METHODS

Selection of participants

The local research ethics committee approved the study. All patients admitted to the medical admissions unit (box 1) and seen by a pharmacist between 1 January and 30 June 2001 were included in the study. The pharmacists saw patients as part of their routine work Mondays to Fridays. They recorded details of all patients seen and reported patients with any suspected drug related morbidity to the principal investigator (RH). Similar methods have been used successfully in at least five previous studies.^6–10

The principal investigator assessed possible cases using a combination of:

• medical note review (throughout the admission and on discharge);

• contact with general practitioners, where necessary, to obtain medication histories, test results, and information regarding the management of patients in primary care;

• interviewing patients, where possible, about the type and duration of symptoms and medication use, using an interview schedule. Those not interviewed included patients discharged home before being seen by the principal investigator, patients too unwell to be interviewed, and patients unable to speak English.

Following assessment by the principal investigator, some cases were excluded from further review (fig 1). These included cases where further investigation suggested that drug related morbidity was unlikely—for example, a low probability VQ scan in a case of suspected pulmonary embolism in a woman taking a combined oral contraceptive. In order to ensure that some patients were included in the study who had been admitted at weekends, the principal investigator identified patients with potential drug related morbidity through medical note review and patient interview on alternate weekends (fig 1).

Classification of potential drug related admissions

Following discharge from hospital, the principal investigator prepared detailed case summaries on patients with suspected drug related morbidity (box 2). These summaries provided reviewers with information on temporal relationships between medications and symptoms, the nature of patients’ symptoms, medication changes resulting from the admission, and the diagnoses made by physicians caring for the patients.

The case summaries were independently scored by a hospital physician (PH), an academic general practitioner with a special interest in prescribing (AA), and a clinical pharmacist (RH). Each reviewer scored the cases using explicit criteria for causality,¹¹ preventability,¹² contribution to the admission,¹¹ and classification of the underlying cause of the drug related morbidity¹² (box 3). All the criteria were validated in pilot study work where 46 possible cases of drug related morbidity were reviewed to ensure the criteria were reliable and gave reasonable agreement between the reviewers. After scoring the cases the reviewers met up and, in keeping with standard practice in this field,³ a majority decision was used to classify the cases where there was disagreement on any of the review criteria.

Admissions were classified as drug related if two or more reviewers scored 4 or 5 using the amended Hallas criteria and judged the drug to have made a dominant or partial contribution to the admission. Admissions were classified as potentially preventable if the above criteria were fulfilled and two or more reviewers scored 4 using the Hepler criteria for preventability.

Recording of data

For all admissions to the medical admissions unit age, sex, and date of admission were recorded on an ACCESS 97 database. The hospital routinely records details of primary diagnoses of patients admitted, classified according to the International Classification of Diseases and Related Health Problems, 10th revision.¹³ These data were available for 7962 (91%) of the admissions in our study and were imported from the hospital database into the study database.

To identify the patients seen by a pharmacist, a record was kept on the database. Data obtained from patient interviews, medical record reviews, and GP contacts were also recorded on the database and were used to generate case summaries for the reviewers. The judgements made by individual reviewers and the final classifications for each case were double entered onto the database. In addition, details of the drugs that were thought to have contributed to the admissions were recorded and were grouped by British National Formulary codes.¹⁴

Analysis of data

Data were exported to SPSS version 10.0 for statistical analysis. To assess differences between the groups of patients seen and not seen by a pharmacist, patient characteristics were compared using an independent t test for age on admission and χ² tests for sex, day of admission, and primary diagnosis (available for 7962 (91%) of all patients admitted). To assess inter-reviewer reliability, case review scores were compared using Cohen’s kappa and intraclass correlation coefficients (moderate to substantial agreement is indicated by kappa 0.41–0.80).¹⁵ The proportion of patients considered to have a drug related admission was calculated as the percentage of patients seen by a pharmacist who were classified as having definite or probable drug related morbidity that had made a dominant or partial contribution to the admission. We also calculated the proportion of drug related admissions that were considered potentially preventable. The proportion of patients considered to have a potentially preventable drug related admission was calculated as the percentage of patients seen by a pharmacist who were judged to have a preventable drug related admission.

RESULTS

Patient characteristics

The flow of patients through the study is shown in box 2. Of the 8781 patients admitted to the medical admissions unit, 4093 (47%) were seen by a pharmacist. The mean (SD) age of patients seen was 62.6 (20.7) years and the mean (SD) age of patients not seen by a pharmacist was 62.0 (20.6) years. Of the patients seen by a pharmacist 2002 (49%) were female and 2091(51%) were male, and of the patients not seen by a pharmacist 2399 (51%) were female and 2289 (49%) were male.

Comparing patients who were either seen or not seen by a pharmacist, there were no significant differences in age (p=0.168), a weakly significant difference in sex (χ² 24.7, df 1, p=0.031), and a marked difference in terms of day of admission (χ² 568, df 1, p<0.001; 3070 (75%) of the patients seen by a pharmacist were admitted Monday to Thursday). Comparison of primary diagnoses indicated that those with infections, nervous system diseases, mental disorders, and respiratory diseases were over-represented in the study, and those with skin and musculoskeletal problems were under-represented (table 1).¹³

Inter-reviewer reliability

Kappa and intraclass correlation coefficients showed good to excellent agreement between the reviewers (table 2).

Proportion of admissions considered drug related

Drug related morbidity was judged to be the cause of 265 (6.5%) of the admissions seen by a pharmacist, and 178 (67%) of these were judged to be potentially preventable. The main underlying causes of preventable drug related admissions were prescribing problems (n=63 (35%)), monitoring problems (n=46 (26%)), and adherence problems (n=53 (30%)). Problems with administration of medication were infrequent (n=9 (5%)). Aspirin associated gastrointestinal bleeds were classified as “potentially preventable with interventions which are not standard care at present” (n=7 (4%)).

The drugs most commonly associated with preventable drug related admissions can be seen in tables 3, 4, and 5. These account for 105 (60%) of the preventable drug related admissions (details of all drugs associated with the preventable admissions are available on the QSHC website at www.qshc.com/supplemental).

DISCUSSION

In this large observational study we found that 6.5% of admissions screened by pharmacists on a medical admissions unit were judged to be drug related and 67% of these were considered preventable. The drugs most commonly associated with potentially preventable drug related admissions were NSAIDs, low dose aspirin, beta-blockers, antiepileptics, diuretics, sulphonylureas, digoxin, inhaled corticosteroids, nitrates, and insulin. These admissions were mainly attributed to problems with prescribing, monitoring, and patient adherence.

In order to minimise the risk of misclassifying cases, we used a methodology which compares well with the most rigorous studies previously reported.^3,8,16,17 We used explicit criteria to guide the assessment of cases and a three member panel of reviewers to independently classify each case. Despite the varied backgrounds of the reviewers, kappa and intraclass correlation coefficients showed good to excellent levels of agreement. Only those admissions where the drug related morbidity was judged to be the dominant cause, or partially contributing, to the admission were included as drug related admissions. Also, when we classified drug related admissions as preventable, it was our view that the problems identified were predictable and controllable in terms of the treatment (Box 4). Nevertheless, our results could be questioned in terms of the extent to which they reflect the true preventability of the observed outcomes. For example, a recent study of hospital deaths that were attributed to medical error¹⁸ suggested that in many cases the error made relatively little difference to expected fatal outcomes in seriously ill patients. While these results are not directly comparable to our study, we recognise that some of the patients in our study were at very high risk of hospital admission regardless of any preventable medication related adverse event.

In our study we reviewed only those patients seen by a pharmacist on the admissions unit. This means that our estimates of the proportion of admissions that were drug related and preventable need to be treated with some caution. The patients we studied were not fully representative of all patients admitted. However, the differences between the groups seen and not seen by a pharmacist were relatively small and, had we studied all the admissions, it is unlikely that the results would have been substantially different. In addition, our findings were similar to median figures from a recent systematic review.³

In recent years there have been calls for researchers to move beyond simply reporting preventable adverse events to finding ways of improving patient safety through avoidance of these events.^1,2 It has been suggested that an approach focusing on systems failures is likely to be most effective.^1,2,19 The key to starting this process is to find out which types of problem are most important.²⁰

Almost 60% of the preventable drug related admissions that we found are shown in tables 3–5. These data describe the drugs most commonly associated with admissions. We have recently completed a systematic review of interventions aimed at reducing medication related morbidity in primary care.²¹ Combining evidence from this review with results from the study presented here means that we are able to suggest the types of interventions that may be most successful in reducing the incidence of preventable drug related admissions.

In terms of prescribing related problems, our study suggests that it may be worth focusing on preventing morbidity associated with NSAIDs and antiplatelet drugs. We know that patients with two or more risk factors for gastrointestinal bleeds are at high risk,²² and that various interventions can reduce the risks of these drugs.^23–25 For example, co-prescription with ulcer healing drugs can reduce the risk of gastrointestinal bleeds²⁴ and educational interventions have been shown to improve the safety of GP prescribing.^23,25 Indeed, in a large Australian study an educational intervention focusing on NSAIDs reduced by 70% the rate of hospital admissions for upper gastrointestinal ulceration, bleeding and perforation.²⁵

In terms of monitoring, some of the best evidence for improvements in primary care relate to the monitoring of warfarin.^26–30 The results of these studies may not be applicable to other drugs, but they suggest that nurse led monitoring clinics,²⁸ computerised decision support systems,^27,28 patient education and involvement,²⁶ and patient self-management^29,30 may help to improve control through improved monitoring. In our systematic review²¹ we did not find many other studies that looked at medication monitoring in primary care, and we believe that there is a need for further research in this area.

In terms of patient adherence, a number of studies have shown that improved education^31,32 and approaches that provide greater involvement of patients in decision making^31,33 improve patient adherence and may reduce drug related admissions. However, few studies have attempted to show clinical benefits.³⁴ Our study suggests that priority should be given to patients taking the drugs listed in table 5.

Many of the problems identified in this study arise from medicines management in primary care, but it should be recognised that drug related admissions are relatively rare from the primary care perspective.³⁵ For example, while NSAIDs were responsible for 12.4% of potentially preventable admissions in our study, a German study estimated that NSAIDs were associated with just 5.5 hospital admissions per 10 000 patients treated with these drugs in primary care.³⁵ This may have important implications for encouraging general practitioners and health service managers that drug related admissions are important enough to justify potentially significant changes in medicines management. The types of changes outlined above could mean increased costs associated with gastroprotection for patients taking NSAIDs, more frequent monitoring for patients taking certain drugs (table 4), and more time spent educating patients and involving them in decision making. Nevertheless, these changes might improve safety for patients and protect healthcare professionals from litigation.