Drug-related deaths among inpatients: a meta-analysis

Purpose To estimate the prevalence of drug-related deaths with regard to total hospital mortality and to explore the heterogeneity in its estimation through subgroup analysis, univariate and multivariate analysis. Methods Two investigators independently searched the PubMed and Google Scholar databases with appropriate key terms to identify observational and randomised studies assessing drug-related problems. The prevalence of drug-related deaths was estimated using a double arcsine method. The heterogeneity was explored through subgroup and univariate analysis for the following study characteristics: study design, age group, study ward, study region, types of drug-related problems, study duration, sample size and study period. The study variables showing significant effects were further explored through a multivariable regression model. The percentage of preventable drug-related deaths was explored as a secondary objective. Results Of the 480 full-text articles assessed, 23 studies satisfying the selection criteria were included. The mean percentage of drug-related deaths was 5.6% (95% CI: 3.8-7.6%; I-2 = 96%). The univariable analysis showed study design (regression coefficient: 4.31) and study wards (regression coefficient: - 6.37) as heterogeneity modifiers. In the multivariable model, only the study ward was considered a significant predictor of drug-related deaths (regression coefficient: - 5.78; p = 0.04). The mean percentage of preventable drug-related deaths was 45.2% (95% CI: 33.6-57.0%; I-2 = 60%). Conclusion Drug-related problems are an important cause of mortality. The variability in its estimation could be explained by admission wards.