n = 17) had more unstable disease and received the same medications as group 1, except that vitamin E was given only 2 days before the operation. Group 4 (n = 19) was their controls. This trial did not meet our inclusion criteria.
SKICAP S/B 1997 Randomised clinical trial. There were no deaths during the trial. Additional information was obtained through personal communication with authors.
Skyrme‐Jones 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Spiller 1985 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Stampfer 1988 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Stark 1985 Observational study in attempt to prevent the senile degeneration of the macula treatment with cosaldon A+E.
Steck‐Scott 2004 Randomised clinical trial. Authors did not report any deaths during the follow‐up period.
Steinberg 1998 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Steiner 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Stich 1986 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Stich 1988 Randomised clinical trial. A short‐term intervention trial of vitamin A therapy. Participants were randomly distributed into two groups; one receiving 200,000 IU vitamin A per week (0.14 mg/kg body wt/per day) for 6 months, and the other receiving placebo capsules. There were deaths, but in which arm it is not reported. The authors did not answer to our requests for additional information.
Stone 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Studinger 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Subakir 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Subudhi 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sumida 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sun 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sureda 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Surmen‐Gur 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Sutherland 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tahir 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tam 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tardif 1997 Randomised clinical trial. One month before angioplasty, 317 patients were randomly assigned to receive one of four treatments: placebo, probucol (500 mg), multivitamins (30,000 IU of beta carotene, 500 mg of vitamin C, and 700 IU of vitamin E), or both probucol and multivitamins‐all given twice daily. Patients were treated for four weeks before and six months after angioplasty. The patients received an extra 1000 mg of probucol, 2000 IU of vitamin E, both probucol and vitamin E, or placebo 12 hours before angioplasty, according to their treatment assignments. Base‐line and follow‐up angiograms were interpreted by blinded investigators using a quantitative approach. This trial did not meet our inclusion criteria.
Tarng 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tarp 1985 Randomised clinical trial. All patients completed the trial.
Tauler 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tauler 2008 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tecklenburg 2007 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Teixeira 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tessier 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tessier 2009 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Thomson 1988 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Title 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tofler 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tolonen 1985 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tousoulis 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Traber 2006 Randomised clinical trial. All participants completed the trial.
Trebble 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Trebble 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information obtained through personal communication with authors.
Trenga 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tsai 1978 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Tutuncu 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Uden 1990 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ullegaddi 2005 Randomised clinical trial. This trial included acutely ill patients.
Ullegaddi 2009 Randomised clinical trial. This trial included acutely ill patients.
Upritchard 2000 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Upritchard 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
van Amsterdam 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Van Gossum 1995 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Van Hoydonck 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
van Poppel 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
van Rhijn 1990 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Van Straten 2002 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
van Tits 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Vannas 1958 This is not a randomised clinical trial. A clinico‐pathological investigation; an attempt to cure the arteriosclerotic changes seen in the fundus and affecting vision, and to compare the results achieved by different methods of treatment inclusive of placebo.
Vasankari 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Vasankari 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Vega‐Lopez 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Vela 2000 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Venn 2003 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Verret 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Vertrugno 2001 Randomised clinical trial to evaluate the effect of a high dose vitamin A and E supplementation on corneal re‐epithelialisation time, visual acuity and haze following photorefractive keratectomy (PRK). Two groups of 20 patients who underwent myopic PRK were supplemented with either 25 000 IU retinol palmitate and 230 mg alpha tocopheryl nicotinate or a placebo. Clinical outcomes were evaluated up to 360 days. This trial did not meet our inclusion criteria.
Vincent 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Viscovich 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Volkovova 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
von Herbay 1997 Randomised clinical trial using a double‐blind cross‐over design to evaluate whether treatment of hepatitis C patients refractory to alpha‐interferon therapy with high doses of vitamin E (2 x 400 IU RRR‐alpha‐tocopherol/day) for 12 weeks improves the aminotransferase status. This trial did not meet our inclusion criteria.
Wander 1996 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wang 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Ward 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Watanabe 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Watanabe 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Watson 1991 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Welch 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Wen 1997 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wen 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wenzel 1993 Quasi‐randomised clinical study including 56 patients suffering from acute alcohol hepatitis. Patients were randimised using the date of birth. This study was originally included in our JAMA review but during assessment of the review we realized our mistake.
Werninghaus 1994 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wesnes 2003 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Wijnen 2002 Randomised clinical trial to examine the effect of antioxidants on the activation and sequestration of white blood cells and muscle injury during intra‐abdominal aortic aneurysm repair. Forty‐two patients undergoing elective infrarenal aneurysm repair were randomised to either standard therapy (22 patients) or standard therapy with additional multiantioxidant supplementation (20 patients). Vitamin E and C, allopurinol, N‐acetylcysteine, and mannitol was administered perioperatively. White blood cell count, serum creatine kinase, aspartateaminotransferase, lactate, and lipofuscine were measured. This trial did not meet our inclusion criteria.
Willett 1983 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Willett 1984 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Williams 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Winkler 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Winterbone 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wittenborg 1998 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wolters 2005 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wolvers 2006 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wood 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Woodside 1999 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Wu 2004 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wu 2007 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Wuyi 2001 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Xia 2005 Randomised clinical trial. There were no deaths during the follow‐up period. Additional information was obtained through personal communication with authors.
Xia 2010 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Xu 1992 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Yu 1990 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.
Yu 1991 Randomised clinical trial. The authors did not report number of deaths during the follow‐up period.
Yu 1997 Randomised clinical trial. The authors did not report number of deaths during the follow‐up period.
Zaridze 1993 Randomised clinical trial. The authors did not report number of deaths during the follow‐up period.
Zhu 2002 Randomised clinical trial. The authors did not report number of deaths during the follow‐up period.
Zielinski 1978 Not a randomised clinical trial. Cosaldon A+E (containing vitamin A and vitamin E) was applied in 61 patients, most of whom presented severe vascular degenerative retinochoroidal circulatory disorders and chronic glaucoma.
Zimmermann 1997 In this study the effect of antioxidant therapy with sodium selenite was investigated in patients with systemic inflammatory response syndrome and multiple organ failure. Forty patients were included and observed over a period of 28 days. This study did not meet our inclusion criteria.
Zollinger 1999 Randomised clinical trial. The authors did not report any deaths during the follow‐up period.

PPT: Polyp Prevention Trial 
 wk = week

Differences between protocol and review

1. We changed QUORUM (Moher 1999) into PRISMA (Moher 2009) as the guideline was updated.

2. We assessed the influence of trials with zero mortality in both intervention groups by re‐calculating the RR with 0.5, 0.01, and 0.001 as empirical continuity corrections. The reason is that we used Trial Sequential Analysis version 0.8 (TSA 2008; Thorlund 2011) for those calculations. This computer program has pre‐defined empirical continuity corrections of 0.5, 0.01, and 0.005.

3. Data collection and analysis. Assessment of risk of bias in included trials has been updated according to the Cochrane Handbook for Systematic Reviews of Interventions. We have now added the following risk of bias domains: incomplete outcome data; selective outcome reporting; and other bias.

4. Data collection and analysis. Data synthesis. We also conducted trial sequential analyses with diversity‐adjusted required information size in addition to inconsistency‐adjusted required information size. The reason is that the diversity‐adjusted required information size seems to give less biased estimates of the required information size than the inconsistency‐adjusted required information size (Wetterslev 2009).

Contributions of authors

GB had full access to all data in the review and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: GB, DN, LLG, RGS, CG. 
 Acquisition of data: GB, DN, CG. 
 Analysis and interpretation of data: GB, DN, LLG, RGS, CG. 
 Drafting of the manuscript: GB, DN, LLG, RGS, CG. 
 Critical revision of the manuscript for important intellectual content: GB, DN, LLG, RGS, CG. 
 Statistical analysis: GB, LLG, RGS, CG. 
 Obtained funding: CG. 
 Administrative, technical, or material support: DN, CG. 
 Study supervision: GB, CG.

Sources of support

Internal sources

  • The Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Denmark.

External sources

  • Knowledge and Research Centre for Alternative Medicine (ViFAB), Denmark.

Declarations of interest

None known. The funding sources had no role in the conduct of the study, collection of data, management, analysis, interpretation of the data, or preparation of the manuscript.

New search for studies and content updated (no change to conclusions)

References

References to studies included in this review

ADCS 1 1997 {published data only}

  1. Drachman DA, Leber P. Treatment of Alzheimer's disease ‐ searching for a breakthrough, settling for less. New England Journal of Medicine 1997;336(17):1245‐7. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  2. Kilander L, Ohrvall M. Alpha‐tocopherol and Alzheimer's disease. New England Journal of Medicine 1997;337(8):572‐3. [MEDLINE: ] [PubMed] [Google Scholar]
  3. Pincus MM. Alpha‐tocopherol and Alzheimer's disease. New England Journal of Medicine 1997;337(8):572. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  4. Sano M, Ernesto C, Klauber MR, Schafer K, Woodbury P, Thomas R, et al. Rationale and design of a multicenter study of selegiline and alpha‐tocopherol in the treatment of Alzheimer disease using novel clinical outcomes. Alzheimer's Disease Cooperative Study. Alzheimer Disease and Associated Disorders 1996;10(3):132‐40. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  5. Sano M, Ernesto C, Thomas RG, Klauber MR, Schafer K, Grundman M, et al. A controlled trial of selegiline, alpha‐tocopherol, or both as treatment for Alzheimer's disease. The Alzheimer's Disease Cooperative Study. New England Journal of Medicine 1997;336(17):1216‐22. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]

ADCS 2 2005 {published data only}

  1. Barnes DE, Yaffe K. Vitamin E and donepezil for the treatment of mild cognitive impairment. New England Journal of Medicine 2005;353(9):951‐2. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  2. Blacker D. Mild cognitive impairment ‐ no benefit from vitamin E, little from donepezil. New England Journal of Medicine 2005;352(23):2439‐41. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  3. Petersen RC, Thomas RG, Grundman M, Bennett D, Doody R, Ferris S, et al. Vitamin E and donepezil for the treatment of mild cognitive impairment. New England Journal of Medicine 2005;352(23):2379‐88. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]

Allsup 2004Low {published data only}

  1. Allsup SJ, Shenkin A, Gosney MA, Taylor S, Taylor W, Hammond M, et al. Can a short period of micronutrient supplementation in older institutionalized people improve response to influenza vaccine? A randomized, controlled trial. Journal of the American Geriatrics Society 2004;52(1):20‐4. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]

ALSRT 2001Low {published data only}

  1. Desnuelle C, Dib M, Garrel C, Favier A. A double‐blind, placebo‐controlled randomized clinical trial of alpha‐tocopherol (vitamin E) in the treatment of amyotrophic lateral sclerosis. ALS Riluzole‐Tocopherol Study Group. Amyotrophic Lateral Sclerosis and Other Motor Neuron Disorders 2001;2(1):9‐18. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]

AMDS 1996Low {published data only}

  1. Age Related macular Degeneration Study Group. Multicenter ophthalmic and nutritional age‐related macular degeneration study ‐ part 2: antioxidant intervention and conclusions. Journal of the American Optometric Association 1996;67(1):30‐49. [MEDLINE: ] [PubMed] [Google Scholar]
  2. Age‐related Macular Degeneration Study Group. Multicenter ophthalmic and nutritional age‐related macular degeneration study ‐ part 1: design, subjects and procedures. Journal of the American Optometric Association 1996;67(1):12‐29. [MEDLINE: ] [PubMed] [Google Scholar]

AREDS 2001Low {published data only}

  1. Age‐Related Eye Disease Research Group. The Age‐Related Eye Disease Study (AREDS) system for classifying age‐related macular degeneration from stereoscopic color fundus photographs. AREDS Report No. 6. American Journal of Ophthalmology 2001;132:668‐81. [DOI] [PubMed] [Google Scholar]
  2. Age‐Related Eye Disease Study Research Group. A randomized, placebo‐controlled, clinical trial of high‐dose supplementation with vitamins C and E and beta carotene for age‐related cataract and vision loss: AREDS report no. 9. Archives of Ophthalmology 2001;119(10):1439‐52. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Age‐Related Eye Disease Study Research Group. A randomized, placebo‐controlled, clinical trial of high‐dose supplementation with vitamins C and E, beta carotene, and zinc for age‐related macular degeneration and vision loss: AREDS report no. 8. Archives of Ophthalmology 2001;119(10):1417‐36. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Age‐Related Eye Disease Study Research Group. Impact of antioxidants, zinc, and copper on cognition in the elderly: a randomized, controlled trial. AREDS Report No. 12. Neurology 2004;63(9):1705–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Age‐Related Eye Disease Study Research Group. Risk factors associated with age‐related macular degeneration. A case‐control study in the age‐related eye disease study: age‐related eye disease study report number 3. Ophthalmology 1999;91(20):1738‐43. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Age‐Related Eye Disease Study Research Group. Risk factors associated with age‐related nuclear and cortical cataract: a case‐control study in the Age‐Related Eye Disease Study, AREDS Report No. 5. Ophthalmology 2001;108(8):1400‐8. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Age‐Related Eye Disease Study Research Group. The Age‐Related Eye Disease Study (AREDS): design implications AREDS Report No. 1. Controlled Clinical Trials 1999;20(6):573‐600. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Age‐Related Eye Disease Study Research Group. The age‐related eye disease study (AREDS) system for classifying cataracts from photographs: AREDS report no. 4. American Journal of Ophthalmology 2001;131(2):167‐75. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Age‐Related Eye Disease Study Research Group. The effect of five‐year zinc supplementation on serum zinc, serum cholesterol and hematocrit in persons randomly assigned to treatment group in the age‐related eye disease study: AREDS Report No. 7. Journal of Nutrition 2002;132(4):697‐702. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Age‐Related Eye Disease Study Research Group, SanGiovanni JP, Chew EY, Clemons TE, Ferris FL 3rd, Gensler G, et al. The relationship of dietary carotenoid and vitamin A, E, and C intake with age‐related macular degeneration in a case‐control study: AREDS Report No. 22. Archives of Ophthalmology 2007;125(9):1225‐32. [DOI] [PubMed] [Google Scholar]
  11. Age‐Related Macular Degeneration Study Group. Multicenter ophthalmic and nutritional age‐related macular degeneration study ‐ Part 1: design, subjects and procedures. Journal of the American Optometric Association 1996;67(1):12‐29. [PubMed] [Google Scholar]
  12. Asensio‐Sanchez VM. AREDS and age‐related macular degeneration [Estudio AREDS y degeneracion macular asociada a la edad]. Archivos de la Sociedad Espanola de Oftalmologia 1999;91(20):1738‐43. [MEDLINE: ] [PubMed] [Google Scholar]
  13. Bartlett H, Eperjesi F. Age‐related macular degeneration and nutritional supplementation: a review of randomised controlled trials. Ophthalmic & Physiological Optics 2003;23(5):383‐99. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  14. Bressler NM, Bressler SB, Congdon NG, Ferris FL 3rd, Friedman DS, Klein R, et al. Potential public health impact of Age‐Related Eye Disease Study results: AREDS Report No. 11. Archives of Ophthalmology 2003;121(11):1621‐4. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Chew EY, Clemons T. Vitamin E and the age‐related eye disease study supplementation for age‐related macular degeneration. Archives of Ophthalmology 2005;123(3):395‐6. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  16. Chew EY, Kim J, Sperduto RD, Datiles MB 3rd, Coleman HR, Thompson DJ, et al. Evaluation of the age‐related eye disease study clinical lens grading system AREDS report No. 31. Ophthalmology 2010;117(11):2112‐9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Chew EY, Lindblad AS, Clemons T, Age‐Related Eye Disease Study Research Group. Summary results and recommendations from the age‐related eye disease study. Archives of Ophthalmology 2009;127(12):1678‐9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Clemons TE, Chew EY, Bressler SB, McBee W. National Eye Institute Visual Function Questionnaire in the Age‐Related Eye Disease Study (AREDS): AREDS Report No. 10. Archives of Ophthalmology 2003;121(2):211‐7. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Clemons TE, Kurinij N, Sperduto RD, AREDS Research Group. Associations of mortality with ocular disorders and an intervention of high‐dose antioxidants and zinc in the Age‐Related Eye Disease Study: AREDS Report No. 13. Archive of Ophthalmology 2004;122(5):716‐26. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Clemons TE, Milton RC, Klein R, Seddon JM, Ferris FL 3rd, Age‐Related Eye Disease Study Research Group. Risk factors for the incidence of Advanced Age‐Related Macular Degeneration in the Age‐Related Eye Disease Study (AREDS) AREDS Report no. 19. Ophthalmology 2005;112(4):533‐9. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Clemons TE, Rankin MW, McBee WL, Age‐Related Eye Disease Study Research Group. Cognitive impairment in the Age‐Related Eye Disease Study: AREDS report no. 16. Archives of Ophthalmology 2006;124(4):537‐43. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Cukras C, Agrón E, Klein ML, Ferris FL 3rd, Chew EY, Gensler G, et al. Natural history of drusenoid pigment epithelial detachment in age‐related macular degeneration: Age‐Related Eye Disease Study Report No. 28. Ophthalmology 2010;117(3):489‐99. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Forooghian F, Agrón E, Clemons TE, Ferris FL 3rd, Chew EY, Age‐Related Eye Disease Study Research Group. Visual acuity outcomes after cataract surgery in patients with age‐related macular degeneration: age‐related eye disease study report no. 27. Ophthalmology 2009;116(11):2093‐100. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lindblad AS, Clemons TE. Responsiveness of the National Eye Institute Visual Function Questionnaire to progression to advanced age‐related macular degeneration, vision loss, and lens opacity: AREDS Report No. 14. Archives of Ophthalmology 2005;123(9):1207‐14. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lindblad AS, Lloyd PC, Clemons TE, Gensler GR, Ferris FL 3rd, Klein ML, et al. Change in area of geographic atrophy in the Age‐Related Eye Disease Study: AREDS report number 26. Archives of Ophthalmology 2009;127(9):1168‐74. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Milton RC, Sperduto RD, Clemons TE, Ferris FL 3rd, Age‐Related Eye Disease Study Research Group. Centrum use and progression of age‐related cataract in the Age‐Related Eye Disease Study: a propensity score approach. AREDS report No. 21. Ophthalmology 2006;113(8):1264‐70. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Rankin MW, Clemons TE, McBee WL. Correlation analysis of the in‐clinic and telephone batteries from the AREDS cognitive function ancillary study. AREDS Report No. 15. Ophthalmic Epidemiology 2005;12(4):271‐7. [MEDLINE: ] [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Sackett CS, Schenning S. The age‐related eye disease study: the results of the clinical trial. Insight 2002;27(1):5‐7. [MEDLINE: ] [PubMed] [Google Scholar]
  29. SanGiovanni JP, Chew EY, Agrón E, Clemons TE, Ferris FL 3rd, Gensler G, et al. The relationship of dietary omega‐3 long‐chain polyunsaturated fatty acid intake with incident age‐related macular degeneration: AREDS report no. 23. Archives of Ophthalmology 2009;126(9):1274‐9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. SanGiovanni JP, Chew EY, Clemons TE, Davis MD, Ferris FL 3rd, Gensler GR, et al. The relationship of dietary lipid intake and age‐related macular degeneration in a case‐control study: AREDS Report No. 20. Archives of Ophthalmology 2007;125(5):671‐9. [DOI] [PubMed] [Google Scholar]
  31. Sangiovanni JP, Agrón E, Meleth AD, Reed GF, Sperduto RD, Clemons TE, et al. {omega}‐3 Long‐chain polyunsaturated fatty acid intake and 12‐y incidence of neovascular age‐related macular degeneration and central geographic atrophy: AREDS report 30, a prospective cohort study from the Age‐Related Eye Disease Study. American Journal of Clinical Nutrition 2009;90(6):1601‐7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Sperduto RD, Clemons TE, Lindblad AS, Ferris FL 3rd, Age‐Related Eye Disease Study Research Group. Cataract classification using serial examinations in the age‐related eye disease study: age‐related eye disease study report no. 24. American Journal of Ophthalmology 2008;145(3):504‐8. [DOI] [PubMed] [Google Scholar]

ASAP 2003Low {published data only}

  1. Bruunsgaard H, Poulsen HE, Pedersen BK, Nyyssonen K, Kaikkonen J, Salonen JT. Long‐term combined supplementations with alpha‐tocopherol and vitamin C have no detectable anti‐inflammatory effects in healthy men. Journal of Nutrition 2003;133(4):1170‐3. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  2. Jialal I, Devaraj S. Antioxidants and atherosclerosis: don't throw out the baby with the bath water. Circulation 2003;107(7):926‐8. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  3. Kaikkonen J, Porkkala‐Sarataho E, Morrow JD, Roberts LJ 2nd, Nyyssonen K, Salonen R, et al. Supplementation with vitamin E but not with vitamin C lowers lipid peroxidation in vivo in mildly hypercholesterolemic men. Free Radical Research 2001;35(6):967‐78. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  4. Porkkala‐Sarataho E, Salonen JT, Nyyssonen K, Kaikkonen J, Salonen R, Ristonmaa U, et al. Long‐term effects of vitamin E, vitamin C, and combined supplementation on urinary 7‐hydro‐8‐oxo‐2'‐deoxyguanosine, serum cholesterol oxidation products, and oxidation resistance of lipids in nondepleted men. Arteriosclerosis, Thrombosis, and Vascular Biology 2000;20(9):2087‐93. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  5. Salonen JT, Nyyssonen K, Salonen R, Lakka HM, Kaikkonen J, Porkkala‐Sarataho E, et al. Antioxidant Supplementation in Atherosclerosis Prevention (ASAP) Study: a randomized trial of the effect of vitamins E and C on 3‐year progression of carotid atherosclerosis. Journal of Internal Medicine 2000;248(5):377‐86. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  6. Salonen RM, Nyyssonen K, Kaikkonen J, Porkkala‐Sarataho E, Voutilainen S, Rissanen TH, et al. Six‐year effect of combined vitamin C and E supplementation on atherosclerotic progression: the Antioxidant Supplementation in Atherosclerosis Prevention (ASAP) Study. Circulation 2003;107(7):947‐53. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]

ATBC 2003Low {published data only}

  1. Albanes D, Heinonen OP, Huttunen JK, Taylor PR, Virtamo J, Edwards BK, et al. Effects of alpha‐tocopherol and beta‐carotene supplements on cancer incidence in the Alpha‐Tocopherol Beta‐Carotene Cancer Prevention Study. American Journal of Clinical Nutrition 1995;62 Suppl(6):1427‐30. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  2. Albanes D, Heinonen OP, Taylor PR, Virtamo J, Edwards BK, Rautalahti M, et al. Alpha‐tocopherol and beta‐carotene supplements and lung cancer incidence in the alpha‐tocopherol, beta‐carotene cancer prevention study: effects of base‐line characteristics and study compliance. Journal of the National Cancer Institute 1996;88(21):1560‐70. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  3. Albanes D, Malila N, Taylor PR, Huttunen JK, Virtamo J, Edwards BK, et al. Effects of supplemental alpha‐tocopherol and beta‐carotene on colorectal cancer: results from a controlled trial (Finland). Cancer Causes & Control 2000;11(3):197‐205. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  4. Blumberg J. The Alpha‐Tocopherol, Beta‐Carotene Cancer Prevention Study in Finland. Nutrition Reviews 1994;52(7):242‐50. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  5. Freedman DM, Tangrea JA, Virtamo J, Albanes D. The effect of beta‐carotene supplementation on serum vitamin D metabolite concentrations. Cancer Epidemiology, Biomarkers & Prevention 1999;8(12):1115‐6. [MEDLINE: ] [PubMed] [Google Scholar]
  6. Hartman TJ, Dorgan JF, Woodson K, Virtamo J, Tangrea JA, Heinonen OP, et al. Effects of long‐term alpha‐tocopherol supplementation on serum hormones in older men. Prostate 2001;46(1):33‐8. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  7. Kataja‐Tuomola M, Sundell JR, Männistö S, Virtanen MJ, Kontto J, Albanes D, et al. Effect of alpha‐tocopherol and beta‐carotene supplementation on the incidence of type 2 diabetes. Diabetologia 2008;51(1):47‐53. [DOI] [PubMed] [Google Scholar]
  8. Leppala JM, Virtamo J, Fogelholm R, Huttunen JK, Albanes D, Taylor PR, et al. Controlled trial of alpha‐tocopherol and beta‐carotene supplements on stroke incidence and mortality in male smokers. Arteriosclerosis, Thrombosis, and Vascular Biology 2000;20(1):230‐5. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  9. Liede KE, Haukka JK, Saxen LM, Heinonen OP. Increased tendency towards gingival bleeding caused by joint effect of alpha‐tocopherol supplementation and acetylsalicylic acid. Annals of Medicine 1998;30(6):542‐6. [MEDLINE: ] [PubMed] [Google Scholar]
  10. Malila N, Taylor PR, Virtanen MJ, Korhonen P, Huttunen JK, Albanes D, et al. Effects of alpha‐tocopherol and beta‐carotene supplementation on gastric cancer incidence in male smokers (ATBC Study, Finland). Cancer Causes & Control 2002;13(7):617‐23. [MEDLINE: ] [DOI] [PubMed] [Google Scholar]
  11. Malila N, Virtamo J, Virtanen M, Albanes D, Tangrea J, Huttunen J. The effect of alpha‐tocopherol and beta‐carotene supplementation on colorectal adenomas in middle‐aged male smokers. Cancer Epidemiology, Biomarkers & Prevention 1999;8(6):489‐93. [MEDLINE: ] [PubMed] [