Prevalence, nature and potential preventability of adverse drug events – a population-based medical record study of 4970 adults

Setting and participants

A random sample of 5025 adult residents (≥18 years on 31 December 2007) in the county council of Östergötland, Sweden, was drawn from the Total Population Register of Statistics Sweden. The sample included all adults with a registered address in the county, including people living in nursing homes etc. We calculated the sample size based on a conservative 8% expected prevalence and for estimating a 50% proportion among individuals with ADEs, with a maximum width of ±5% for the 95% confidence interval (CI), requiring a minimum of 384 individuals with ADEs. Medical care in all care units of the study population was reviewed retrospectively for 3 months in 2008. To account for seasonal variation, the study population was randomly divided into four groups for each quarter of the year.

Outcome measures

The primary outcome measure was an ADE, defined as ‘an injury resulting from medical intervention related to a drug’ 5, which could be associated with prescribed, non-prescribed or complementary, but not illicit drugs. Some consider ADEs to consist of non-preventable ADRs, and medication errors that are by definition preventable 33, while others consider also part of ADRs preventable 34, 35. In our study, an ADR could be preventable and was defined according to the World Health Organization 36 as ‘a response to a drug which is noxious and unintended, and which occurs at doses normally used in man for the prophylaxis, diagnosis, or therapy of disease, or for the modification of physiological function’. We excluded drug dependence (DD) from ADRs, as DD could occur in higher doses than normally used. Apart from ADRs, medication errors such as omission of a dose 37 may result in other types of injury, which could be included in the broad definition for ADEs 5, but are not detailed in most studies on ADEs. Thus, we identified additional, mutually exclusive ADE categories from the literature 6, 8-12, 25-32, 38-40. To differentiate from ADRs, we defined drug intoxications from overdose (DIs) as ‘a noxious, intended or unintended drug reaction that occurs at higher doses than normally used in man for prophylaxis, diagnosis or treatment. The intention for administrating the drug(s) may or may not be therapeutic’. DD and drug abuse (DA) were defined according to the American Psychiatric Association as ‘a maladaptive pattern of substance use leading to clinically significant impairment or distress’, which had to be manifested according to specific criteria 41. Sub-therapeutic effects of drug therapy (STEs) included absence of therapeutic response that could be linked causally either to dose that was too low, drug non-compliance, recent dose reduction/discontinuation or inadequate monitoring 40. We also included STEs due to improper drug selection or when treatment had been rational (e.g. first line treatment not effective). Morbidity due to drug-related untreated indication (UTI) occurred when a person had a clinical condition that under normal circumstances would have required pharmacological therapy but none was received. Secondary outcome measures were preventable 42 and serious 34 ADEs.

Data sources and case assessment

Data from multiple sources were linked using the personal identity number (Figure 1). Data on all individuals' dispensed drugs were retrieved from the Swedish Prescribed Drug Register (SPDR) 43, which covers all prescribed drugs dispensed in pharmacies (also, for example, low dose acetylic salicylic acid and small benzodiazepine packages), including prescription drugs for residential care. The SPDR excludes non-prescription and complementary drugs bought without a prescription, drugs administered in hospitals and emergency drugs administered in residential care. Data on healthcare encounters were retrieved from the regional patient register, Care Data Warehouse of Östergötland 44, including administrative data on all inpatient and outpatient care provided in the county in all medical specialities. In the study area, private outpatient care (including dental care) constituted 3% of all healthcare expenditure in 2008 45, practically all of which is recorded (personal communication from Lars Svensson, Östergötland County Council). Thus, the coverage of the care data is considered full. Based on the administrative care data, electronic medical records in all care units were retrieved for individuals with one or more healthcare encounters (nurse or physician, visit or telephone contact, outpatient or inpatient, specialized or not, excluding dental and paramedical care) during the 3 month study period. Healthcare providers were contacted for paper copies when electronic records were missing, including private outpatient care providers (Östergötland had no private inpatient care providers in 2008).

ADEs were detected in the medical records in a stepwise manner similar to previous studies 5, 17, using manuals and standardized, pilot-tested data extraction tools. Pharmacists extracted information on suspected ADEs from the medical records for the 3 month study period, and 9 months before and 3 months after it. For prescribed drugs, data were extracted both from the SPDR and from the medical record, as medical records include drug use in inpatient care and emergency drug use in residential care, which are excluded from the SPDR. Information on non-prescribed and complementary drugs was extracted exclusively from the medical records. Used triggers included diagnoses (e.g. urticaria) 33, drugs (e.g. warfarin) 33 and drug–drug interactions 46. A clinical pharmacologist and another pharmacist independently assessed the causality 47 between the suspected ADEs and drug therapies, detected possible additional ADEs, and assessed preventability 42, contribution to hospitalization 42 and seriousness 34. Conflicting assessments were solved by consensus. Suspected ADEs with at least possible causality 47 were considered ADEs, and ADEs with at least possible preventability 42 preventable ADEs. An ADE was considered to contribute to a hospitalization if its significance for the admission was ‘dominant’, ‘partially contributing’, or ‘less important’ 42. This assessment was also used for determining seriousness, among other seriousness criteria 34. All assessors were trained in the process. Data from the medical records were combined with register data for all individuals (Figure 1).

Aggregated data

The individuals' sociodemographic characteristics were compared with the adult general population in Sweden by retrieving aggregated data on age, gender, marital status, area of residence, country of birth, education and income from Statistics Sweden.

Data analysis

In the main analyses, the 3 month prevalences and 95% confidence intervals were calculated for different categories of ADEs and preventable ADEs. Individuals with at least one ADE were used in the numerator in the prevalence calculations. All individuals in the study population were chosen as the denominator in the prevalence calculations, because ADEs could occur without dispensed drugs (non-prescription drugs or stockpile), and UTIs, STEs and prolonged ADEs do not require current drug use. The ADE prevalences were compared by age group using χ² or Fisher's exact test, choosing cut-off ages based on changing patterns in morbidity and mortality: 18–44, 45–64 and ≥65 years 48. The prevalences of serious ADEs, and preventable serious ADEs were also calculated. For the prevalence of hospitalizations contributed by ADEs, individuals with hospitalizations during the study period were used in the denominator. For preventability and seriousness, the number of preventable and/or serious ADEs was divided by the number of all ADEs. STATA software version 11.2 was used.

Drugs associated with each ADE category were classified according to the Anatomical Therapeutic Chemical (ATC) Classification 49, including main groups (first level) and pharmacological subgroups (third level) representing >1% of the ADE category, and chemical substances (fifth level) representing ≥20% of the given pharmacological subgroups. Psycholeptics (N05) and psychoanaleptics (N06) were also classified into the fourth level drug classes. For comparison, the most common drugs dispensed to all individuals were described, from 6 months before the study period until the last day before the study period.

According to the Medical Dictionary for Regulatory Activities (MedDRA) 50, organ systems (System Organ Classes 50) and symptoms (Preferred Terms 50) representing >1% of all or preventable ADEs were presented. For ADRs and STEs, chemical substances 49 associated with the Preferred Terms at least twice were reported.

As sensitivity analyses, the ADE prevalences were calculated varying the denominator: individuals with dispensed drugs during 6 months before the study period, and individuals with ≥1 healthcare encounters during the study period. In further sensitivity analyses, the ADE prevalence was calculated without UTIs and non-preventable STEs, to mimic previous ADE definitions. This was done using different denominators: all individuals, individuals with dispensed drugs and individuals with healthcare encounters.

Ethical considerations

An ethical approval was received from the Regional Ethical Review Board in Gothenburg (644-08). According to Swedish legislation, no informed consents from the participants were required, because participation could not change the participants' healthcare or health status and the results were expected to improve care for future patients.

Study population

After excluding 55 individuals (Figure 1), the study population consisted of 4970 individuals, of which 2434 (49.0%) had healthcare encounters, in 29 departments of inpatient care in three hospitals, 110 specialized outpatient clinics, and 51 primary care units. The sociodemographic characteristics of the study population and the general population were similar (Table 1), although a larger proportion of the study population was born in Sweden.

Prevalences of persons with ADEs

ADEs were detected in 596 of the 4970 individuals, resulting in a total prevalence of 12.0% (95% CI 11.1, 12.9%) in the total general population (Table 2). As described in Table 2, ADRs and STEs were more prevalent than the other ADE categories. The prevalences differed by age group for all ADEs, ADRs and STEs, being higher in older age groups. The prevalence of preventable ADEs was 5.6% (95% CI 5.0, 6.2%), also differing by age group. The 3 month prevalence of serious ADEs in the general population was 1.2% (95% CI 0.9, 1.6%), and preventable serious ADEs 0.7% (95% CI 0.5, 1.0%). ADEs contributed to admissions for 0.6% (95% CI 0.4, 0.8%) of the general population, and for 22.1% (95% CI 15.1, 29.1%) of 136 individuals with hospitalizations during the 3 month study period. When only preventable ADEs were analyzed, they contributed to admissions for 0.4% (95% CI 0.2, 0.6%) of the general population, and for 14.0% (95% CI 8.1, 19.8%) of individuals with hospitalizations.

Numbers of events

Of all 981 ADEs, 52.4% were ADRs, 38.8% STEs, 5.3% UTIs, 2.6% DD and DA cases and 0.8% DIs. Of the 596 individuals with ADEs, 30.4% had two to three, and 6.5% four or more ADEs. Among individuals with ADRs, 30.7% had two or more ADRs, while 15.3% of individuals with STEs had two or more STEs. Of individuals with ADEs, 78.5% experienced exclusively one ADE category.

Preventability and seriousness of events

The preventability of all ADEs was 38.8% (95% CI 35.8, 41.9%), varying by ADE category (Table 3). Serious ADEs represented 9.5% (95% CI 7.6, 11.3%) of all ADEs, of which DIs were the most and ADRs the least serious. Of all serious ADEs, 55.9% (95% CI 45.8, 66.0%) were preventable. By ADE category, the preventability of serious ADEs was similar to all ADEs, apart from the 54.8% (95% CI 36.3, 73.4%) preventability of serious ADRs. Of the ADEs contributing to hospitalizations, 62.8% (95% CI 48.3, 77.2%) were judged preventable.

Drugs associated with events

Drugs for the nervous system were associated with 39.3% of ADRs, 30.4% of STEs, all DD and DA cases and 62.5% of DIs (Table 4). Also cardiovascular drugs attributed to 29.6% of ADRs and 28.3% of STEs. Among nervous system drugs, psychoanaleptics were the most common (19.8%) among ADRs and analgesics (12.1%) among STEs (Table S1).

By and large, the main drug classes associated with all and preventable ADRs and STEs were similar. Drugs for the nervous system contributed to 43.7% of preventable ADRs (psychoanaleptics 17.8%, psycholeptics 15.6%, analgesics 14.1%), and drugs for the cardiovascular system to 37.8% (β-adrenoceptor blocking agents 15.6%, diuretics 14.1%, agents acting on the renin-angiotensin system 11.9%). Drugs for blood and blood forming organs composed 9.6% (antithrombotic agents 8.1%) of preventable ADRs, and drugs for the musculoskeletal system 8.2% (anti-inflammatory and anti-rheumatic products 7.4%). Of preventable STEs, cardiovascular drugs represented 31.4% (β-adrenoceptor blocking agents 16.9%, agents acting on the renin-angiotensin system 15.1%, diuretics 11.1%), nervous system drugs 21.5% (analgesics 8.1%, psychoanaleptics 8.1%, psycholeptics 4.1%), alimentary tract and metabolism drugs 20.4% (drugs used in diabetes 18.0%), and drugs for the musculoskeletal system 9.3% (anti-inflammatory and anti-rheumatic products 9.3%).

Organs affected by events

ADRs were most frequently gastrointestinal (21.6%) or general disorders (12.3%) (Table 5, Table S2), the most frequent ADR symptoms being fatigue, nausea, dizziness and increased weight. Analogously to all ADRs, preventable ADRs were the most frequently gastrointestinal (20.7%) or general disorders (13.3%).

STEs were most frequently vascular (18.9%), dominated by hypertension (Table 6, Table S3). Preventable STEs were frequently vascular (23.8%), psychiatric (12.2%), or musculoskeletal (9.9%), as for all STEs, but endocrine disorders were more common among preventable (14.5%) than all STEs (8.7%).

UTIs were the most commonly psychiatric (17.3%) or vascular (13.5%), hypertension being the most common individual symptom (12.5%). All DD and DA cases were psychiatric (100%), while DIs were distributed in different organ classes.

Sensitivity analyses for total prevalence of persons with ADEs

The prevalences of ADEs and their categories were higher compared with the main analysis when alternative denominators were used (Table 7), with 18.3% (95% CI 16.9, 19.7%) total ADE prevalence for individuals with dispensed drugs, and 24.5% (95% CI 22.8, 26.2%) for individuals with healthcare encounters. The prevalence of all ADEs remained similar to the main analysis when UTIs were omitted from ADEs: 11.4% (95% CI 10.5, 12.3%) for all individuals, 17.6% (95% CI 16.2, 19.0%) for individuals with dispensed drugs, and 23.2% (95% CI 21.5, 24.9%) for individuals with healthcare encounters. When both UTIs and non-preventable STEs were omitted from ADEs, the prevalence was lower: 9.2% (95% CI 8.4, 10.0%) for all individuals, 14.3% (95% CI 13.1, 15.5%) for individuals with dispensed drugs, and 18.9% (95% CI 17.3, 20.4%) for individuals with healthcare encounters.

Strengths and weaknesses

This is the first study investigating ADEs in both inpatient and outpatient settings of a random population sample, making our results generalizable to the county and by and large the entire nation. However, the under-representation of persons born outside Sweden in our study population somewhat limits generalizability to the entire nation. We chose a retrospective study design, because recruiting a representative population prospectively for detecting their ADEs in healthcare units would have been practically infeasible and resulted in drop-outs, limiting generalizability. As the retrospective data collection enabled assessing symptoms of ADEs based only on the medical records, symptoms of ADEs that patients had not communicated to care providers or care providers had not recorded are underestimated in our study. To minimize the underestimation, our comprehensive case assessment with clinical experts' causality assessment was designed to detect ADEs that were not recognized, diagnosed, or reported as ADEs, but were otherwise detectable, for example, based on free text or diagnostic tests in the medical records. ADEs from prescribed drugs were probably detected to a greater extent than ADEs from other drugs, because non-prescribed and complementary drugs bought without a prescription are excluded from the SPDR and also less commonly recorded in medical records.

Our study benefited from a thorough ADE definition with categories, enabling the investigation of all adverse events related to drugs combined or divided into categories. However, differing definitions in other studies hindered direct comparisons. We applied established methods for case detection and assessing causality, preventability and seriousness, but the varying reliability and the lack of validation of the current methods 51-54 warrant cautious interpretation of the results.

Comparison to prior research

Our finding that 12% of the adult general population experienced ADEs during a 3 month period was, considering our methods, of similar magnitude with most previous studies 9, 17, 20-26, 55, demonstrating that ADEs are a significant burden in the entire healthcare setting. Our lower prevalence than the recently reported 19% 1 month prevalence of self-reported ADEs among Swedish adults 25 may be explained by our retrospective study design 9, the incompleteness of medical records, and patients' unique ability to report events 21, 56. In particular, the higher prevalence of UTIs in self-reports 25 implies their insufficient detection from medical records exclusively. Our inclusive ADE definition and thorough case detection facilitated by access to all medical records, including before and after the study period, probably contributed in our reasonably high 25% ADE prevalence among individuals with healthcare encounters, compared with studies on outpatients 9, 17, 20-24, 26, 55. The higher proportion of admissions contributed by ADEs in our study, 22%, compared with prior studies 6, 9, 10, may in addition be explained by the ambition in Sweden to hospitalize only the most severely ill, at high risk of ADEs. Despite our relatively high ADE prevalences compared with previous observational studies, expert panels have estimated ADEs even more common 57-59, indicating that our prevalences are not overestimations.

Drug classes and organ systems associated with ADEs among the general public varied between the ADE categories, but were by and large similar to previous descriptions for ambulatory care patients 9, 12, 15, 17, 25, 26 and different from hospitalized patients 60-62. As reported previously 9, 17, 25, 26, the most commonly dispensed drugs, nervous system and cardiovascular drugs, contributed to ADEs the most frequently. Within the nervous system drugs, antidepressants dominated ADRs and analgesics STEs, as found before for self-reports 25, while analgesics, hypnotics and sedatives, and anxiolytics caused DD and DA. Although gastrointestinal ADRs have been described as common 12, 15, 25, we found them the most common among ADRs. However, if nervous system and psychiatric ADRs and fatigue were combined in our study, ADRs affecting the ‘central nervous system’ would become the most common, in line with others' findings 12, 15. ADRs or ADEs related to electrolyte, renal, hepatic, and haematologic functions are reported more common among hospitalized patients 60-62 than in our and others' general population samples 15, 25, probably due to the differing nature of outpatient care, patients' age, and possibly overseeing such events in retrospective studies or studies using patient reports. Similarly to disease specific studies 63, 64, but unlike in studies on all ADEs 12, we found hypertension and hyperglycaemia as STEs of antihypertensives and antidiabetics, and psychiatric and musculoskeletal STEs common in the general population. ADEs have previously been described to constitute of heterogeneous events 6, 8-12, 25-32, but our results illustrate that also reporting the associated drugs and affected organs by ADE category further contributes in understanding their nature.

Our 39% preventability of ADEs in the general population is comparable with previous estimates 8-12, 14, as is the similarity of all and preventable ADEs 25, 26. As for all ADEs, nervous system and cardiovascular drugs were the most commonly associated with preventable ADEs, in line with the findings of others 12, 25, 26, 65. Complementary findings to previous research were our high frequencies of preventable STEs of antihypertensives (resulting in hypertension) and antidiabetics (resulting in hyperglycaemia), which in most studies have not been separated from all preventable ADEs from antidiabetics and antihypertensives 12. These results reveal the usefulness of categorizing ADEs also for investigating preventability and developing preventive strategies.

Implications and future research

The heterogeneous nature of ADEs in our study reinforces the demand for improving and harmonizing definitions and classifications for ADEs and preventable ADEs, and methods for assessing them 7, 51, 52, 66, 67. Apart from the traditionally emphasized ADRs, the other categories of ADEs combined caused harm more frequently, and differed in their nature from each other and from ADRs in terms of associated drugs, affected organs, preventability and seriousness. The ADE categories should therefore be considered in research and clinical practice for preventing, detecting and mitigating ADEs. Considering the reduction of ADEs more strongly as part of patient safety and quality of care would probably also benefit conceptualizing ADEs.

Although the results of this study reflect the Swedish healthcare system and the prevalence and pattern of ADEs vary depending on the patient population, settings, ADE definitions and methods 6, ADEs are most likely a significant health concern also in other countries and regions. Our results are the most generalizable to other countries with a similar disease burden dominated by non-communicable diseases 68, a similar pattern of drug use with a high annual prevalence of anti-infective, anti-inflammatory, cardiovascular and nervous system drug use 69, and a similarly structured publicly funded healthcare system 70. Further, ADEs are unlikely to be exceptionally common in Sweden, considering the high quality of the Swedish healthcare system concerning patient safety indicators, compared with other high-income countries 71.

Despite the higher preventability of serious ADEs and ADEs in hospitals, also described by others 9, 26, ADEs should be prevented, detected and mitigated in the entire healthcare system, because a large quantity of non-serious events combined may result in considerable direct resource consumption 72, and indirect costs 73. Further, the high burden of ADEs and preventable ADEs from widely used drugs warrants large scale efforts to redesign safer, higher quality healthcare systems, as urged previously 74-76. Significant improvements in quality and safety require the commitment of clinicians and care units, collaboration with patients, researchers and safety experts, and strong political will and leadership. As framed by Charles Vincent on improving patient safety: ‘Only very few systems have probably understood the nature and scale of capacity development that is actually needed; most have relied on enthusiasm, culture change and people doing quality improvement work in their non-existent spare time’ 74.

In conclusion, the considerable burden of ADEs and preventable ADEs from commonly used drugs in the adult general public warrants large-scale efforts to redesign safer, higher quality healthcare systems, across care settings. The heterogeneous nature of the ADE categories should be considered in research and clinical practice for preventing, detecting and mitigating ADEs.