Adverse drug reactions and hospital admissions: Large case‐control study of patients aged 65–100 years using linked English primary care and hospital data

Adverse drug reactions (ADRs) are common and a leading cause of injury. However, information on ADR risks of individual medicines is often limited. The aim of this hypothesis‐generating study was to assess the relative importance of ADR‐related and emergency hospital admission for large group of medication classes.

and propensity scores for hospital admission risk.Medication groups with systemic administration as listed in the British National Formulary (used by prescribers for medication advice).Prescribing in the 84 days before the index date was assessed.
Only medication groups with 50+ cases exposed were analysed.The outcomes of interest were ADR-related and emergency hospital admissions.Conditional logistic regression estimated odds ratios (ORs) and 95% confidence intervals (CI).

Results:
The overall population included 121 546 cases with an ADR-related and 849 769 cases with emergency hospital admission.The percentage of hospitalizations with an ADR-related code for admission diagnosis was 1.83% and 6.58% with an ADR-related code at any time during hospitalization.A total of 137 medication groups was included in the main ADR analyses.Of these, 13 (9.5%) had statistically non-significant adjusted ORs, 58 (42.3%) statistically significant ORs between 1.0 and 1.5, 37 (27.0%) between 1.5-2.0,18 (13.1%)between 2.0-3.0 and 11 (8.0%) 3.0 or higher.Several classes of antibiotics (including penicillins) were among medicines with largest ORs.Evaluating the 14 medications most often associated with ADRs, a strong association was found between the number of these medicines and the risk of ADR-related hospital admission (adjusted OR of 7.53 (95% CI 7.15-7.93)for those exposed to 6+ of these medicines).
• This study evaluated the effects of many medication groups on the risk of ADR-related and emergency hospital admission using two large datasets linking primary care to hospital admission data.
• Of all hospitalizations, 1.83% had an ADR code for admission diagnosis and 6.58% for diagnosis at any time during hospitalization.
• There is a need for a regular systematic assessment of the harm-benefit ratio of medicines, harvesting the information in large healthcare databases and combining it with causality assessment of individual case histories.

Plain Language Summary
Side-effects to medicines (also known as adverse drug reactions or ADRs) are common and a leading cause of injury.However, information on ADR risks of individual medicines is often limited.This study used a large database containing anonymized medical records from elderly patients in primary care in England that was linked to data from hospitals.Of all hospitalizations, 1.83% had an ADR code for admission diagnosis.Of the 137 medication groups included in the analyses, the majority of medication groups had risks for a hospital admission possibly related to an ADR which were higher than unexposed patients.Penicillins doubled the while thiazides and lipid-regulating medicines showed no significant difference in hospitalization risk.In conclusion.
there is a need for a regular systematic assessment of the harm-benefit ratio of medicines, harvesting the information in large healthcare databases and combining it with causality assessment of individual case histories.

| INTRODUCTION
Elderly patients have more long-term conditions and so take more medicines concurrently.This often results in polypharmacy (5+ concurrent medicines), which is particularly common in the frail elderly population, and may be problematic, as highlighted in the recent UK Government Overprescribing Review. 1 This review estimated at least 10% of the current volume of medicines in the United Kingdom may be overprescribed and stressed the need to reduce overprescribing. 1 The importance of optimizing prescribing of medications is also highlighted by the World Health Organization which has identified Medication Without Harm as the theme for the third Global Patient Safety Challenge. 2verse drug reactions (ADRs) are common and can account for considerable morbidity and mortality.A study in 2004 found that 6.5% of hospital admissions were related to ADRs-directly so in 80% of these admissions.Low dose aspirin, diuretics, warfarin, and nonsteroidal anti-inflammatory drugs (NSAIDs) other than aspirin were the medicines most commonly implicated. 3In a US study of emergency hospital admissions in older Americans in 2011, four medication groups-warfarin, insulins, oral antiplatelet agents and oral hypoglycemic agents-were implicated in 67.0% of hospitalizations. 4A metaanalysis of 81 studies reported that around one in 30 patients are exposed to preventable medication harm. 5Another meta-analysis of 42 studies considered hospital admissions due to ADRs in the elderly, finding 8.7% of admissions due to ADRs, with NSAIDs frequently implicated. 6However, the number of cases for medication groups was generally small in this review, highlighting the need for more research with large high-quality data.The aim of this hypothesis-generating study was to assess the relative importance of ADR-related and emergency hospital admission for large group of medication classes.This top-down approach of evaluating composite outcomes for large number of medicines may allow for prioritization of research into the specific adverse effects of specific medicines with potentially largest effects.

| Data
Data sources were Clinical Practice Research Databank (CPRD GOLD) 7 and Aurum. 8Both databases contain longitudinal, anonymized, patient level electronic health records (EHRs) from general practices in the United Kingdom.Almost all UK residents are registered with a general practice, which typically provides most of the primary healthcare.In case that a patient receives emergency care (e.g., at accident & emergency department) or inpatient or outpatient hospital care, the general practice is informed about this.General practices can use one of several clinical information systems, of which EMIS is the commonest.Vision software was used more frequently in the past although use has reduced substantially in recent years in England. 9The CPRD GOLD databases includes general practices that use the Vision EHR software system while Aurum practices use EMIS Web.Practices can change their EHR software although this will be reflected in the start and end of data collection for each practice.CPRD GOLD includes data on about 11.3 million patients 7 and Aurum 19 million patients, 8 although practices and patients may have contributed data for varying durations of time.These databases include the clinical diagnoses, medication prescribed, vaccination history, lifestyle information, clinical referrals, as well as patient's age, sex, ethnicity, smoking history and body mass index.The patientlevel data from the general practices in England only has been linked through a trusted third party to hospital data (hospital episode statistics) using unique patient identifiers. 7The hospital data contained information on the date of hospital admission and the clinical diagnoses established at and during admission and coded using ICD10.Linked data were available, starting 1 April 2007, for visits to emergency departments, including the visit day, but presenting diagnosis data was less complete for these visits.The general practices included in this study were from England agreeing to record linkage.
Patient-level socioeconomic information was approximated from small area level index of multiple deprivation (IMD) linked to the patient's residential postcode. 10Patient-level IMD was aggregated into quintiles for the current analysis.Prescriptions were classified using the British National Formulary (BNF).Information on medical history, ethnicity, lifestyle variables, and medication prescribing was obtained from the general practice data.

| Study population
The source population consisted of patients aged 65-100 years at any time during the observation period.Follow-up observation started on 1 January 2000 and ended on 1 July 2020 (CPRD GOLD) or 1 September 2020 (Aurum).Follow-up of individual patients also considered their start date of registration with a general practice, prior history of registration in the practice of at least 3 years, time of reaching age 65 as well as end date due to moving away or death and time of reaching age 101.The follow-up of each patient was divided into 3-month periods with risk factors such as presence of morbidity assessed at each of these time-periods.These data were used in the matching process.
A case-control study nested within this source population was conducted.Cases of interest were patients with a first hospital admission of interest during follow-up with no prior admission in the preceding year (as recorded in HES).The reason for excluding recent prior hospital admissions was to restrict the ADRs to those related to medicines given outside hospitals.Cases were matched to up to six controls without hospital admission in the year before.
The objective of the matching was to closely match on extent of morbidity based on disease (although not on treatments).Matching was done using propensity matching (using the QAdmission Score) as well as matching by variables including age, sex, morbidity, presence of frailty, practice coding level and calendar time (i.e., number of days between hospital admission date of case and index dates of matched controls).The QAdmissions score estimates the risk of emergency hospital admission for patients aged 18-100 years in primary care. 11It is based on variables such as deprivation score, ethnicity, lifestyle variables (smoking, alcohol intake), and chronic diseases. 11Predictors such as prescribed medications and laboratory values were not used in the calculation as medications were the exposure of interest and laboratory values were not extracted.
Age and calendar time matching was done stepwise (age same year up to difference of up to 5 years; calendar time from within 3 months up to difference up to 5 years).Patients were also matched by groups of medical history of diseases.Clustering techniques using the k-means method were used to group patients into more similar groups of morbidity.Using 38 conditions, 12 the number of clusters was increased stepwise until the number of patients in smaller clusters exceeded 5% of the size of the population.For each practice, the mean level of coding was assessed for each general practice.Nine inception cohorts of starters of medications were identified (including antiarrhythmics, drugs for hypertension/ heart failure, thyroid disorders, anti-Parkinson drugs, anti-dementia drugs, antidepressants, antiepileptics, antihyperglycemic therapy and inhaled bronchodilators).The presence of a code for the indication of treatment was measured and then averaged across the practice.Cases and controls were matched on the quintile of practice coding level (mean in CPRD of 64.6% with 5%-95% range of 54.4 to 76.6; Aurum 74.4%, 61.6%-85.7%).Matching was done separately for CPRD GOLD and Aurum and the risk-set approach to control sampling was used (with control patients potentially included as controls for multiple cases although only once for a particular case).
Two sets of hospital admissions were analysed in this study, including those with an admission code for ADR and emergency status.For ADR-related hospital admissions, we used a code list based on a systematic search and assessment of lists in 41 publications identifying ADRs from administrative data. 13This review classified codes according to level of likely causality based on the ICD-10 code (as determined in the ADR review 13 ).The categories used in the current study included (i) ICD-10 codes with phrase "induced by medication/drug," (ii) ICD-10 codes with phrase "induced by medication or other causes" or "poisoning by medication," (iii) ADRs deemed to be very likely or (iv) likely although the ICD-10 code description does not refer to a drug. 13A large study of emergency hospital admissions reported that unintentional overdoses were frequently a reason for admission. 4Code lists used in this study are included in Supplementary Table 1.Emergency hospital admissions were defined as hospital admissions with a visit to the Accident & Emergency on the same day as the hospital admission (following the approach by Budnitz 4 ).One sensitivity analysis was based on ADRrelated codes recorded at any time during hospitalization (i.e., codes at admission or discharge or during hospitalization).Such codes may include either ADRs that may not be fully determined at the time of admission but only get documented some time into the admission following further investigation or ADRs due to treatments given in the hospital.
The exposures of interest were based on medication groups as listed in the British National Formulary, which is used by prescribers for medication advice.Prescribing in the 84 days before the index date was assessed.The medication groups included medications with systemic administration (including oral, suppositories, injections and inhaled).Nutritional medicines, fluids, electrolytes and vitamins, skin, eye and ear applications, vaccines and anaesthetics were excluded in the analyses of individual medication groups.Also, only medication groups with 50+ cases exposed were analysed.
Patients were also classified at each 3-monthly period into four frailty groups (based on the QFrailty classification). 14The Charlson co-morbidity index was also estimated as indicator of overall level of morbidity. 15

| Statistical analysis
The propensity matching procedure used a caliper (i.e., pre-specified maximum difference) of 0.25 of the logit of the propensity score. 16eedy nearest neighbor matching was used to select the control unit nearest to each treated unit.The SAS procedure PSMATCH was used to conduct the matching.Conditional logistic regression was used to estimate the odds ratios (ORs) and 95% confidence intervals

| RESULTS
The overall population included 121 546 cases hospitalized with an ADR-related admission and 849 769 cases with emergency hospital admission who were matched to controls.A small number of cases (1.2%) were excluded as they could not be matched to at least one control.Of the matched cases, 90.0% were matched by year of birth and within 3 months (the rest were matched up to 5 years of age and calendar time difference).As shown in Table 1, age and sex were comparable between ADR-related cases and controls.The majority of cases were retrieved from the Aurum database (N = 101 992) while 19 554 came from CPRD GOLD.Younger cases had on average more controls per case than elderly cases.Comparing medical history between cases and one randomly sampled control (per case) showed that medical histories were broadly comparable.History of atrial fibrillation was present in 12.6% of CPRD GOLD cases and 14.2% Aurum cases; for one randomly selected control these numbers were 12.9% and 14.8%, respectively.Supplementary Table 2 provides the characteristics at the index date of cases of emergency hospital admission matched to controls.
Table 2 shows the incidence of ADR-related hospital admissions.The rate of hospital admissions with an ADR recorded at admission was 2.69 per 1000 patient-years of follow-up, while this was 12.11 for ADRs recorded at any time during hospital admission.
The incidence of ADR-related hospital admissions increased sharply with frailty (rate of 1.37 in patients without frailty to 7.30 in those with severe frailty).Taking the total number of hospital admissions as denominator, the percentage with an ADR-related code was 1.83% (as coded at admission) or 6.58% (as coded during hospitalization).
A total of 137 medication groups was included in the main ADR analyses.Of these, 13 (9.5%) had statistically non-significant adjusted OR, 58 (42.3%) had statistically significant ORs between 1.0 and 1.5, 37 (27.0%) between 1.5-2.0,18 (13.1%)between 2.0-3.0 and 11 (8.0%) 3.0 or higher.Table 3 shows the ORs for both the medicines with largest increases in risk of ADR-related hospital admission and the medicines with highest prescribing rates.Several antibiotic classes were among the medicines with largest ORs (quinolones had an adjusted OR for ADR-related hospital admission of 3.32; 95% CI 3.16-3.49).Supplementary Tables 3 and 4 provide results for all medication groups included in the main ADR analyses as well as results for hospital admissions with ADRs recorded at any time during hospital admission.Use of a different time-window for exposure (30 instead of 84 days before the index date did not find substantive differences in results (Supplementary Table 5).
Evaluating the 14 medications most often associated with ADRs, a strong association was found between the number of these medicines and the risk of ADR-related hospital admission (adjusted OR of 7.53 for those exposed to 6+ of these medicines as shown in Table 5).However, the effects of individual medicines varied substantially.

| DISCUSSION
This study found that ADR-related hospital admissions are frequent in patients aged 65-100 (with estimates ranging between 2% and 7%).
Frail elderly patients showed the highest absolute risks.A large/ comprehensive number of medication groups was evaluated in this study and one in five medication groups had substantive risks of ADRs (ORs of 2.0+).Antibacterials, laxatives, and drugs used in nausea and vertigo were amongst the medication groups that had particularly increased risks of ADR-related hospital admissions.these data have major utility in identifying new ADRs (signal detection), they have significant limitations in estimating ADR-related risks (signal quantification).Systematic use of linked electronic health data, such as used in this study, could enhance the timeliness, quantification, and actionability of ADR reporting.In the United States, the Sentinel System uses medical billing information and electronic health records to answer regulatory questions on approved medicines. 18wever, these large research databases are not yet used to systematically provide clinicians with information on the magnitude of ADRrelated risks.Further, it can be argued that any modern health system should be able to import, export and act upon this kind of ADR intelligence.Payers, regulators and providers of healthcare widely recognize that the value of a medicine depends on its harm-benefit ratio. 19But despite the high risk of ADR-related hospital admissions observed in this and other studies, 6 clinicians and patients seldom receive properly quantified information on the magnitude of ADR risks for specific patient populations.The need for risk quantification may be particularly important for patients with polypharmacy, which involves a very large number of different combinations of medicines.
A related study found substantial differences between medicine combinations in risks of ADR-related and emergency hospital admissions. 20A B L E 4 Odds ratios (ORs) for emergency hospital admission for medicines with highest ORs and for most frequently prescribed medicines (top 15 for both).This study used a generic code list to identify ADRs based on structured hospital data 13 in order to conduct analyses across a broad set of medicines and estimate an overall ADR incidence.However, not all ADRs included in the analyses may be causally related to each of the medicines.For example, it is unlikely that antibiotics will cause peptic ulcers.Rather antibiotics such as clarithromycin are used to treat the Helicobacter pylori infections that caused these ulcers. 21ere is, thus, an opportunity to systematically combine the strengths of the spontaneous reporting system and of healthcare databases not just for signal detection 22 but also signal quantification.The careful causality assessment of case histories to assess possible causal relation between medicine and ADR 23 could also be applied to case histories as collected in healthcare databases.The outcomes of the case classifications could then be used in signal quantification with detailed data on characteristics of medicines users as obtained in the healthcare databases.[27] Antibiotics are widely used in primary care for a variety of common infections. 28There is a major concern about patients becoming resistant to antibiotics 29 and possible lack of effect in patients who use antibiotics frequently. 30,31There are also concerns about frequent inappropriate antibiotic prescribing 32 which may reduce the antibiotic benefits to an individual patient.Patients at low risk of infectionrelated complications are also as likely to receive an antibiotic as higher risk patients. 33Other than narrative reviews and information on relative effects of ADRs with antibiotics, 34 there is limited evidence on the magnitude of harm-benefit ratio overall and by specific patient groups.In three large studies that evaluated ADR-related hospital admissions, [35][36][37] no information was provided on the relative and absolute effects of individual medication groups including antibiotics.
Even for widely used medicines such as antibiotics, there is a need for a systematic assessment of the harm-benefit ratio of medicines. 38,39ere are several limitations of this study.One is that residual confounding due to differences between cases and controls in underlying disease severity cannot be excluded.Despite our efforts to control for presence of disease, this matching did not include disease severity.However, it would be unrealistic for randomized trials to be conducted to address the multitude of ADRs as assessed in this study.
The results of this study should be interpreted as potential risks of medicines to be considered in the context of the benefits of the treatment and possible causal mechanisms.Another limitation was that this study relied on ADRs as coded by hospital reviewers of hospital records.A smaller study in a single pediatric center found that the incidence of ADR-related hospital admissions was five times higher based on classification by clinical pharmacists compared to coded records. 40The effects of any under-coding in this study may be dependent on whether this was similar or not between those exposed or non-exposed (non-differential misclassification typically biases towards the null).However, we did use two different approaches to ADR-related hospital admission (i.e., admission only or any time).The admission ADR analyses had often higher ORs as could be expected as ADRs at any time during hospital admission could also be related to treatments given in the hospital.Another limitation was that we used a generic definition for ADRs, 13 which will not have included all possible ADRs for all medicines (for example, bleeding outcomes were not included in the ADR list while important for oral anticoagulants).
While specific ADR code lists for specific medicines would be best to evaluate hypotheses of particular drug effects, this study aimed to assess the relative importance of ADRs for large group of medicine classes.Exposure misclassification is a further misclassification as exposure data were based on prescribing data rather than actual consumption by patients.
In conclusion, we found that ADR-related hospital admissions were common in patients aged 65-100 years.There is a need for a regular systematic assessment of the harm-benefit ratio of medicines, harvesting the information in large healthcare databases and combining it with causality assessment of individual case histories.We propose that large scale data sources are systematically evaluated for possible ADRs with medicines (signal generation including assessment of possible excess absolute risks).This signal generation would then be followed by an in-depth evaluation and testing of the signals using appropriate methods.

(
CIs) for hospital admission.Potential confounders included IMD quintiles, smoking history and indicator of missingness, ethnicity (other ethnicity and missing ethnicity with Caucasians as reference) and individual morbidities (to further minimize residual confounding by underlying morbidity).The morbidities included atrial fibrillation, cancer, asthma or chronic obstructive pulmonary disease, congestive heart failure, cardiovascular disease, dementia, epilepsy, falls, learning disability, leg ulcer, chronic liver or renal disease or pancreatitis, Parkinson's disease, rheumatoid arthritis, chronic kidney disease, type 1 diabetes, type 2 diabetes, schizophrenia, dementia and venous thromboembolism.All analyses were performed using SAS software version 9.4.
T A B L E 2 Incidence rate in the source population of ADR-related hospital (as recorded at admission and during hospital admission) stratified by age, sex, and frailty.Characteristics of cases with ADR-related hospital admission and controls (propensity matched by variables such as age, sex and clinical characteristics) stratified by data source.
a Number of cases per 1000 person-years of follow-up (follow-up was based on period of patient's registration with practice).T A B L E 1 Odds ratios (ORs) for ADR-related hospital admission for medicines with highest ORs and for most frequently prescribed medicines (top 15 for both).
Large systems have been developed to collect data on ADRs, particularly spontaneous reporting systems that collect data on suspected ADRs from clinicians and patients.Many regulatory authorities collect these data and the WHO has the largest database (called VigiBase), with over 30 million reports of suspected ADRs of medicines. 17While T A B L E 3 2441Medicines listed as single group in Ref.41.variable data available in spontaneous reports24by using the longitudinal data in healthcare databases linking, for example, primary care to hospital records. Plorms that enable distributed queries of linked data from health systems such as OpenSAFELY (www.opensafely.org)embedded in CIPHA (www.cipha.nhs.uk) or the European Health Data & Evidence Network (www.ehden.eu)could provide a systematic and secure approach for large-scale ADR analytics.This has provided substantial, prompt intelligence in COVID-19 responses.
a Medicines listed as single group in Ref. 4. b