Number of drugs most frequently found to be independent risk factors for serious adverse reactions: a systematic literature review

Abstract

In order to reduce the numbers of medication errors (MEs) that cause adverse reactions (ARs) many authors have tried to identify patient-related risk factors. However, the evidence remains controversial. The aim was to review systematically the evidence on the relationship between patient-related risk factors and the risk of serious ARs. A systematic search in Pubmed, Embase, Cochrane Systematic Reviews, Psychinfo and SweMed+ was performed. Included full text articles were hand searched for further references. Peer reviewed papers including adults from primary and secondary healthcare were included if they clearly defined seriousness of the ARs and described correlations to risk factors by statistical analysis. A total of 28 studies were identified including 85 212 patients with 3385 serious ARs, resulting in an overall frequency of serious ARs in 4% of patients. Age, gender and number of drugs were by far the most frequently investigated risk factors. The total number of drugs was the most consistent correlated risk factor found in both univariate and multivariate analyses. The number of drugs is the most frequently documented independent patient-related risk factor for serious ARs in both the general adult population as well as in the elderly. The existing evidence is however conflicting due to heterogeneity of populations and study methods. The knowledge of patient-related risk factors for experiencing ARs could be used for electronic risk stratification of patients and thereby allocation of healthcare resources to high risk patients.

Introduction

An adverse reaction (AR) is a clinical response and is a response to a medicinal product which is noxious and unintended and the term includes medication errors (MEs), non-compliance and intentional overdose (Table1) 1. The definition of an AR was updated by the World Health Organization (WHO) and adopted by the European Authorities in 2012. In order to accumulate knowledge about drug safety and communication across countries a consistent terminology is essential. In order to facilitate and support a consistency in terminology this paper will use the new official term 1,2.

Table 1.

The characteristics of the included studies that investigated risk factors of ARs

Adverse reaction
	An adverse reaction is a response to a medicinal product which is noxious and unintended. This includes adverse reactions which arise from:
	• Use of a medicinal product within the terms of the marketing authorization
	• Use outside the terms of the marketing authorization, including overdose, misuse, abuse and medication errors
	• Occupational exposure 1,2

The overall rate of ARs in hospitalized patients reported in meta-analyses varies between 6.1–9.2 % 31–37. An estimate of prolonged hospitalization of 1.2–8.5 days per AR per patient has been suggested 35,38,39. Preventing ARs causing these stays is therefore important for both patients and the healthcare systems.

In order to reduce the numbers of ARs many authors have searched for a relationship between patient-related risk factors and ARs 40–42. However the evidence remains controversial, probably due to differences in definitions and the use of statistical methods. Some authors have reported that more than 50% of patients admitted with an AR were females and likewise that patients over 65 years of age accounted for more than 50% of all hospitalizations caused by ARs 43,44. However, adjusting for confounding factors can ameliorate both age and gender differences. In a meta-analysis from 2007 11 studies estimated risk factors for ARs of which the most important were polypharmacy, female gender, use of drugs with a narrow therapeutic range, age > 65 years and reduced renal function 35. However, other studies did not verify a correlation to some of these risk factors but revealed correlation to other risk factors such as length of hospitalization, ischaemic heart disease, depression and cognitive problems 11,13,42,45. Reviews of ARs causing hospitalizations are available as well as reviews from hospitalizations 31–37. To the best of our knowledge reviews of risk factors for serious ARs are currently not available.

If patient-related risk factors for serious ARs are known it might be possible to intervene prior to their occurrence, provided that some degree of certainty of the risk factors has been established. Therefore, the aim of this study was to perform a systematic review in order to establish the most important patient-related risk factors for serious ARs.

Methods

Search strategy

The following databases were searched in October 2011: Pubmed, Embase, Cochrane Systematic Reviews, Psychinfo and SweMed+ using the terms (medication errors OR pharmaceutical preparations/adverse effects OR drug toxicity OR drug therapy/adverse effects) AND risk factors and epidemiologic studies AND drug therapy/adverse effects (see appendix for a full search strategy). The reference lists of all included full text articles were hand searched for further references. An updated search was performed in December 2014 revealing no new studies.

Inclusion criteria

Peer reviewed papers were included that reported on either prospective or retrospective detection of ARs that were serious and clearly defined seriousness according to WHO and the European Medical Authorities as any untoward medical occurrence that at any dose results in death, is life-threatening, requires inpatient hospitalization or prolongation of existing hospitalization, results in persistent or significant disability or incapacity and is a congenital anomaly/birth defect 1,2,46. When seriousness could be interpreted from the article e.g. hospitalizations due to ARs the studies were also included. Further to this, papers presenting statistical correlation methods in order to distinguish serious ARs separately from non-serious, to patient-related risk factors, were included. Papers in the following languages were included: Danish, Swedish, Norwegian, German and English. Finally, papers reporting older definitions of ARs, which are ADRs and ADEs, were also included, provided that seriousness according to the WHO could be evaluated.

Exclusion criteria

Papers were excluded if they 1) investigated ARs that could potentially harm patients, but had not resulted in harm according to the above mentioned seriousness criteria, 2) included non-compliance and intentional drug overdose unless they were handled separately in the statistical analysis and thereby be excluded from the analysis and 3) included children at age < 18 years if mentioned in the paper. Children were excluded as drug use is different from adult use. One major difference is the off-label use which might result in different safety problems from those seen in adults.

Data extraction

All citations were uploaded to Refworks, an online reference management tool 47. First, all duplicates were removed. Second, each title and abstract was screened to determine whether the full text research paper should be retrieved or whether it was evident that it did not fulfill the inclusion criteria.

The following data were extracted from the full text papers that met the inclusion criteria: reference identification number, first author, title and year, country of origin, study design, number of patients, patient category, age group, type of ARs investigated, statistical methods used, correlation investigated and results. Authors were not contacted for further information.

Risk factors of individual drugs and their correlation with ARs are not presented in this review.

Results

After removing duplicates 2859 references were identified (Figure1). Titles and abstracts were screened and full text was retrieved for 144 references. Of these, 122 papers were excluded based on the exclusion criteria. From the reference lists a further six references were identified and included.

Figure 1.

Flow chart of the literature search

The 28 full text papers are shown in Table2. Thirteen countries were represented and they were mainly from the western part of the world. Study designs were primarily observational, and most of them were prospective. One case-control study was included and one study included psychiatric patients. Two studies had been published or partly published twice. Thus the total number of studies included was 26. Patients included in the studies varied from being all ages (19 studies) to include only the elderly (seven studies), most frequently past the age of 65 years. All studies investigated risk factors in hospital settings except for one study that was performed in a nursing home. Most studies (20 in total) investigated hospitalizations caused by ARs. Four studies investigated ARs during hospitalization, one study investigated both and finally one study investigated ARs causing transfer to an intensive care unit.

Definition of terms

Reference	Year	Country	Patients (5)	Age (years)	Study design	Seriousness criteria
ARs during hospitalization
Cabello et al. 3	2009	Spain	289	>18	Retrospective analysis	Fatality
Buajordet et al.	2001	Norway	732	All	Prospective observational	Fatality
Ebbesen et al. 4,5*
Fattinger et al. 6	2000	Switzerland	3624	?	Prospective observational	Hospitalization
Darchy et al. 7	1999	France	623	?	Retrospective study	Life-threatening (ICU)
ARs causing and during hospitalization
Bordet et al. 8	2001	France	16916	Adults	Prospective observational	Hospitalization
Moore et al. 9	1998	France	329	?	Prospective observational	Hospitalization, fatality, life-threatening, prolongation of hospitalization
ARs causing hospitalization
Santamaria-Pablos et al. 10	2009	Spain	163	>18	Prospective cohort	Hospitalization
Wawruch et al. 11	2009	Slovakia	600	≥65	Retrospective cohort	Hospitalization
Helldén et al. 12	2009	Sweden	153	≥65	Retrospective observational	Hospitalization
Leendertse et al. 13	2008	Holland	12793	≥18	Prospective case-control	Hospitalization
Mjörndal et al. 14	2002	Sweden	681	?	Prospective observational	Hospitalization
Onder et al. 15	2002	Italy	28411	All	Prospective observational	Hospitalization
Malhotra et al. 16	2001	India	578	≥65	Prospective observational	Hospitalization
Mannesse et al. 17	2000	Holland	106	≥70	Prospective observational	Hospitalization
Pouyanne et al. 18	2000	France	3137	?	Prospective cross-sectional	Hospitalization
Cooper et al. 19	1999	USA	332	Nursing home	Prospective observational	Hospitalization
Hallas et al. 20	1992	Denmark	1999	?	Prospective observational	Hospitalization
Hallas et al. 21	1991	Denmark	294	Geriatric patients	Prospective observational	Hospitalization
Col et al. 22	1990	USA	315	≥65	Prospective observational	Hospitalization
Hallas et al. 23	1990	Denmark	366	<36–85	Prospective observational	Hospitalization
Colt et al. 24	1989	USA	244	All	Retrospective analysis	Hospitalization
Davidsen et al. 25	1988	Denmark	426	?	Prospective observational	Hospitalization
Hermesh et al. 26	1985	Israel	321	≥18	Prospective follow-up	Hospitalization
Bergman et al. 27	1981	Sweden	285	16–97	Prospective observational	Hospitalization
Levy et al.	1980	Israel + Germany	I: 2499	>20–81	Prospective observational	Hospitalization
Levy et al. 28,29*	1979		G: 2933
Caranasos et al. 30	1974	USA	6063	11–100	Prospective observational	Hospitalization

^*The study has been published or partly published twice. The paper mentioned first was included. ICU, intensive care unit.

A total of 85 212 patients were included in the 26 studies, with a total of 3385 serious ARs resulting in an overall frequency of 4% of patients experiencing an AR. The frequency of serious ARs varied from 0.5–23.6 % of patients (Tables3 and 4) with the smaller studies finding larger frequencies (Figure2). In the studies exclusively investigating the elderly population the frequency of serious ARs was 11.9 %. However, they only included 2387/85 662 (2.8 %) of patients. The most frequently investigated risk factors were gender, age, co-morbidity, number of drugs and impaired renal function and are presented in Tables3 and 4. Some of the less frequent risk factors identified that are not presented here were ischaemic heart disease, heart failure, depression, cognitive problems, dependent living situation, length of hospital stay, previous history of AR, geriatric ward, general internal medicine ward and smaller hospital (not university), cardiovascular complications and coagulation disorders 8,9,11,13,29,48.

Table 3.

Risk factors correlated with adverse reactions in adult patients

Reference	Demography					Correlation with risk factors
	Patients (n)	Serious ARs or hospitalizations (n)	ARs (%)	Gender		Age		Number of drugs		Co-morbidities		Impaired renal function
ARs during and/or causing hospitalization:
Bordet et al. 8	16 916	86	0.5			O
Fattinger et al. 6	3624	144	4	F		No		No
Cabello et al. 3	289	17	5.9	No		No		F		No
Darchy et al. 7	623	41	6.6	F		O		Yes
Moore et al. 9	329	31	9.4	F		O		Yes
Buajordet et al. 4	732	133	18.2	No		M		Yes		Yes
ARs causing hospitalization:
Caranasos et al. 30	6063	177	2.9	F		O
Pouyanne et al. 18	3137	100	3.2	F		O
Onder et al. 15	28 411	964	3.4			O				Yes
Hallas et al. 23	366	15	4.1	No		O		No
Levy et al. 29	5432	270	5	F		No		Yes				Yes
Leendertse et al. 13	12 793	714	5.6	No		O		Yes		Yes		Yes
Hermesh et al. 26	321	24	7.5			O
Hallas et al. 20	1999	157	7.9	F		O		Yes
Colt et al. 24	244	23	9.4	No				Yes		No
Bergman et al. 27	285	31	10.9	F		No		Yes
Davidsen et al. 25	426	49	11.5					Yes
Mjörndal et al. 14	681	99	14.5	No		No		Yes
Santamaria-Pablos et al. 10	163	27	16.6	No		Y		No		No
Mean	4360	163	3.7
	Risk factor (yes/no)			No	Yes	No	Yes	No	Yes	No	Yes	No	Yes
	Univariate analysis (number of studies)			7	8 (F)	5	10 (O)	3	10	3	3	0	2
							1 (M)		1 (F)
				1 (Y)				1
	Multivariate analysis (number of studies)			2	1 (F)	3	1 (O)	1	3
							1 (Y)

F, Female; M, Male; O, correlation with older patients; Y, correlation with younger patients.

Table 4.

Risk factors correlated with adverse reactions in elderly patients only

Reference	Demography				Correlation to risk factors
	Patients (n)	Age (years) (mean)	Serious ARs or hospitalizations (n)	ARs (%)	Gender	Age	Number of drugs	Co-morbidities	Impaired renal function
Malhotra et al. 16	578	72.5	39	6.7	No
Wawruch et al. 11	600	76.6	47	7.8	No	≥75	≥6	≥4	No
Hallas et al. 21	294	81	33	11.2	No	No	No
Helldén et al. 12	153	82.1	22	14.4	No	F	No		F
Col et al. 22	315	76.6	53	16.8			Yes
Cooper et al. 19	332	82	64	19.3	No	No	Yes
Mannesse et al. 17	106	78	25	23.6	No		≥3	≥5
Mean	340	78.4	40.4	11.9
				Multivariate analysis
				Reference	Gender	Age	Number of drugs	Co-morbidities	Impaired renal function
				Wawruch et al. 11
				Malhotra et al. 16			≥3
				Mannesse et al. 17			≥3	No
				Hallas et al. 21	No	No	No
				Col et al. 22			>3

Figure 2.

The relationship between study size and percentage of patients with ARs in the studies. Studies including only the elderly and ARs as cause of hospitalization. Studies including all adults and ARs as cause of hospitalization. Studies including ARs during hospitalization (all adults). AR, Adverse reaction.

The importance of study size

When analyzing studies with a study size below 2000 patients (17 studies), a total of 7815 patients were included of whom 10 % experienced an AR. Ten of these studies included all adults and seven studies only the elderly. Studies comprising more than 2000 patients showed a mean frequency of ARs of 3.2 % and none of these studies included only the elderly. Figure2 illustrates that studies including only the elderly population seem to find larger frequencies of ARs, but are of smaller size.

Risk factors in studies including all adults

Table3 shows studies with statistically significant correlations (univariate analysis) between ARs in adult patients and risk factors. The studies were ranked according to frequency of ARs. The top of the table shows the studies investigating ARs during hospitalizations and the bottom shows the studies investigating ARs causing hospitalization. The mean frequency of ARs was 3.7 % (0.5–18.2) and the mean study size was 4360 (163–28 411) patients. The types of risk factors were equally distributed among the smaller and larger studies and no differences were seen among these. Age, gender and number of drugs were most frequently investigated.

Univariate analysis showed age, number of drugs and co-morbidity to be risk factors in 10 of 17, 11 of 14 and three of six studies, respectively. Multivariate analysis was only performed in four of those studies and number of drugs was confirmed as a risk factor in three out of four. Co-morbidity and age as risk factors were not confirmed. No studies investigated impaired renal function in multivariate analysis. Considering the studies of ARs at admission and ARs during admission separately did not change the overall picture.

Risk factors in studies including only the elderly

Studies including the elderly population exclusively are shown in Table4. The definition of elderly was ≥ 65 years of age (four studies), ≥ 70 years (one study), geriatric patients (one study) or nursing home residents (one study). The mean rate of ARs was higher in these studies, namely 11.9 % (6.7–23.6), while the studies were smaller, with an average of 340 patients in each. In univariate analysis the most convincing risk factor was number of drugs with four out of six studies and co-morbidity with two out of two studies showing an association. In multivariate analysis the number of drugs was still a risk factor in three out of four studies, while co-morbidity was not a risk factor in one out of one study. Impaired renal function was found to be a risk factor in women in one of two studies and was not investigated in multivariate analysis. Finally, gender was not found to be a risk factor in univariate analysis in this population.

Discussion

We found that 85 212 patients had a total of 3385 serious ARs resulting in an overall frequency of serious ARs in 4% of patients. The primary risk factor correlated with the risk of ARs was the total number of drugs when considering studies using multivariate analysis.

Is study size of importance?

In the studies exclusively investigating the elderly population the frequency of serious ARs was 11.9 %. However, they represented only a minority of the patients. Considering studies with less than 2000 patients, there were 10 studies of all adults and seven of the elderly. This simple subdivision points in the direction of small study size rather than age being the reason for increased frequencies of ARs. The tendency of smaller studies uncovering more ARs could be related to practical issues such as scrutiny of hospital records, which is manageable when only few patients are investigated in comparison with studies of larger populations. Other possible explanations are publication bias, as studies displaying large frequencies are more interesting to publish, or it could be a question of dilution as larger populations might include less ill patients. In a meta-analysis from 2002 a total of 68 studies were analyzed with the purpose of estimating the number of patients hospitalized due to ARs 32. Eleven of those studies were also included in the present review as they examined patient-related risk factors. Beijer et al. found a total rate of 4.9 % of hospitalizations due to ARs, 4.1 % in the non-elderly population and 16.6 % in the elderly population and a tendency towards larger studies to display a lower percentage of ARs were identical.

Is the number of drugs a risk factor for ARs?

The most consistent variable in both the general population and in the elderly population was the total number of drugs. An exponential relationship between the number of concurrently used drugs and the likelihood of an AR is described in a paper by Smith et al. from 1966, and the paper concludes that the absolute drug number probably is the only truly independent variable 49. Today an exponential curve might not be entirely true, as drugs have become safer, and the morbidity from drugs has likely decreased. Instead of an exponential curve, it might rather be flat or even regress at first as morbidity due to drug treatment will lessen until a certain level, where too many drugs exceed the advantages of drug treatment.

Could knowledge of the total number of drugs as a patient-related risk factor be used in relation to drug treatment? A reduction of the number of prescribed drugs would be the natural consequence. However, knowledge about the plateau of decreased morbidity due to disease and increased morbidity due to drug treatment is necessary. The total number of drugs along with the individual drug’s risk could be incorporated into an electronic algorithm and thereby capture patients with expected higher risk of serious ARs. In this way allocation of healthcare resources could be focused on high risk patients.

Is age a risk factor for ARs?

Age does not seem to be a risk factor for ARs based on this review. Polypharmacy and polymorbidity often go hand in hand which might hamper distinction between symptoms of the disease and symptoms of the ARs. Consequently, this might lead to both over- and under-interpretation of ARs when hospital records are available and studied in detail. Studies have found that polymorbidity and polypharmacy were more frequently present in the group of patients with ARs resulting in hospitalization 11,50–53. In addition, elderly use more drugs than younger individuals 51,54–57 and an exponential increase between number of ARs and drugs taken has been described 49,58. These might serve as confounding factors.

Some studies investigated the impact of reduced cognition and found it to be negatively associated with the number of ARs, both serious and non-serious 50,53. Impaired cognition is expected to cause more mistakes in drug consumption. However, patient notification might be significantly lower and the patients might have difficulty in verbalizing or recalling symptoms resulting in underdiagnosing of ARs. Similarly, patients older than 80 years of age were significantly less likely to experience ARs than younger patients in some studies 59,60. This could be explained by the under-reporting of ARs by both healthcare professionals and patients or simply reflect the survival of the fittest.

Studies investigating the correlation with both serious and non-serious ARs found that age was not an independent risk factor for ARs. However, they revealed that the elderly patients had more severe ARs and the numbers of non-serious ARs were reduced 51,61. Carbonin et al. analyzed the data in a multivariate logistic model, which confirmed that age was not a real risk factor. The elderly population is most likely more fragile to the impact of medication and at greater need of it but, it is not evident that age per se is a risk factor for ARs.

Is female gender a risk factor for ARs?

Seven studies found females to be at higher risk, while eight studies did not find gender to be a risk factor. For the elderly population six out of six studies did not find gender to be a risk factor for ARs. Several studies found that women more often had non-serious ARs, while men had, at least, as often serious ARs. In fact, some studies demonstrated that men more frequently had serious ARs than women 41,55,62. This is in line with the perception that women seek medical attention more often than men, whereas men tend not to accept their disease, and therefore, do not resort to medical services as often 63.

Further to this, several studies have shown that women in general take a higher number of drugs than men 16,22,30,55–57 and their co-morbidity was higher 53. In the present review we chose to exclude papers that included non-compliance if impossible to distinguish ARs, that were caused or not caused by non-compliance. It might partly explain why studies including non-compliance as ARs find female gender to be a risk factor, as a higher number of drugs are found to cause less compliance 64,65.

Strengths and limitations

Karch & Lasagna investigated the difficulties in evaluating ARs by having three clinical pharmacologists independently evaluate 60 selected cases and found that they agreed on only 50% 66. Many studies have found the same problem with imprecise judgments and it is an important reason for the heterogeneity of these studies. Even though we tried to narrow the inclusion criteria it is not possible to judge if the evaluation of ARs across studies or even within studies is consistent.

A number of other factors may explain the discrepancy in results between studies, like the heterogeneity of the populations being studied with regard to age, size and co-morbidities. Another problem is the descriptive nature of most reports and thereby a lack of control group. Most studies included in this review were cross sectional studies and as such investigated the immediate relation between drug exposure and ARs. This method will provide information concerning prevalence of ARs and determine possible risk factors for ARs. However it cannot provide information about a causal relation, and the studies should be interpreted carefully. Control patients should ideally be a representative sample from those at risk of an AR that would necessitate hospital admission. The only included study aiming at this was the case-control study by Leendertse et al. 13.

Thus, it was not possible to evaluate adequately the influence of independent variables, the confounding nature of these and the possible interactions among them. Likewise the data extracted were not amenable to meta-analysis due to the heterogeneity of study methods and definitions and we did not formally assess the risk of bias.

It is possible that the reason for the marked variability in results is the lack of control for, or consideration of, many factors that may influence development of drug-related illnesses separately.

Conclusion

Based on this systematic review the number of drugs is the most frequently documented independent patient-related risk factor for serious ARs. Further, we found an overall frequency of serious ARs in 4% of patients based on 26 studies including 85 212 patients. Interestingly, the size of studies was important for the interpretation of reliable frequencies of ARs, consistent with results from review of hospitalizations.

The knowledge of patient-related risk factors for experiencing ARs could be used for electronic risk stratification of patients and thereby allocation of healthcare resources to high risk patients.

Competing Interests

All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organization for the submitted work, no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years and no other relationships or activities that could appear to have influenced the submitted work.