Prevalence of medication administration errors in hospitalized adults: A systematic review and meta-analysis up to 2017 to explore sources of heterogeneity.

Previous estimates to meta-analyze administration error rates were limited by the high statistical heterogeneity, restricting their use. This study aimed to investigate sources of heterogeneity in pooled administration error rates in hospitalized adults. We systematically searched scientific databases up to November 2017 for studies presenting error rates/relevant numerical data in hospitalized adults. We conducted separate meta-analyses for the numerators: One Medication Error (OME) (each dose can be correct or incorrect) and Total Number of Errors (TNE) (more than one error per dose could be counted), using the generic inverse variance with a 95% confidence interval. Heterogeneity was assessed using the I and Cochran's Q test. We meta-analyzed 33 studies. The global pooled analyses based on the OME and TNE numerators showed very high heterogeneity (I  = 100%; p < 0.00001). For each meta-analysis, subgroup analyses based on study characteristics (countries, wards, population, routes of administration, error detection methods, and medications) yielded results with consistently elevated heterogeneity. Beyond these characteristics, we stratified the studies according to the mean error prevalence level as the threshold. Based on the OME numerator, we identified two subgroups of low (0.15[0.13-0.17]; I  = 0%; p = 0.43) and high (0.26[0.24-0.27]; I  = 38%; p = 0.17) pooled prevalence rates, with controlled heterogeneity. Similarly, for the TNE numerator, we identified two subgroups of low (0.10[0.09-0.10]; I  = 0%; p = 0.76) and high (0.28[0.27-0.29]; I  = 0%; p = 0.89) pooled prevalence rates, with controlled heterogeneity. These subgroups differed regarding the denominators used: Total opportunities for errors versus others (doses, observations, administrations). Calculation methods, specifically the denominator, seem a primary factor in explaining heterogeneity in error rates. Standardizing numerators, denominators, and definitions is necessary.