Prevalence of bias attributable to composite outcome in clinical trials published in 2019–2020: a systematic review

TL;DR

This study found that 25% of clinical trials using composite outcomes showed bias, often underestimating treatment effects on mortality.

Contribution

The study introduces the BACO index to quantify bias in composite outcomes and highlights its prevalence in clinical trials.

Findings

25.2% of 91 studies showed significant or suggestive bias in composite outcomes.

73.9% of biased studies underestimated treatment effects on mortality.

No studies showed overestimation of effects attributable to composite outcomes.

Abstract

The aim of this study was to investigate the prevalence of bias attributable to composite outcome (BACO) in clinical trials.

We searched PubMed for randomized clinical trials where the primary outcome was a binary composite that included all-cause mortality among its components from January 1, 2019, to December 31, 2020. For each trial, the BACO index was calculated to assess the correspondence between effects on the composite outcome and mortality. This systematic review was registered in PROSPERO (CRD42021229554).

After screening 1,076 citations and 171 full-text articles, 91 studies were included from 13 different medical areas. The prevalence of significant or suggestive BACO among the 91 included articles was 25.2% (n=23), including 12 with p<0.005 and 11 with p between 0.005 and <0.05. We observed that in 17 (73.9%) of these 23 studies, the BACO index value was between 0 and <1, indicating an underestimation of the effect. The other six studies showed negative values (26.1%), indicating an inversion of the association with mortality. None of the studies showed significant overestimation of the association attributable to the composite outcome.

These findings highlight the need to predefine guidelines for interpreting effects on composite endpoints based on objective criteria such as the BACO index.

Figures