Percentage of Patients with Preventable Adverse Drug Reactions and Preventability of Adverse Drug Reactions – A Meta-Analysis

Background Numerous observational studies suggest that preventable adverse drug reactions are a significant burden in healthcare, but no meta-analysis using a standardised definition for adverse drug reactions exists. The aim of the study was to estimate the percentage of patients with preventable adverse drug reactions and the preventability of adverse drug reactions in adult outpatients and inpatients. Methods Studies were identified through searching Cochrane, CINAHL, EMBASE, IPA, Medline, PsycINFO and Web of Science in September 2010, and by hand searching the reference lists of identified papers. Original peer-reviewed research articles in English that defined adverse drug reactions according to WHO’s or similar definition and assessed preventability were included. Disease or treatment specific studies were excluded. Meta-analysis on the percentage of patients with preventable adverse drug reactions and the preventability of adverse drug reactions was conducted. Results Data were analysed from 16 original studies on outpatients with 48797 emergency visits or hospital admissions and from 8 studies involving 24128 inpatients. No studies in primary care were identified. Among adult outpatients, 2.0% (95% confidence interval (CI): 1.2–3.2%) had preventable adverse drug reactions and 52% (95% CI: 42–62%) of adverse drug reactions were preventable. Among inpatients, 1.6% (95% CI: 0.1–51%) had preventable adverse drug reactions and 45% (95% CI: 33–58%) of adverse drug reactions were preventable. Conclusions This meta-analysis corroborates that preventable adverse drug reactions are a significant burden to healthcare among adult outpatients. Among both outpatients and inpatients, approximately half of adverse drug reactions are preventable, demonstrating that further evidence on prevention strategies is required. The percentage of patients with preventable adverse drug reactions among inpatients and in primary care is largely unknown and should be investigated in future research.


Introduction
Drug-related adverse events, including adverse drug reactions (ADRs), have been reported to be among leading causes of morbidity and mortality [1,2]. ADRs occur in both outpatients and inpatients [2][3][4][5][6][7]. In a meta-analysis in 2002, 4.9% of hospital admissions were associated with ADRs, ranging between 0.2 and 41.3% in individual studies [4]. Further, 28.9% of the ADRrelated hospitalisations were considered preventable. Of inpatients, 10.9% is estimated to experience an ADR during hospitalisation [2]. According to the World Health Organization (WHO), costs of ADRs, including hospitalisations, surgery and lost productivity, exceed the cost of medicines in some countries [8]. As drug-related adverse events are estimated to cost USD 422-7062 per drug-related admission and USD 2284-5640 per inpatient with drug-related adverse events (2000 values) [9], significant costs may be saved if drug-related adverse events, including ADRs, were prevented.
Previous review studies have investigated preventable drugrelated adverse events and the preventability of the events [4,6,[10][11][12][13]. However, most of them summarised studies on all drugrelated adverse events, including adverse events due to noncompliance and overdose, without a clear definition for the adverse event, and used medians as summary measures [6,[11][12][13]. Two previous reviews investigated the preventability of ADRs among outpatients being admitted to hospital [4,10], but no standardised definition for ADRs was required, outpatients without hospitalisation were not studied and no meta-analysis technique was used to pool the results. Further, no previous review has investigated the preventability of ADRs among inpatients or the percentage of outpatients or inpatients with preventable ADRs (PADRs). Therefore, we applied meta-analysis techniques using a standardised definition for ADRs to estimate the percentage of adult outpatients and inpatients with PADRs, and the preventability of ADRs.

Methods
We carried out a meta-analysis on studies on PADRs in adults, adapting methods recommended by the Statement for Reporting Systematic Reviews and Meta-analyses [14], and following our study protocol. We searched seven databases; MEDLINE, Excerpta Medica Database (EMBASE), the Cochrane database of systematic reviews, Cumulative Index to Nursing & Allied Health Literature (CINAHL), International Pharmaceutical Abstract (IPA), PsycINFO and Web of Science (-September 2010) for relevant publications. References of included articles and previous reviews and meta-analysis on ADRs were reviewed to identify additional relevant articles and consider their inclusion.
The databases' search fields for titles, abstracts and index terms were searched using the databases' index terms and other commonly utilised terminology on drug-related adverse events and preventability ( Figure 1). No limits for the years of publication were set. The search was limited to English. Multiple publications of the same study were carefully reviewed [15]. The titles and abstracts were screened by one researcher (KMH). Studies were selected for inclusion from full-text articles in collaboration by two researchers (KH, KMH). An additional reviewer (SH) participated in the review process when uncertainty about eligibility criteria arose.
We included original peer-reviewed research articles published in English regardless of the study design (prospective, retrospective, cross-sectional, or interventional). To avoid inconsistent estimates, ADRs had to be defined according to the WHO: ''a response which is noxious and unintended, and which occurs at doses normally used in humans for the prophylaxis, diagnosis, or therapy of disease, or for the modification of physiological function'' [16], or according to Edwards and Aronson's similar definition [17]. Small changes in wording were overlooked if the two researchers (KH, KMH) agreed that the functional meaning remained the same. Studies representing one or two specific disease areas (inclusion of patients, setting or sampling frame) or specific treatments were excluded. Included studies had to report the percentage of patients with PADRs or the preventability of ADRs. We excluded articles summarising previous results without original assessment of ADRs and the preventability of ADRs and studies conducted in paediatrics or intensive care, or focusing only on specific types of, life threatening or fatal ADRs. We also excluded studies if ADRs were identified exclusively through spontaneous reporting or International Classification Disease (ICD-9 or 10) codes, as these two strategies are known to underestimate the rate of ADRs [7,18]. Although we set no limitations on how preventability must be defined in original studies, we required a case by case preventability assessment. Thus, we excluded studies that considered all type A ADRs, defined as dose dependent and predictable from the known pharmacological characteristics of the drug [19], as preventable without a separate preventability assessment. Studies were also excluded if the percentage of patients with PADRs and preventability of ADRs were inadequately reported.

Data extraction
To increase the reliability and efficiency of data extraction, we developed and piloted a paper data extraction form by adapting the checklist for Strengthening the Reporting of Observational Studies in Epidemiology [20]. Studies' characteristics such as the study design, data source, sampling frame, population characteristics, and definition for ADRs, as well as the number of included patients, healthcare visits, ADRs, PADRs, patients with PADRs, and healthcare visits with PADRs were extracted by two researchers. The first (KH) extracted the data, and the second (KMH) confirmed the accuracy and completeness of the extraction. Any disagreements were noted and resolved by consensus. The extracted data were based on information reported in or calculated from the included articles. Authors were not contacted for complementary information.
The number of PADRs in each study represented the sum of definitely, probably or possibly preventable cases, as reported in original studies. The percentage of patients with PADRs was calculated by dividing the reported number of healthcare visits (such as primary or emergency care visits or hospitalisations) with PADRs by the total number of healthcare visits. The percentages could be calculated only if the number of healthcare visits were interpretable in the original studies. The preventability of ADRs was calculated by dividing the number of PADRs by the total number of ADRs. If the number of PADRs, ADRs, healthcare visits or healthcare visits with PADRs was not directly reported, the two reviewers assessed whether they were interpretable based on other presented data.

Risk of bias
The two reviewers (KH, KMH) assessed independently the quality and the risk of bias in the original studies in conjunction with data extraction. To minimise inconsistent estimates, we required a standardised definition for ADRs [16,17], a case by case assessment of preventability, and more inclusive data sources than spontaneous reports or International Classification Disease (ICD-9 or 10) codes exclusively. No scoring system for quality assessment of the individual studies was applied, as no consensus on quality scoring of observational epidemiological studies exists [21].

Statistical analysis
Meta-analysis was performed using DerSimonian and Laird random effect model with the estimate of heterogeneity being taken from the inverse variance fixed effect model [22]. The summary measures for the percentage of patients with PADRs and for the preventability of ADRs were calculated separately for ADRs occurring in outpatients and for ADRs present among inpatients during hospitalisation. Studies on the elderly were analysed separately. For the percentage of patients with PADRs, we first converted the individual percentage estimates to logit to meet the normal distribution assumption, conducted the analysis on the logit and converted the final results into non-logit for interpretation.
Preventability estimates were calculated without converting them into logit in the analysis, because overall estimates using direct and logit methods differed less than one percentage when they were compared. Confidence intervals (95%) for each summary measure were calculated. STATA software version 10 was used for data analysis.
To investigate the robustness of the overall estimates, we conducted sensitivity analyses. Each analysis was conducted separately for studies with less than six months' study period and with six months' or longer study period, because studies with longer study periods may more likely include revisits of the same patients. Moreover, outlier studies whose estimates differed 20% or more from the overall estimate were omitted from each analysis. We also assessed the possibility of publication bias by evaluating funnel plots. No asymmetry was evident. Heterogeneity was explored using Cochrane's Q test of heterogeneity and I 2 statistics.

Ethics Statement
According to Swedish regulations on medical research on humans, approval by an ethical review board is not required in review studies and meta-analyses that use aggregated patient data from previous studies. As no individual patient data was used or stored for the current study, informed consents from the participants of the original studies were not required.

Results
A total of 5770 citations were found from electronic database searches and additional 59 records were identified from reference lists ( Figure 2). After removal of duplicate records, the inclusion and exclusion criteria were applied on 4220 unique citations. After title and abstract review, the inclusion and exclusion criteria were applied on 399 articles' full texts. Most full-text articles were excluded due to not assessing or reporting the percentage of patients with PADRs or the preventability of ADRs (n = 290). Excluded articles commonly focused on potential drug-related problems or other types of adverse events without a category for ADRs or their preventability. Many also lacked a denominator for calculating the percentage of patients or preventability. After applying all inclusion criteria, we finally included 22 articles in the review. Of these, 14 studies reported PADRs among outpatients exclusively, six among inpatients, and two separately for outpatients and inpatients (Tables 1,2,3).

Sensitivity analyses
The preventability of ADRs among outpatients and inpatients and the percentage of outpatients with PADRs were higher in studies with shorter study periods. When outliers were removed from the main analysis, the preventability of ADRs among outpatients was 45% (95% CI: 40-59%), lower than in the main analysis. The preventability of ADRs among inpatients was 54% (95% CI: 51-56%), higher than in the main analysis, when outliers were removed. However, all sensitivity analyses' estimates, both according to study period and after removal of outliers, were within the confidence intervals of the overall estimates of the main analyses.

Discussion
We found that 2% of adult outpatients being hospitalised or visiting emergency care experience PADRs. Approximately half of all ADRs among outpatients were preventable. As no studies in primary care were identified, the percentage of patients with PADRs and the preventability of ADRs remain unknown among outpatients without an admission or emergency visit. Among inpatients, close to half of ADRs present during hospitalisation were preventable, but the percentage of inpatients with PADRs could not be estimated precisely.

Strengths and limitations of the review
Our study is the first to estimate in a meta-analysis the percentage of patients with PADRs and the preventability of ADRs, among both outpatients and inpatients. Our search strategy was comprehensive, but laborious with a large number of citations. We included published studies in English, as research is to be shared internationally, but this may have lead to overlooking some relevant studies. Authors were not contacted for unpublished data which may have lead to excluding studies that would have fulfilled the inclusion criteria if more data had been available. Further, although an excluded full-text article could include several exclusion criteria, only one exclusion criterion per article was recorded in the order of notifying the criterion. As it was not interpretable in most original studies whether ADRs were incident or prevalent, we used the percentage of patients with PADRs as an outcome measure. To prevent heterogeneous estimates and avoid bias, we required a standardised definition for ADRs, inclusive data sources and an original preventability assessment. Yet, the included studies were heterogeneous, perhaps due to varying study designs, settings and criteria for preventability. However, our overall estimates did not differ substantially when two sensitivity analyses were performed.
We found ADRs among adult outpatients more preventable than in earlier reviews, in which the median and pooled preventability of ADRs among patients being hospitalised has been 31% and 29%, respectively [4,10]. The difference may arise from inconsistent definitions for ADRs. For providing consistent estimates, we required WHO's [16] or a similar [17] definition for ADRs. Studies included in previous reviews on outpatients accepted various definitions for ADRs [4,10], some of which found ADRs less preventable [47][48][49][50]. The criteria for preventability may also have influenced differing preventability estimates. While it was not an inclusion criterion, most studies in our metaanalysis used the criteria introduced by Hallas et al [39] or Schumock et al [40] to establish preventability. Some studies in Figure 5. Percentage of inpatients with preventable adverse drug reactions, during hospitalisation. *not provided directly in the study, interpreted from other presented data. doi:10.1371/journal.pone.0033236.g005 Figure 6. Preventability of adverse drug reactions among inpatients, during hospitalisation. *not provided directly in the study, interpreted from other presented data. doi:10.1371/journal.pone.0033236.g006 previous reviews used more narrow criteria, considering exclusively inappropriate drug selection or dose [47] and unnecessary therapy [51] preventable. Thus, preventability due to other reasons, such as lack of monitoring or prescribing a prophylactic medication for an expected ADR, may have been overlooked in previous studies resulting in lower preventability estimates. In addition, 12 out of 16 articles on outpatients in our meta-analysis were published in the 21 st century while all articles in the previous reviews on PADRs were published in the 20 th century. Even though it may not be concluded from our results that the preventability of ADRs among outpatients would have increased over time, the increasing interest in and discussion on patient safety since the late 1990s' [52] may have fostered acknowledging preventability in newer studies.
The lack of other review studies on the percentage of outpatients with PADRs or the preventability of ADRs among inpatients hinders comparison to previous evidence. In one review on all drug-related adverse events, at median 4.3% of all admissions among outpatients were considered drug-related and preventable [11]. Their higher estimate compared to ours on PADRs exclusively (2%) was expected, as they included events beyond ADRs, such as therapeutic failures, drug intoxications and misuse. Among inpatients, previous reviews have found that the median preventability of all drug-related adverse events is 35% and 46% [6,12], ranging between 19% and 90% in individual studies. These are comparable to our estimate that 45% of ADRs are preventable among inpatients experiencing ADRs during hospital stay. Compared to previous reviews, our meta-analysis provides more consistent estimates on the preventability of ADRs and the proportion of patients with PADRs as our outcome measure is more standardised.

Study implications
This meta-analysis demonstrates that PADRs are a significant cause of morbidity among outpatients and that roughly half of all ADRs among adult outpatients and inpatients may be prevented. In the included articles resulting in these preventability estimates, a common criteria for preventability was ''the drug event was due to a drug treatment procedure inconsistent with present-day knowledge of good medical practice or was clearly unrealistic, taking the known circumstances into account'' [39]. Others considered ADRs preventable when they occurred due to contraindications, inappropriate dose or monitoring, interactions, ignoring toxic serum drug concentrations or previous allergic reactions, or noncompliance [40]. ADRs occurring for these reasons need to be diminished to reduce the burden of ADRs, related costs [31,44], and unnecessary patient harm. Thus, effective intervention strategies and safety measures for preventing ADRs need to be incorporated into healthcare in system-level. As support for prevention strategies, such as medication reviews, to reduce medication-related harm is limited [53], further evidence on interventions to prevent ADRs and their implementation in healthcare is required. However, errors related to use of medications will to some extent always occur mainly due to the human imperfection in mental functioning and due to the complex nature of medical practice [54].
Our meta-analysis also highlights the lack of evidence on PADRs. Despite our thorough search strategy, we did not identify studies on PADRs occurring in primary care. Thus, our findings are likely to represent only the most serious PADRs among outpatients. Further, only two studies allowed estimating the percentage of inpatients with PADR, and the generated overall estimate was imprecise. Therefore, future research should investigate PADRs in the general population, especially among people without emergency visits or hospitalisation and among inpatients during hospital stay. As identified in previously [55,56], better consensus on defining and assessing preventability should also be reached to decrease heterogeneity between studies and enable more precise estimates in future meta-analyses.
This meta-analysis corroborates that PADRs are a significant burden to healthcare among adult outpatients. Among both outpatients and inpatients, approximately half of all ADRs are preventable. Although preventability estimates vary across studies, our results demonstrate that further evidence on prevention strategies is required. The percentage of patients with PADRs among inpatients and in primary care is largely unknown and should be investigated in future research.

Supporting Information
Checklist S1