In the context of the literature
Previous reviews of interventions to reduce hospital admissions have focused on non-pharmacological initiatives [
10,
43‐
45]. Our study, therefore, fills an important gap as the first systematic investigation of medications that affect hospital admissions. We mapped a large evidence base and prioritised therapies based on the quality of the evidence and support of clinical guidelines.
Our results complement a growing body of evidence about drug-related hospital admissions. Several systematic reviews of observational studies have shown that approximately 3% of all emergency hospitalisations are related to suspected adverse drug reactions and drug–drug interactions [
46‐
51]. Drugs often associated with hospital admissions include antiplatelet drugs, diuretics, renin-angiotensin system blockers, NSAIDs and anticoagulants. However, the reviews have not considered the number of admissions these drugs help to avoid and therefore, have not provided evidence about their net effects on admissions. Our study complements this literature, as it shows a strong body of evidence supporting a beneficial reduction in admissions for renin-angiotensin system blockers [angiotensin-converting-enzyme (ACE) inhibitors and angiotensin II receptor blockers] and aldosterone receptor antagonists. We have also identified strong evidence for a harmful increase in cause-specific admissions due to heart failure from the use of COX2 inhibitors and pneumonia from the use of inhaled corticosteroids in COPD patients. Many of the other drugs commonly associated with drug-related admissions did not appear in our study. Several reviews of RCTs focusing on antiplatelet drugs and anticoagulants were excluded, because they reported admissions as part of composite outcomes. The net effect of antiplatelet drugs and anticoagulants on hospital admissions is, therefore, not clear from the published systematic review literature.
Implications for practice
Policymakers and commissioners may use these results to prioritise quality measurement and improvement efforts. We have systematically identified a list of 11 evidence- and guideline-based treatments for five major chronic diseases: heart failure, stable coronary artery disease, COPD, asthma and schizophrenia. These diseases cause millions of hospital admissions each year globally [
52‐
56] and account for about 5% of all emergency admissions in high-income countries [
57‐
59]. Yet, there is evidence of significant variation in the prescribing of some these 11 therapies in the United States and Europe, including under-dosing and under-prescribing [
60‐
62]. Therefore, improving utilisation of these medications could translate to substantial reductions in hospital admissions. Potential improvement targets include minimising gaps in prescribing, correcting over- and under-dosing, and improving patient adherence, although the specific target of any improvement initiatives should be driven by locally identified shortfalls in care.
The results of this study may be fed into existing mechanisms for tracking and improving clinical practices in health systems. Prescribing data for some of the 11 evidence- and guideline-based medications are currently monitored in several health systems, for example, use of ACE inhibitors, angiotensin II receptor blockers and beta blockers is currently measured as part of the UK Quality and Outcomes Framework, a pay-for-performance incentive structure that has demonstrably reduced prescribing gaps for these medications and improved prescribing efficiency for other medications [
63‐
65]. All of the top 11 medications we identified could be considered for inclusion in these types of quality assurance and incentive structures. This pathway to optimising medication utilisation may be a feasible strategy to help reduce emergency hospital admissions.
For some of the therapies in this review, bridging treatment gaps and improving prescribing may result in immediate benefits and may rely only on a small number of stakeholders. For example, we found evidence that treating patients with acute exacerbations of asthma in the emergency department with inhaled corticosteroids or anticholinergics helped to avoid hospital admissions. Optimal utilisation of these medications could be achieved through direct efforts by emergency department physicians and hospital pharmacists.
However, many of the therapies in this review form part of ongoing chronic disease management, and beneficial effects occur after months of use. Optimal utilisation of these medications would require coordination of multiple stakeholders, including physicians in different specialties, home care or case management nurses, community pharmacists and patients.
We can estimate potential reductions in hospital admissions by combining the results of our study with data on existing treatment gaps and disease prevalence. For example, in studies of prescribing for heart failure in the United States and Europe, 13% of patients with reduced left ventricular ejection fractions who are eligible for treatment do not receive the first-line therapy [
60,
66]. It has been estimated that 1.5% of people in developed countries have heart failure, of whom 40% have reduced ejection fractions [
52,
67‐
69]. Therefore, about 400,000 Europeans and 250,000 Americans with heart failure and reduced ejection fractions are eligible for first-line therapy but are not receiving it [
70]. Based on the numbers needed to treat and baseline hospitalisation event rates that we have reported for ACE inhibitors and angiotensin II receptor blockers, closing these treatment gaps could help to avoid approximately 28,000 (95% CI 24,000 to 37,000) hospital admissions in Europe and 18,000 (95% CI 15,000 to 24,000) admissions in the United States per year.
Our results also reinforce the dangers of prescribing COX2 inhibitors, particularly for patients at risk of cardiovascular disease, inhaled corticosteroids in patients with moderate to severe stable COPD, and intermittent antipsychotic drug therapy for patients with schizophrenia. It is, therefore, reassuring that the harms associated with the use of these drugs and steps to ensure that they are used appropriately, if at all, have been reported widely in clinical and government prescribing guidelines [
71‐
73].
In addition to high- and moderate-quality evidence, this study identified 28 medications with low- and very low-quality evidence for reducing admissions. According to the GRADE working group, these estimates for reducing admissions are likely to change if additional research were conducted [
18]. Therefore, given the uncertainty around the estimates of effects for these 28 medications, we would not recommend prioritising these for reducing emergency admissions. However, we recognise that the level of evidence required to act may vary depending on the circumstances, the stakeholders involved and the available evidence (which may change over time). If limited evidence is available and there is a pressing need to act, lower-quality evidence may be sufficient to justify cautious implementation of an intervention. Our prioritisation of medications with high- and moderate-quality evidence does not prevent stakeholders from using the low- and very low-quality evidence if justified in the context of their health-care settings; however, we recommend a robust evaluation to ensure resources are appropriately allocated to those interventions most likely to impact on practice.
Implications for research
Only 1% of the reviews examined the effect of medications in patients with multi-morbidity. Given the challenges of effective clinical management and high hospitalisation risk for patients with multiple diseases [
47,
74], we need to identify which medication combinations most help multi-morbid patients to avoid hospitalisation.
Low- and very low-quality evidence indicates the need for high-quality research to increase confidence in the reliability of effect estimates. Some of the medications in this overview were supported by low- and very low-quality evidence, suggesting a need for additional high-quality research. Hospitalisations, however, are only one important patient and health system outcome. A larger set of core outcomes that reflect patient and health system priorities should be considered when establishing research priorities, including assessment of mortality, adverse events, quality of life and cost.
Similarly, there were 17 drugs with high- and moderate-quality evidence that were not supported by clinical guidelines. We did not record whether the medications had been evaluated. In formulating guidelines, the effect of an intervention on reducing emergency admissions would form only one of many criteria considered by multi-disciplinary panels of stakeholders. We would not recommend that drugs be included in guidelines or considered for inclusion solely because they reduce hospital admissions.
Researchers should consider reporting rates of hospital admissions, as opposed to ratio measures; 476 of the 517 (92%) effect estimates we reviewed were reported as odds or risk ratios. These are crude measures of hospitalisation, as they assess admission as a binary outcome: present or absent. These effect measures equate a patient who has had one admission during follow-up with a patient who has had five admissions. In patients with chronic diseases, such as heart failure or COPD, for which hospital admissions are common, rate-based measures may have greater utility in evaluating the effectiveness of interventions.
Strengths and limitations of the study
This study has three key strengths. First, it was comprehensive. We identified, analysed and synthesised information on nearly one million patients to identify the highest quality evidence for medications that affect emergency hospital admissions. Secondly, we minimised the impact of duplicate RCT evidence between reviews by excluding outdated reviews; every review we included has a unique set of hospital admission data. Thirdly, we classified all patient populations using ICD-10 and therapies using WHO ATC. This helped to homogenise and simplify the data, which was heterogeneously reported in the systematic reviews. When possible, we also converted quantitative data to comparable measures and units (i.e. risk ratios for estimates and months for follow-up duration). This will enable users of our review to navigate and interpret this large body of evidence.
This study has some limitations. First, although we extracted and reported all effect estimates, we have conducted only quality assessments on significant estimates. While it may be useful for decision makers to know the quality of evidence for all tested interventions, our aim was to support decision-making by identifying and prioritising therapies for which an effect has been demonstrated. All the estimates are listed in the online database (Additional file
2). Secondly, we planned to examine secondary outcomes, such as mortality and cost; however, feasibility concerns emerged during the conduct of the review. Therefore, we analysed only our primary outcome, hospital admissions. To provide information about other outcomes, we have extracted and presented conclusions from the abstract of each review. Furthermore, medications that were supported by high- or moderate-quality evidence were cross-referenced with clinical guidelines to identify those for which the overall balance of benefit to harm was judged to be acceptable by key health-care stakeholders. Thirdly, we excluded reviews that reported hospital admissions only as part of a composite end point. We may, therefore, have excluded potentially valuable therapies. However, by excluding composite outcomes, we can be confident that the effective therapies we identified have a significant effect on hospital admissions. Fourthly, we analysed clinical guidelines from at least one national organisation in the UK, Europe or America and identified support for 11 of the medications in this overview. However, it is possible that there are other relevant clinical guidelines that we did not analyse that support additional medications from this study. Readers of the overview may combine the findings with relevant clinical guidelines in their field to identify additional medications that may be considered for quality measurement and improvement to reduce hospital admissions. Fifthly, our GRADE ratings of indirectness were assessed by generalist clinicians and it is possible that clinical specialists (e.g. cardiologists) may have different opinions regarding the comparability of certain subgroups of patients and interventions. Finally, we planned to include reviews in all languages; however, feasibility concerns emerged during the conduct of the review and as a result we included only reviews published in English. Therefore, we may have missed effective therapies examined in non-English reviews.