Calcium intake and risk of fracture: systematic review

Abstract

Objective To examine the evidence underpinning recommendations to increase calcium intake through dietary sources or calcium supplements to prevent fractures.

Design Systematic review of randomised controlled trials and observational studies of calcium intake with fracture as an endpoint. Results from trials were pooled with random effects meta-analyses.

Data sources Ovid Medline, Embase, PubMed, and references from relevant systematic reviews. Initial searches undertaken in July 2013 and updated in September 2014.

Eligibility criteria for selecting studies Randomised controlled trials or cohort studies of dietary calcium, milk or dairy intake, or calcium supplements (with or without vitamin D) with fracture as an outcome and participants aged >50.

Results There were only two eligible randomised controlled trials of dietary sources of calcium (n=262), but 50 reports from 44 cohort studies of relations between dietary calcium (n=37), milk (n=14), or dairy intake (n=8) and fracture outcomes. For dietary calcium, most studies reported no association between calcium intake and fracture (14/22 for total, 17/21 for hip, 7/8 for vertebral, and 5/7 for forearm fracture). For milk (25/28) and dairy intake (11/13), most studies also reported no associations. In 26 randomised controlled trials, calcium supplements reduced the risk of total fracture (20 studies, n=58 573; relative risk 0.89, 95% confidence interval 0.81 to 0.96) and vertebral fracture (12 studies, n=48 967. 0.86, 0.74 to 1.00) but not hip (13 studies, n=56 648; 0.95, 0.76 to 1.18) or forearm fracture (eight studies, n=51 775; 0.96, 0.85 to 1.09). Funnel plot inspection and Egger’s regression suggested bias toward calcium supplements in the published data. In randomised controlled trials at lowest risk of bias (four studies, n=44 505), there was no effect on risk of fracture at any site. Results were similar for trials of calcium monotherapy and co-administered calcium and vitamin D. Only one trial in frail elderly women in residential care with low dietary calcium intake and vitamin D concentrations showed significant reductions in risk of fracture.

Conclusions Dietary calcium intake is not associated with risk of fracture, and there is no clinical trial evidence that increasing calcium intake from dietary sources prevents fractures. Evidence that calcium supplements prevent fractures is weak and inconsistent.

Introduction

Older men and women are recommended to take at least 1000-1200 mg/day of calcium for bone health and prevention of fractures.¹ The average intake in the diet in Western countries is 700-900 mg/day, and lower in Asia and Africa, meaning that most older people would need to take calcium supplements to meet these recommendations. These guidelines for calcium intake have been widely implemented, and, in some Western countries, more than 30-50% of older women take calcium supplements.^{2 3 4 5} Clinical trials of calcium supplements at doses of 1000 mg/day, however, have reported adverse effects, including cardiovascular events,^{6 7 8} kidney stones,⁹ and hospital admissions for acute gastrointestinal symptoms.¹⁰ Consequently, older people have been encouraged to improve bone health by increasing their calcium intake through food rather than by taking supplements.¹¹ This advice assumes that increasing dietary calcium intake to the recommended level of >1200 mg/day prevents fractures without causing the adverse effects of calcium supplements.

We assessed the evidence supporting the recommendation to increase dietary calcium intake to prevent fractures and compared the anti-fracture efficacy of increasing calcium intake through dietary sources with the anti-fracture efficacy of calcium supplements. We undertook a systematic review of studies of dietary sources of calcium or calcium supplements in older adults (>50) with fracture as an endpoint. We primarily focused on the results of randomised controlled trials, but when insufficient evidence from such trials was available, we considered results of observational studies.

Methods

Literature search

In July 2013, we searched Ovid Medline and Embase since inception for English language studies of calcium, milk, or dairy intake, or calcium supplements that reported on a broad range of skeletal and non-skeletal endpoints including fracture. The full text of the search was designed with assistance from a professional librarian and is shown in appendix 1. From this search, we also identified 120 systematic reviews or meta-analyses on these topics and hand searched these articles, any other articles included in our review, and recent review articles on fracture risk for other relevant articles. In September 2014, we updated the results with a focused search (no language restrictions) of PubMed (appendix 1) and Embase for studies with fracture or bone mineral density as an endpoint.

Study selection

We included randomised controlled trials and cohort, case-control, or cross sectional studies with fracture as an outcome in which participants were aged >50 at baseline, or for cohort studies, where most follow-up occurred in participants aged >50. We excluded studies where most participants had a major systemic pathology at baseline other than osteoporosis, such as renal failure or malignancy. We included studies of calcium supplements used in combination with other treatment provided that the other treatment was given to both arms (for example, calcium plus oestrogen v placebo plus oestrogen), and included studies of co-administered calcium and vitamin D supplements (CaD). We classified milk, dairy products, and dietary calcium intake from food as dietary sources of calcium. We treated hydroxyapatite as a dietary source of calcium, though it is not a food because hydroxyapatite supplements are made from bone and contain other minerals, hormones, protein, and amino acids in addition to calcium. Several cohort studies reported analyses of calcium intake and fracture risk in more than one publication. We included the results from the publication that reported the longest duration of follow-up for the cohort. Superseded publications are listed in appendix 1. Titles and abstracts were screened by one author (WL or MJB) and the full text of potentially relevant studies reviewed by two authors independently (WL, MJB, VT, or SB). The flow of articles is shown in appendix 2.

Data extraction

From each study we extracted information on characteristics of participants, study design, funding source and conflicts of interest, and numbers of participants with total, hip, forearm, and vertebral fractures. When data were reported for non-vertebral fracture but not total fracture, we treated non-vertebral fractures as total fractures. A single author (WL, MJB, or VT) extracted data, which were checked by a second author (MJB or SB). Risk of bias was assessed as recommended in the Cochrane Handbook,¹² and we planned a subgroup analysis for each fracture outcome stratified by risk of bias. Any discrepancies were resolved through discussion.

Incorporation of studies

In one randomised controlled trial¹³ it was not clear whether the data reported were total number of fractures or number of participants with a fracture. Another was described as a cluster trial of three different fracture prevention programmes: CaD, an environmental programme, or both.¹⁴ Treatment was randomly assigned to each cluster, however, which was based on location of residence and there were only four clusters (one cluster per treatment group), so in effect participants were quasi-randomised by location. The CaD and environmental programmes included an intervention—a home visit by a nurse to review treatment—which was not offered to the control group. Thus, the best estimate of the effect of CaD in the study is a comparison of both programmes (CaD and environmental) with the environmental programme, whereas the comparison of CaD versus no CaD assesses a multifactorial intervention. For these reasons, we considered these two randomised controlled trials to be at high risk of bias and included them only in sensitivity analyses. One trial was described in the methods as a cluster randomised controlled trials but was analysed as individually randomised.^{15 16} We analysed the trial as a cluster trial in the primary analyses, using the approach recommended in the Cochrane handbook¹² with an intracluster correlation coefficient of 0.023^{17 18} and an estimated average cluster size of 3.5. In sensitivity analyses we analysed the trial as individually randomised. In one trial⁹ there was an interaction between oestrogen treatment, CaD treatment, and risk of hip fracture.¹⁹ In women taking oestrogen, CaD reduced risk of hip fracture (relative risk 0.59, 95% confidence interval 0.38 to 0.93), whereas in women not taking oestrogen, CaD had no effect on risk (1.20, 0.85 to 1.69).¹⁹ We included the data for all participants in the trial in the primary analyses but used results of participants not taking oestrogen from this reanalysis in sensitivity analyses.

Statistics

For randomised controlled trials, data were pooled with random effects meta-analyses and heterogeneity was assessed with the I² statistic (I² >50% was considered significant heterogeneity). We used funnel plots and Egger’s regression model to assess for bias. For the primary analyses, we assessed the effects of calcium with or without vitamin D, and in subgroup analyses we assessed calcium monotherapy and co-administered CaD separately. Randomised controlled trials of CaD versus vitamin D, in which the groups differed only in treatment by calcium, were included in subgroup analyses of calcium monotherapy, while trials of CaD versus placebo or controls were included in the CaD subgroup analyses. For trials with factorial designs or more than two arms, in which multiple comparisons can occur, we included all available data from the study. Thus, for factorial randomised controlled trials we included all study arms that allowed a comparison of calcium versus no calcium in the primary analyses and the calcium monotherapy subgroup analysis, but only arms comparing CaD with controls in the CaD subgroup analysis. For multi-arm randomised controlled trials, we pooled data from the separate treatment arms for the primary analyses, but each treatment arm was used only once. We undertook analyses of prespecified subgroups (risk of bias, calcium monotherapy versus CaD, participants living in the community versus residential care, and baseline dietary calcium intake <800 mg/day) with a random effects model and performed a test for interaction between subgroups. Sensitivity analyses were performed to explore the effects of incorporating different study designs and risk of bias. All tests were two tailed and P<0.05 was considered significant. All analyses were performed with Comprehensive Meta-Analysis (Version 2, Biostat, Englewood, NJ, USA).

For prospective cohort studies, authors reported their data in four different ways: the risk of fracture by group with the cohort divided into two to five groups by baseline dietary intake; pooled risk of fracture per unit of dietary intake; mean baseline dietary intake in individuals with or without subsequent fracture; or a written description of any association. We used only one association from each study for each fracture outcome with priority assigned in the order listed. These four different types of data cannot be combined in a meta-analysis and therefore we did not pool the results of different studies. Instead, we assessed whether there was an association between dietary intake and risk of fracture for each study. We classified associations into four groups: no association, inverse association (where a higher intake was associated with a lower risk of fracture, or a lower intake with a higher risk), a positive association (where a higher intake was associated with a higher risk of fracture or a lower intake with a lower risk), or a U shaped association (where both higher and lower intakes were associated with a higher risk of fracture). We considered associations to be present when there were significant differences between mean baseline dietary intakes (assessed by t tests either reported in the paper or calculated post hoc with OpenEpi; www.OpenEpi.com) or when the confidence interval for a group excluded 1. For studies that reported data from three or more groups of dietary intake, we assessed the results for the group furthest from the reference group. Thus, when the reference group had the lowest dietary intake, we assessed results from the group with the highest intake; when the reference group had the highest dietary intake, we assessed results from the group with the lowest intake; and when the reference group had intermediate dietary intake, we assessed results from the groups with both highest and lowest intake.

Results

Dietary sources of calcium

Randomised controlled trials

We identified two randomised controlled trials of dietary sources of calcium: milk powder in one (n=200, calcium dose 800 mg/day, vitamin D dose 240 IU/day)²⁰ and a preparation of hydroxyapatite in the other (n=62, calcium dose 800 mg/day).²¹ Table 1 and table A in appendix 3 show the study designs and selected baseline characteristics. For the randomised controlled trial of milk powder, there was one fracture in the milk group and three in the controls (relative risk 0.33, 95% confidence interval 0.04 to 3.2; P=0.34). For the trial of the hydroxyapatite preparation, fracture data were not reported separately for the hydroxyapatite arm (n=31 participants) but were reported for the 62 participants receiving hydroxyapatite or calcium supplements and are included in the analyses of calcium supplements.

Table 1.

Study design and selected baseline characteristics of randomised controlled trials of calcium intake that reported on fractures. Data are means (SD) unless stated

Trial	Design	Calcium dose (mg/day)	Vitamin D dose	Duration	Care setting	Primary endpoint	Participants Ca/controls	Age (years)	% Female
Dietary calcium trials
Chevalley 199421	3 arm RCT OMC/D; CaD; P/D	800	300 000 IU IM stat	18 m	Community	BMD	62/31	72 (6)	88
Lau 200120	2 arm RCT milk powder; control	800	240 IU/d	2 y	Community	BMD	100/100	57 (6)	100
Calcium supplement trials
Inkovaara 198313	2*3 factorial RCT Ca; D; M; CaD; CaM; DM; CaDM; P	1200	1000 IU/d	12 m	Residential care	Biochemistry	171/156	80 (7)	83
Hansson 198773	4 arm RCT 30 mg NaF/Ca; 10 mg NaF/Ca; Ca; P	1000	—	3 y	Community	BMC	25/25	66 (6)	100
Chapuy 1992,199415 16	2 arm cluster RCT CaD; P	1200	800 IU/d	3 y	Residential care	Fracture	1634/1636	84 (6)	100
Reid 1993,199574 75	2 arm RCT Ca; P	1000		4 y	Community	BMD	68/67	58 (5)	100
Chevalley 199421	3 arm RCT CaD; OMC/D; P/D	800	300,000 IU IM stat	18 m	Community	BMD	62/31	72 (6)	88
Recker 199676	2 arm RCT Ca; P	1200	—	4.3 y	Community	Fracture	95/102	73 (7)	100
Dawson-Hughes 199777	2 arm RCT CaD; P	500	700 IU/d	3 y	Community	BMD	187/202	71 (5)	55
Riggs 199878	2 arm RCT Ca; P	1600	—	4 y	Community	BMD	119/117	66 (3)	100
Baron 199979 80	2 arm RCT Ca; P	1200	—	4 y	Community	Colorectal adenoma	464/466	61 (9)	28
Ruml 199981	2 arm RCT Ca; P	800	—	2 y	Community	BMD	29/34	52 (4)	100
Peacock 200082	3 arm RCT Ca; 25OHD; P	750	—	4 y	Community	BMD	126/135	74 (8)	72
Chapuy 200283	3 arm RCT CaD; CaD; P	1200	800 IU/d	2 y	Residential care	25OHD	389/194	85 (7)	100
Avenell 200484	2*2 factorial RCT Ca; D; CaD; control	1000	800 IU/d	46 m	Community	Compliance/ retention	64/70	77 (5)	82
Fujita 200485	3 arm RCT Ca; Ca; P	900	—	2 y	Residential care	BMD	38/20	80 (7)	100
Harwood 200486	4 arm RCT CaD; CaD; D; control	1000	300 000 IU IM stat or 800 IU/d	12 m	Community	Biochemistry	75/75	81	100
Larsen 200414	4 arm cluster RCT Env; CaD; Env/CaD; control	1000	400 IU/d	42 m	Community	Fracture	4957/4648	75	60
Grant 200587	2*2 factorial RCT Ca; CaD; D; P	1000	800 IU/d	45 m	Community	Fracture	2617/2675	77 (6)	85
Porthouse 200588	2 arm RCT CaD; control	1000	800 IU/d	25 m	Community	Fracture	1321/1993	77 (5)	100
Jackson 20069	2 arm RCT CaD; control	1000	400 IU/d	7 y	Community	Fracture	18176/18106	62 (7)	100
Prince 200689	2 arm RCT Ca; control	1200	—	5 y	Community	Fracture	730/730	75 (3)	100
Reid 200690	2 arm RCT; Ca; control	1000	—	5 y	Community	Fracture	732/739	74 (4)	100
Bolton-Smith 200791	2*2 factorial RCT CaD; CaD/vit K; vit K; P	1000	400 IU/d	2 y	Community	BMD	62/61	68 (6)	100
Bonnick 200792	3 arm RCT CaD/alend; CaD; Alend/D	1000	—	2 y	Community	BMD	282/281	66 (9)	100
Reid 200893	3 arm RCT Ca; Ca; P	600 or 1200	—	2 y	Community	BMD	216/107	56 (10)	0
Salovaara 201094	2 arm RCT CaD; control	1000	800 IU/d	3 y	Community	Fracture	1718/1714	67 (2)	100
Sambrook 201218	3 arm cluster RCT Ca/UV; UV; control	600	UV exposure	1 y	Residential care	Falls	207/190	86 (6)	69

RCT=randomised controlled trial; OMC=ossein-mineral complex (hydroxyapatite); D=vitamin D; CaD=co-administered calcium and vitamin D; P=placebo; Ca=calcium; M=methanedione; NaF=sodium fluoride; BMC=bone mineral content; BMD=bone mineral density; IM=intramuscular; 25OHD=25-hydroxyvitamin D; Env=environmental programme; vit K=vitamin K; alend= alendronate; UV=ultraviolet light.

Cohort studies

As there were too few randomised controlled trials of dietary calcium intake that reported fracture to draw conclusions, we analysed observational studies. We identified 50 publications^{22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71} from 44 cohort studies reporting relations between dietary calcium (n=37), milk (n=14), dairy intake (n=8), or calcium supplements (n=11) and fracture outcomes. There were sufficient cohort studies to analyse, so we did not analyse case-control or cross sectional studies, which are considered a lower level of evidence. Table 2 and table C in appendix 3 show the study design and selected characteristics of the cohort studies.

Table 2.

Study design and selected characteristics of cohort studies reporting fractures. Data are mean (SD) or range unless stated. For dietary calcium, milk, and dairy intake, and calcium supplement, “yes” indicates data reported for this variable in article

Author	No in group	% Female	Duration	Age (years)	Dietary calcium intake	Milk intake	Dairy intake	Calcium supplement	No with fracture
									Total	Hip	Vertebra	Forearm
Riggs 198222	72	100	5 y	64	—	—	—	Yes	—	—	107*	—
Holbrook 198823	957	55	14 y	50-79	Yes	—	—	—	—	33	—	—
Wickham 198924	1419	49	15 y	≥65	Yes	—	—	—	—	44	—	—
Paganini-Hill 199125	13 649	NS	7 y	73	Yes	—	—	Yes	—	418	—	—
Looker 199326	2226†	100	14.6 y	50-74	Yes	—	—	—	—	122	—	—
Huang 199627	2513†	100	13.4 y	62 (9)	—	—	Yes	—	—	130	—	—
Cumming 199728	9704	100	6.6 y	72	Yes	Yes	—	Yes	1950	332	389	467
Fujiwara 199729	4573	65	14 y	59 (12)	—	Yes	—	—	—	55	—	—
Meyer 199730	39 787	50	11.4 y	47 (5)	Yes	Yes	—	—	—	213	—	—
Owusu 199731	43 063	0	8 y	54 (10)	Yes‡	Yes	—	Yes‡	—	56	—	201
Mussolino 199832	2879†	0	22 y	61	Yes	—	—	—	—	71	—	—
Munger 199933	32 050	100	3.3 y	61 (4)	Yes	Yes	Yes	Yes	—	44	—	—
Honkanen 200034	11 798†	100	5 y	52 (3)	Yes	—	—	—	—	—	—	368
Huopio 200035	3068†	100	3.6 y	53	Yes	—	—	—	257	—	—	—
Kato 200036	6250	100	7.6 y	58	Yes	—	—	—	1025	—	—	193§
Nguyen 200137	1844†	60	7.6 y	70 (7)	Yes	—	—	—	—	—	—	121
Dargent-Molina 200238	1588	100	3.7 y	81	Yes	—	—	—	—	NS	—	—
Albrand 200339	672	100	5.3 y	59	Yes	—	—		75	—	—	—
Feskanich 200340	72 337	100	18 y	60	Yes	Yes	—	Yes		603	—	—
Michaelsson 200341	60 689†	100	11 y	54	—	Yes	Yes	—	3986	1535	—	—
Melton 200342	225	100	14 y	68	Yes	—	—	—	126	—	—	—
Roy 200343	6575	52	3.8 y	63 (8)	—	Yes	—	—	—	—	224	—
van def Klift 200444	3001	54	6.3 y	66 (7)	Yes	—	—	—	—	—	157	—
Kanis 200545	39 563**	69	3.8 y	64	—	Yes	—	—	2469	413	—	—
Papaioannou 200546	5143	100	3 y	63 (10)	Yes¶	—	—	—	280	—	34	—
Cauley 200747	159 579	100	8 y	63 (7)	Yes¶	—	—	—	23 270	—	—	—
Diez-Perez 200748	5146	100	3 y	72 (5)	Yes	—	—	—	311	49	—	104
Key 200749	34 696	77	5.2 y	47	Yes	—	—	—	1898	—	—	—
Kung 200750	1435	100	5 y	63 (8)	Yes	—	—	—	80	—	—	—
Lewis 200751	5876	0	4.1 y	74	Yes¶	—	—	—	275	—	—	—
Nguyen 200752	924†	100	10 y	69 (6)	Yes	—	—	—	221	24	76	—
Van Geel 200753	2367	100	10 y	62 (7)	Yes	—	—	—	380	—	—	—
Dargent-Molina 200854	36 217	100	8.4 y	56 (6)	Yes	—	—	Yes	2408	—	—	—
Meier 200855	609†	0	5.8 y	73 (6)	Yes	—	—	—	113	27	55	—
Nieves 200856	52 144	100	3.3 y	65	Yes	—	—	—	2205	337	—	—
Koh 200957	63 154	56	7.1 y	56	Yes¶	—	—	Yes	—	968	—	—
Nakamura 200958	75 879	54	10 y	52 (8)	Yes	Yes	—	—	—	—	364	—
Thomas-John 200959	257	0	3 y	77 (4)	—	—	Yes	Yes	41	—	—	—
Gronskag 201060	4851	100	9.3 y	73	—	Yes	—	—	—	391	—	—
Benetou 201161	29 122	64	8 y	64	Yes	—	Yes	—	—	275	—	—
Nakamura 201162	773	100	5.5 y	75 (4)	Yes	—	—	—	51	—	—	—
Warensjo 201163	61 433†	100	19 y	54	Yes	—	—	—	14 738	3871	—	—
Khan 201264	12 528	NS	13-14 y	45-64	Yes	—	—	—	824	—	—	—
Rouzi 201265	707	100	5.2 y	61 (7)	Yes	—	—	—	138	—	—	—
Feart 201366	1482†	63	8 y	76 (5)	—	Yes	Yes	—	155	57	43	73
Prentice 201367	46 892	100	7.2 y	50-79	—	—	—	Yes	6640	451	—	—
Samieri 201368	1482†	63	8 y	76 (5)	Yes	—	—	Yes	155	—	—	—
Sahni 201369	3212	56	12 y	55 (10)	—	Yes	Yes	—	—	43	—	—
Domiciano 201470	707	64	4.3 y	73 (5)	—	—	Yes	—	—	—	111	—
Sahni 201471	764	NS	11.6 y	77 (5)	—	Yes	—	—	—	97	—	—

NS=not stated, IF=funding by grants from independent funders; Ind=funded by grants from industry and/or run by industry.

*Data are number of vertebral fractures not number of participants with vertebral fractures.

†Reports from same cohort studies. Report with longest duration of follow-up and/or most number of fractures for each association included.

‡Reported total calcium intake divided into dairy and non-dairy intake. Dairy calcium intake treated as dietary intake, and non-dairy intake treated as supplemental calcium intake.

§Data for forearm and hip fracture not reported separately; includes 34 hip fractures.

¶Reported total calcium intake only. Treated as dietary calcium intake because most total calcium intake was from dietary sources.

**Individual patient meta-analysis of six cohort studies.

Tables 3-5 and tables E-F in appendix 3 summarise the results of these cohort studies . For dietary calcium, 14/22 studies (32 853 with fracture/291 273 participants) reported no relation between calcium intake and total fracture (table 3), 17/21 no relation with hip fracture (2629 with fracture/329 414 participants) (table 4), 7/8 no relation with vertebral fracture (711 with fracture/54 140 participants) (table 5), and 5/7 no relation with forearm fracture (1065 with fracture/65 268 participants) (table 5). Thus, 43 of the 58 (74%) reported associations between dietary calcium intake and fracture outcomes were neutral. When relations were reported, they were usually inverse (13/15 associations), with one study describing a positive relation and one study a U shaped relation. Of these 15 associations, 14 reported a numerical relative risk estimate, and 11 of these 14 estimates were between 0.5 and 2.0, which are considered weak associations in observational studies.⁷² For milk and dairy intake (tables D and E in appendix 3), nearly all studies reported no association with fracture risk, with 25/28 neutral associations for milk intake and fracture risk and 11/13 for dairy intake.

Table 3.

Association between dietary calcium intake and risk of total fracture in cohort studies

Study	Fracture*/participants	Association	Groups	Risk or daily calcium intake†		Group 3 risk	Group 4 risk	Group 5 risk	Cut points between each group (mg/d)‡ or unit for pooled risk
				Group 1	Group 2
Cauley 200747	23 270/159 579	Nil	—	NR§	NR§	NR§	—	—	—
Lewis 200751	275/5876	Nil	—	NR§	NR§	NR§	—	—	—
Albrand 200339	75/672	Nil	2	824 (313)	804 (270)	—	—	—	No fracture; fracture
Nguyen 200752	221/924	Nil	2	583 (284)	555 (300)	—	—	—	No fracture; fracture
Samieri 201368	155/1482	Inverse	2	871 (439)	796 (398)	—	—	—	No fracture; fracture
Huopio 200035	257/3068	Nil	—	1.10 (0.99 to 1.23)	—	—	—	—	Per quartile decrease
Melton 200342	126/225	Inverse	—	1.29 (1.06 to 1.56)	—	—	—	—	Per SD decrease
Papaioannou 200546	280/5143	Nil	—	1.005 (0.925 to 1.093)	—	—	—	—	Per 500 mg/d increase
Meier 200855	113/609	Nil	—	1.43 (1.17 to 1.78)	—	—	—	—	Per SD (322 mg/d) decrease
Diez-Perez 200748	311/5146	Inverse	2	1.92 (1.30 to 2.86)	1	—	—	—	250
Kung 200750	80/1435	Inverse	2	3.1 (1.9 to 5.2)	1	—	—	—	400
Van Geel 200753	380/2367	Nil	2	1.0 (0.8 to 1.2)	1	—	—	—	900
Khan 201264	824/12 528	Inverse	2	1	0.75 (0.60 to 0.94)	—	—	—	Lowest quintile; highest quintile
Rouzi 201265	138/707	Inverse	2	1.66 (1.08 to 2.53)	1	—	—	—	391
Cumming 199728	1950/9704	Nil	4	1	1.0 (0.9 to 1.1)	0.9 (0.7 to 1.1)	0.9 (0.7 to 1.1)	—	400; 800; 1200
Kato 200036	1025/6250	Nil	5	1	1.06 (0.9 to 1.3)	0.93 (0.8 to 1.1)	1.10 (0.9 to 1.3)	0.92 (0.8 to 1.1)	569; 689; 799; 949
Key (F>50 y) 200749	888/NS	Inverse	5	1.53 (1.05 to 2.23)	1.31 (0.98 to 1.77)	1.10 (0.87 to 1.39)	1.05 (0.87 to 1.27)	1	525; 700; 900; 1200
Key (M) 200749	343/7947	Nil	5	1.15 (0.63 to 2.09)	0.94 (0.59 to 1.49)	0.91 (0.62 to 1.32)	1.02 (0.76 to 1.37)	1	525; 700; 900; 1200
Dargent-Molina 200854	2408/36 217	Nil	4	1	1.05 (0.94 to 1.19)	1.00 (0.89 to 1.13)	0.91 (0.80 to 1.03)	—	829; 995; 1201
Nieves 200856	2205/52 144	Nil	3	1	0.94 (0.80 to 1.10)	0.92 (0.81 to 1.06)	—	—	500; 800
Nakamura 201162	51/773	Nil	4	0.64 (0.29 to 1.41)	0.81 (0.39 to 1.69)	0.73 (0.32 to 1.64)	1	—	410; 544; 722;
Warensjo 201163	14 738/61 433	Inverse	5	1.18 (1.12 to 1.25)	1.04 (0.98 to 1.10)	1	1.02 (0.96 to 1.07)	1.00 (0.95 to 1.06)	751; 882; 996; 1137

Nil=no association between calcium intake and risk of fracture; inverse=higher calcium intake associated with decreased risk of fracture or lower calcium intake associated with higher risk of fracture; SD=standard deviation; M=male; F=female; Q=quartile (values not reported in paper); NS=not stated.

*Number of participants with fracture.

†Hazard ratio or relative risk (95% CI) or mean (SD) mg/d.

‡For example, cut point of 250 indicates 2 groups of <250 and ≥250 mg/d; cut points of 400; 800; and 1200 indicate 4 groups <400; 400-799; 800-1199; ≥1200 mg/d.

§No numerical data were reported, but it was stated that there was no association between calcium intake and risk of fracture.

Table 4.

Association between dietary calcium intake and risk of hip fracture in cohort studies

Study	Fracture*/participans	Association	Groups	Risk or daily calcium intake†		Group 3 risk	Group 4 risk	Group 5 risk	Cut points between each group (mg/d)‡ or unit for pooled risk
				Group 1	Group 2
Dargent-Molina 200238	NS/1588	Nil	—	NR§	NR§	NR§	NR§	NR§	—
Munger 199933	44/32 050	Nil	2	842 (322)	778 (267)	—	—	—	No fracture; fracture
Nguyen 200752	24/924	Nil	2	583 (284)	489 (367)	—	—	—	No fracture; fracture
Holbrook 198823	33/957	Inverse	—	0.6	—	—	—	—	Per 198 mg/1000 kcal/d increase
Meier 200855	27/609	Nil	—	1.32 (0.81 to 2.16)	—	—	—	—	Per SD (322mg/d) decrease
Benetou 201161	275/29 122	Nil	—	1.02 (0.91 to 1.13)	—	—	—	—	Per quintile increase
Diez-Perez 200748	49/5146	Inverse	2	2.52 (1.07 to 5.92)	1	—	—	—	250
Wickham 198924	44/1419	Nil	3	0.7 (0.1 to 3.9)	0.9 (0.2 to 4.3)	1	—	—	641; 901
Paganini-Hill (F) 199125	332/8600	Nil	3	1	1.02 (0.77 to 1.33)	1.11 (0.85 to 1.44)	—	—	280; 500;
Paganini-Hill (M) 199125	86/5049	Nil	3	1	0.87 (0.50 to 1.51)	1.25 (0.75 to 2.08)	—	—	280; 500;
Looker 199326	122/2226	Nil	4	1	0.86 (0.5 to 1.5)	1.03 (0.6 to 1.7)	0.72 (0.4 to 1.3)	—	300; 501; 776
Cumming 199728	332/9704	Nil	4	1	1.0 (0.7 to 1.3)	0.8 (0.5 to 1.2)	0.9 (0.5 to 1.6)	—	400; 800; 1200
Meyer (F) 199730	150/19 752	Nil	4	1	0.86 (0.55 to 1.35)	0.87 (0.56 to 1.35)	0.67 (0.42 to 1.08)	—	435; 569; 718
Meyer (M) 199730	55/20 035	Nil	4	1	0.96 (0.46 to 2.00)	1.08 (0.53 to 2.21)	0.64 (0.28 to 1.45)	—	623; 823; 1030
Owusu 199731	56/43 063	Nil	5	1	1.47 (0.65 to 3.28)	1.14 (0.50 to 2.64)	0.86 (0.35 to 2.13)	0.64 (0.24 to 1.69)	134; 248; 364; 591
Mussolino 199832	71/2879	Nil	4	1	0.83 (0.42 to 1.63)	0.76 (0.34 to 1.66)	0.76 (0.32 to 1.79)	—	417; 680; 1033
Feskanich 200340	603/72 337	Nil	5	1	1.13 (0.87 to 1.49)	1.29 (0.98 to 1.71)	1.04 (0.76 to 1.42)	1.08 (0.78 to 1.49)	500; 625; 750; 900
Nieves 200856	337/52 144	Nil	3	1	0.89 (0.61 to 1.31)	0.87 (0.63 to 1.21)	—	—	500; 800
Koh (F) 200957	692/35 241	Positive	4	1	1.16 (0.92 to 1.47)	1.36 (1.07 to 1.73)	1.45 (1.16 to 1.82)	—	259; 327; 425
Koh (M) 200957	276/27 913	Nil	4	1	1.23 (0.90 to 1.69)	0.87 (0.60 to 1.27)	1.24 (0.86 to 1.79)	—	259; 327; 425
Warensjo 201163	3871/61 433	U shaped	5	1.29 (1.17 to 1.43)	1.09 (0.98 to 1.21)	1	1.13 (1.01 to 1.26)	1.19 (1.06 to 1.32)	751; 882; 996; 1137

Nil=no association between calcium intake and risk of fracture; inverse=higher calcium intake associated with decreased risk of fracture or lower calcium intake associated with higher risk of fracture; SD=standard deviation; M=male; F=female; Q=quartile (values not reported in paper); NS=not stated.

*Number of participants with hip fracture.

†Hazard ratio or relative risk (95% confidence interval) or mean (SD) mg/d.

‡For example, cut point of 250 indicates 2 groups of <250 and ≥250 mg/d; cut points of 400; 800; and 1200 indicate 4 groups <400; 400-799; 800-1199; ≥1200 mg/d.

§ No numerical data were reported, but it was stated that there was no association between calcium intake and risk of fracture.

Table 5.

 Association between dietary calcium intake and risk of vertebral or forearm fracture in cohort studies

Study	Fracture*/participants	Association	Groups	Risk or daily calcium intake†		Group 3 risk	Group 4 risk	Group 5 risk	Cut points between each group (mg/d)‡ or unit for pooled risk
				Group 1	Group 2
Vertebral fracture
van def Klift (M) 200444	44/1377	Nil	2	1162 (399)	1148 (341)	—	—	—	No fracture; fracture
van def Klift (F) 200444	113/1624	Nil	2	1108 (333)	1089 (305)	—	—	—	No fracture; fracture
Papaioannou 200546	34/5143	Nil	2	1133 (681)	1274 (823)	—	—	—	No fracture; fracture
Nguyen 200752	76/924	Nil	2	583 (284)	559 (292)	—	—	—	No fracture; fracture
Meier 200855	55/609	Nil	—	1.08 (0.77 to 1.51)	—	—	—	—	Per SD (322 mg/d) decrease
Cumming 199728	389/9704	Nil	4	1	1.2 (0.9 to 1.6)	1.2 (0.8 to 1.8)	1.5 (0.9 to 2.5)	—	400; 800; 1200
Nakamura (M) 200958	NS/34 759	Nil	4	1.46 (0.82 to 2.61)	1.20 (0.68 to 2.09)	1.68 (1.02 to 2.74)	1	—	Q1; Q2; Q3; Q4
Nakamura (F) 200958	NS/41 120	Inverse	4	1.92 (1.28 to 2.88)	1.30 (0.86 to 1.98)	1.50 (0.99 to 2.26)	1	—	Q1; Q2; Q3; Q4
Forearm fracture
Nguyen (M) 200137	21/739	Inverse	—	1.98 (1.00 to 3.58)	—	—	—	—	Per 300 mg/d decrease
Nguyen (F) 200137	100/1105	Nil	—	1.01 (0.82 to 1.25)	—	—	—	—	Per 300 mg/d decrease
Diez-Perez 200748	104/5146	Nil	2	1.52 (0.74 to 3.12)	1	—	—	—	250
Cumming 199728	467/9704	Nil	4	1	1.0 (0.8 to 1.3)	1.4 (1.0 to 2.0)	0.9 (0.6 to 1.6)	—	400; 800; 1200
Owusu 199731	201/43 063	Nil	5	1	1.01 (0.66 to 1.55)	0.75 (0.47 to 1.20)	1.08 (0.70 to 1.68)	1.11 (0.71-1.75)	134; 248; 364; 591
Honkanen 200034	368/11 798	Inverse	4	1	0.7 (0.53 to 0.92)	0.61 (0.43 to 0.85)	0.48 (0.25 to 0.92)	—	500; 1000; 1500
Kato 200036	193/6250	Nil	5	1	1.11 (0.7 to 1.7)	0.93 (0.6 to 1.5)	1.12 (0.7 to 1.7)	0.78 (0.5-1.3)	569; 689; 799; 949

Nil=no association between calcium intake and risk of fracture; inverse=higher calcium intake associated with decreased risk of fracture or lower calcium intake associated with higher risk of fracture; SD=standard deviation; M=male; F=female; Q=quartile (values not reported in paper); NS=not stated.

*Number of participants with vertebral fracture.

†Hazard ratio or relative risk (95% confidence interval) or mean (SD) mg/d.

‡For example, cut point of 250 indicates 2 groups of <250 and ≥250 mg/d; cut points of 400; 800; and 1200 indicate 4 groups <400; 400-799; 800-1199; ≥1200 mg/d.

Calcium supplements

Randomised controlled trials

We identified 26 randomised controlled trials (n=69 107 participants) of calcium supplements that reported fracture outcomes.^{9 13 14 15 16 18 21 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94} Table 1 and table A in appendix 3 shows the study design and selected baseline characteristics of the randomised controlled trials. Fourteen studied calcium monotherapy, eight studied CaD, and four were multi-arm or factorial studies of both agents. Twenty trials used a dose of ≥1000 mg/day of calcium; 21 were in individuals living in the community; 15 had a duration of three or more years; in 16, the mean age of participants at baseline was ≥70; in 24 most participants were women; and in 10 of 19 randomised controlled trials that reported baseline dietary calcium intake, the level was <800 mg/day. Table B in appendix 3 shows our assessment of the risk of bias: three trials were assessed as low risk of bias, one as high risk of bias for hip fracture but low risk for other outcomes, nine as moderate risk of bias, and 13 as high risk of bias.

Figures1-4 show that calcium supplements reduced the risk of total fracture (20 studies, n=58 573; relative risk 0.89, 95% confidence interval 0.81 to 0.96; P=0.004; fig 1) and vertebral fracture (12 studies, n=48 967; 0.86, 0.74 to 1.00; P=0.04; fig 3) but not hip fracture (13 studies, n=56 648; 0.95, 0.76 to 1.18; P=0.63; fig 2) or forearm fracture (eight studies, n=51 775; 0.96, 0.85 to 1.09; P=0.54; fig 4). With Egger’s regression model and visual inspection of funnel plots, data seemed biased toward reduction in risk with calcium supplements for total (P=0.006), vertebral (P=0.002), and forearm fracture (P=0.06), raising the possibility of publication bias. Furthermore, the pooled effect estimates for all fracture outcomes seemed related to the risk of bias. Figures 1, 3, and 4 and table 2 show that the effect size was smallest and not significant for total, forearm, and vertebral fracture in the subgroup of studies at lowest risk of bias, and that results also differed by risk of bias for hip fracture (fig 2).

Fig 1 Random effects models of effect of calcium supplements on risk of total fracture. Trials with no events are not included in meta-analyses

Fig 3 Random effects models of effect of calcium supplements on risk of vertebral fracture. Trials with no events are not included in meta-analyses

Fig 2 Random effects models of effect of calcium supplements on risk of hip fracture. Trials with no events are not included in meta-analyses

Fig 4 Random effects models of effect of calcium supplements on risk of forearm hip fracture. Trials with no events are not included in meta-analyses

Table 6 shows the results of the prespecified subgroup analyses. There was no evidence of a difference in the results between the subgroups of calcium monotherapy or CaD, or between the subgroups based on residential status and baseline dietary calcium intake for total, vertebral or forearm fracture. Fig 1 and table 6 show that there were differences in all subgroup analyses for hip fracture, which were largely because of the results of a single large trial of CaD with a 23% reduction in hip fractures that was carried out in women living in residential care with a low dietary calcium intake and low vitamin D concentrations.^{15 16} In all four subgroup analyses (risk of bias, calcium or CaD, residential status, and baseline dietary calcium intake), whichever subgroup this study was in had markedly different results to the other subgroup, in which there were non-significant increases in risk of hip fracture.

Table 6.

 Subgroup analyses by fracture site in randomised controlled trials of calcium supplements

Subgroup	Total				Hip				Vertebral				Forearm
	No of studies	RR (95% CI)	P value*		No of studies	RR (95% CI)	P value*		No of studies	RR (95% CI)	P value*		No of studies	RR (95% CI)	P value*
Risk of bias:
Low	4	0.96 (0.91 to 1.01)	0.03		3	1.68 (0.84 to 3.36)	0.05		4	0.89 (0.75 to 1.06)	0.37		4	0.98 (0.83 to 1.16)	0.24
Moderate/high	16	0.80 (0.69 to 0.93)			10	0.82 (0.71 to 0.94)			8	0.76 (0.56 to 1.03)			4	0.77 (0.54 to 1.11)
Treatment:
Calcium monotherapy	13	0.85 (0.73 to 0.98)	0.25		7	1.51 (0.93 to 2.48)	0.02		10	0.80 (0.64 to 1.01)	0.47		4	0.92 (0.69 to 1.23)	0.70
Co-administered CaD†	10	0.92 (0.86 to 0.99)			9	0.84 (0.74 to 0.96)			3	0.90 (0.74 to 1.09)			5	0.98 (0.86 to 1.13)
Residential status:
Community	17	0.88 (0.80 to 0.98)	0.63		11	1.10 (0.83 to 1.46)	0.03		10	0.86 (0.75 to 1.00)	0.30		8	0.96 (0.85 to 1.09)	—
Residential care	3	0.85 (0.74 to 0.98)			2	0.75 (0.62 to 0.92)			1	0.35 (0.06 to 1.93)			0	—
Calcium intake:
<800 mg/d	7	0.83 (0.73 to 0.95)	0.78		4	0.75 (0.61 to 0.91)	0.05		6	0.77 (0.55 to 1.07)	0.45		2	0.50 (0.11 to 2.18)	0.42
>800 mg/d	9	0.86 (0.74 to 0.99)			6	1.32 (0.77 to 2.26)			4	0.89 (0.75 to 1.05)			5	0.92 (0.77 to 1.09)

RR=relative risk.

*P value for interaction.

†Co-administered calcium and vitamin D.

Table 7 shows the results of the sensitivity analyses. Inclusion of two randomised controlled trials at high risk of bias^{13 14} and analysis of one cluster randomised controlled trial^{15 16} as an individually randomised trial did not alter the results. We used the result from the reanalysis of the Women’s Health Initiative restricting participants to those not using oestrogen (relative risk 1.20, 95% confidence interval 0.85 to 1.69)¹⁹ instead of the result for the entire cohort (0.88, 0.72 to 1.07).⁹ This had a modest effect, moving the results toward those of the trials at low risk of bias. We repeated our analyses excluding the influential trial with the outlying results.^{15 16} The relative risk was 0.90 (0.82 to 0.98) for total fracture and 1.02 (0.78 to 1.34) for hip fracture.

Table 7.

 Sensitivity analyses of randomised controlled trials of calcium supplements and risk of fracture

Analysis and fracture site	No of studies	Relative risk (95% CI)
Include Inkovaara 198313 and Larsen 200414*
Total fracture	22	0.90 (0.83 to 0.96)
Include Inkovaara 198313 and Larsen 200414†
Total fracture	22	0.89 (0.83 to 0.95)
Analyse Chapuy 199415 16 as individually randomised
Total fracture	20	0.88 (0.81 to 0.96)
Hip fracture	13	0.95 (0.76 to 1.18)
Restrict Jackson 20069 to women not using oestrogen19
Hip fracture-all studies	13	1.04 (0.80 to 1.34)
Hip fracture-CaD subgroup	9	0.90 (0.75 to 1.08)
Hip fracture-community dwelling	11	1.20 (0.97 to 1.48)
Hip fracture-calcium intake >800 mg/d	6	1.41 (0.92 to 2.18)
Exclude Chapuy 199415 16
Total fracture	19	0.90 (0.82 to 0.98)
Hip fracture	12	1.02 (0.78 to 1.34)

*Comparison of both environmental programme and calcium and vitamin D programme with environmental programme only.

†Comparison of any calcium and vitamin D versus no calcium and vitamin D.

Cohort studies

Table 2 and table C in appendix 3 show the study design and selected characteristics of the 11 cohort studies that reported associations between calcium supplements and fracture outcomes. Most studies reported no association between calcium use and fracture (table F in appendix 3). Of the 20 reported associations, 13 were neutral, five were positive, and two were inverse.

Discussion

There is insufficient evidence to assess the effect of increasing calcium intake in the diet from randomised controlled trials as only two small trials of dietary sources of calcium have reported fracture outcomes. Some 42 cohort studies, however, have assessed relation between dietary calcium intake, milk or dairy intake and fracture. Most analyses (≥75%) found no associations, and where there were relations reported, most relative risks were between 0.5 and 2.0, which are considered weak associations in observational studies.⁷² The recommended dietary calcium intake for older adults is 1200 mg/day.¹ Most studies, however, did not report reduced risk of fracture in individuals with this level of calcium intake compared with lower intakes. Thus, observational research does not support a hypothesis of dietary “calcium deficiency” in which there are reductions in fracture risk from increasing dietary calcium intake across the range of intakes (<300->1200 mg/day) in studies in this review.

In 26 randomised controlled trials, calcium supplements reduced the risk of total fracture by 11% and vertebral fracture by 14% but had no effect on forearm or hip fracture. The results, however, were not consistent. There was no effect of calcium supplements on any fracture outcome in the largest trials at lowest risk of bias. Only one trial in frail elderly women in residential care with low dietary calcium intake and vitamin D concentrations showed significant reductions in fracture risk. Funnel plots were also asymmetric with more small-moderate sized studies than expected reporting risk reductions in total, vertebral, and forearm fracture with calcium supplements, raising the possibility of publication bias. Results from randomised controlled trials of calcium monotherapy were similar to those with CaD, with no evidence of additional benefit of vitamin D on risk. These results suggest that widespread untargeted use of calcium supplements in older individuals is unlikely to result in meaningful reductions in incidence of fracture.

Strengths and limitations

The strength of this review is its comprehensive nature, including both randomised controlled trials and observational studies, and assessment of four fracture outcomes: total, hip, vertebral, and forearm. An important limitation is the difficulty of identifying all cohort studies that reported relations between calcium intake and fracture risk. Many of the reports of cohort studies included in our review were not identified by the database searches because the relation between calcium intake and fracture was not the focus of the report, with the results reported in the text or tables of the article but not the abstract. This was more likely to occur when there was no association between calcium intake and fracture, so the current analysis might overestimate the relation between diet and fracture. We did not perform a quality assessment of the cohort studies, although we included only those studies with a prospective cohort design, considered to be the strongest observational methods.

Generally, observational studies are considered to have a higher risk of bias than large well conducted randomised controlled trials. Tools for assessing quality of observational studies are available, but they often focus on reporting of studies rather than topic specific issues, such as methods of assessment of dietary calcium intake, methods of fracture assessment, categorisation of dietary calcium intake in statistical models, and inclusion of covariates in those models. Such factors are likely to be extremely influential in the results of the cohort studies but are either not easily assessed or not able to be assessed. If we limited our results to cohort studies with more than 100 fractures in which fracture risk by baseline dietary calcium intake was reported for at least three groups, most studies reported no association between baseline dietary calcium and fracture (5/7 for total fracture, 6/8 for hip fracture, 1/1 for vertebral fracture, and 3/4 for forearm fracture). The results from these large studies are similar to the overall results, and each study has adequate power to detect clinically relevant effect sizes.

We did not perform meta-regression analyses because there were few studies that reported sufficient data for such an analysis. Individual patient data analyses might be of value in further exploring the relation between baseline calcium intake and fracture risk. Other important limitations include that many of the randomised controlled trials were of short duration and did not have fracture as the primary endpoint. The trials were generally carried out in healthy populations or those at risk of osteoporosis, and so the findings might not apply to other population groups.

Results in context

Overall, there is little evidence currently to suggest an association between calcium intake and fracture risk or that increasing calcium intake through dietary sources will alter risk. Although calcium supplements produced some small inconsistent reductions in fractures, the doses used of 500-1600 mg/day gave an average total daily calcium intake of 1780 mg/day (range 1230-2314 mg/day). This is considerably higher than the dietary calcium intake in the highest quarter or fifth in the prospective observational studies. If calcium supplements are correcting dietary “calcium deficiency” it might be necessary to increase dietary calcium intake to about 1800 mg/day to achieve equivalent effects to calcium supplements. Dietary manipulation to increase calcium intake by ≥1000 mg/day or to achieve total daily intakes of this size is unlikely to be sustainable.

The pooled analyses of all randomised controlled trials showed reductions in risk with calcium supplements for all fractures (by 11%) and vertebral fractures (by 14%). The incidence of vertebral fracture and any fracture in the control groups in our pooled analyses was 1.5% and 12%, respectively, after a participant weighted average duration of follow-up of 6.2 and 5.5 years, respectively. With these values and the observed risk reductions from the meta-analyses, the number needed to treat (NNT) with calcium to prevent one vertebral fracture is 489 for 6.2 years and to prevent one fracture at any site is 77 for 5.5 years. These benefits are unlikely to be attractive for an individual and would be even smaller for individuals at lower risk of fracture, who are often advised to take calcium supplements, or if relative risks from the randomised controlled trials at lowest risk of bias were used in the calculations. There was no benefit from calcium supplements for hip fractures, which have the greatest clinical consequences.

Small benefits might be useful at a population level if calcium supplements were used widely, well tolerated, and safe. Persistence with calcium supplements in clinical trials is low, however, at about 40-60%,^{9 87 89 90} and in one recent randomised controlled trial, there were 24 more women admitted to hospital for acute gastrointestinal symptoms in the calcium group than the placebo group, and 16 fewer women with a fracture.^{10 89} In another randomised controlled trial, there were 68 more women with a kidney stone in the CaD group and 56 fewer women with a fracture.⁹ In our randomised controlled trial and subsequent meta-analyses, the cardiovascular risks of calcium were similar to^{6 7} or exceeded⁸ the benefits of calcium on fracture prevention. In addition, 10-20% of people experience gastrointestinal side effects such as constipation, which cause a considerable number to stop taking the supplements. Thus, because of the small benefits of use and unfavourable risk:benefit profile, calcium supplements should not be recommended for fracture prevention either at an individual or population level.

An important point emerging from our analyses is the impact of one randomised controlled trial¹⁵ on previous meta-analyses. Chapuy and colleagues studied frail elderly French women (mean age 84) in residential care with low baseline dietary calcium intake (513 mg/day) and low baseline vitamin D concentrations (mean about 20 nmol/L in modern assays⁸³). Of these participants, 16% died within 18 months of randomisation. Co-administered CaD (1200 mg/day, 800 IU/day) reduced hip fractures by 23% and all fractures by 17% at three years.¹⁶ These results are in contrast to all six other large randomised controlled trials (n>1000) of calcium or CaD, none of which reported significant reductions in total or hip fracture risk (fig 1). Based on the average vitamin D concentrations in the Chapuy study (about 20 nmol/L), it is possible that many participants had unrecognised osteomalacia, the treatment of which might have led to the benefits observed. Therefore, the benefits of CaD in this study should not be expected to be reproduced in cohorts with higher vitamin D concentrations. In our subgroup analyses, whichever subgroup the Chapuy study was in had reductions in risk of hip fracture that were markedly different to the other subgroup (table 7). The influence of this single trial is also a feature of previous meta-analyses that concluded that high dose but not low dose vitamin D prevents fractures,⁹⁵ co-administered CaD but not vitamin D prevents fractures,⁹⁶ and CaD administered to people living in residential care but not in the community prevents fractures.¹⁷ Our analyses highlight that the results from this study of a frail population with marked vitamin D deficiency are so different to those from other large randomised controlled trials and so influential in any pooled analysis that they should probably not be combined in pooled analyses with studies that enrolled different patient groups. Furthermore, recommendation of use of calcium and vitamin D supplements generally for older adults to prevent fracture based on results heavily influenced by this study of frail women in residential care is inappropriate.

On the basis of the trial data summarised here, we do not think further randomised controlled trials of calcium supplements with or without vitamin D with fracture as the endpoint in the general population are needed. In the population of frail elderly women with low dietary calcium intake and low vitamin D concentrations studied by Chapuy and colleagues,¹⁵ co-administered CaD was clearly beneficial. Important adverse events such as cardiovascular events, however, were not reported, and it remains uncertain whether the benefit was due to vitamin D or calcium or both. Trials to compare the effects of CaD with vitamin D monotherapy in this population group and also to assess whether reduction in fracture risk with anti-resorptive agents requires co-administration of either vitamin D or CaD would be valuable. Surrogate endpoints, such as bone mineral density, allow biological effects of agents to be assessed in much smaller randomised controlled trials. The effects of increasing dietary calcium intake on bone mineral density in the general population and in specific subgroups considered most likely to benefit from this intervention should be examined before large trials with fracture as an endpoint are considered, though it should not be assumed that short term changes in bone density will be sustained or translate into fracture prevention.⁹⁷

Conclusions

In summary, our analyses indicate that dietary calcium intake is not associated with risk of fracture, and there is no evidence currently that increasing dietary calcium intake prevents fractures. Calcium supplements have small inconsistent benefits on fracture reduction but probably have an unfavourable risk:benefit profile. There was no risk reduction in fracture at any site in pooled analyses of the randomised controlled trials of calcium supplements at lowest risk of bias, and there was evidence of publication bias in small-moderate sized trials. Collectively, these results suggest that clinicians, advocacy organisations, and health policymakers should not recommend increasing calcium intake for fracture prevention, either with calcium supplements or through dietary sources.

What is already known on this topic

• Older men and women are recommended to take at least 1000-1200 mg/day of calcium to prevent fractures, and many people take calcium supplements to meet these recommendations

• Recent trials have raised concerns about the safety of calcium supplements

• Experts have therefore encouraged older people to increase their calcium intake through food rather than by taking supplements, but it is not known whether increasing dietary calcium intake prevents fractures

What this study adds

• Dietary calcium intake is not associated with risk of fracture, and there is currently no evidence that increasing calcium intake prevents fractures

• Calcium supplements have small inconsistent benefits on fracture prevention

• Increasing calcium intake, through calcium supplements or dietary sources, should not be recommended for fracture prevention

Cite this as: BMJ 2015;351:h4580

Web Extra. Extra material supplied by the author

Appendix 1: Literature searches and superseded reports of cohort studies

Appendix 2: Flow of articles

Appendix 3: Supplementary tables A-F