Evidence synthesis based on non-randomised studies—a critical review of studies leading to conclusions on fall risk properties of loop diuretics/beta-blockers

Abstract

Abstract Purpose To describe methodological and reporting issues in non-randomised comparative drug safety studies pooled in meta-analyses, with focus on confounding by indication. Methods All studies included in statistically significant meta-analyses in a recent publication investigating fall risk properties of cardiovascular drugs were reviewed. Study characteristics were extracted and assessed. Results Nine studies, including between 498 and 321,995 individuals, contributed data to the significant meta-analyses in which loop diuretics and beta-blockers were associated with falls, five published in 2015. Five individual studies reported a statistically significant association. In the five cohort studies, characteristics of exposed vs unexposed individuals were either not reported (n = 3) or differed substantially regarding morbidity (n = 2). Drug treatment was determined at baseline, and data on falls were collected for up to 2 years thereafter. Out of the four case-control studies, the cases and controls in only one study were matched for morbidity. Morbidity characteristics of fallers compared with non-fallers were either not reported (n = 2) or they differed (n = 1) or were reported according to the matched-for diseases (n = 1). Confounding by indication was explicitly discussed in two studies. None of the abstract conclusions considered causality issues or the possibility of confounding by indication. Conclusions Confounding by indication is a major issue in non-randomised comparative drug safety studies, a problem which may be concealed in meta-analyses. To enhance such research, compared groups need to be balanced regarding relevant factors including morbidities and characteristics adequately reported. Confounding by indication needs to be explicitly discussed and highlighted in the abstract conclusion.