Publication date: Oct 10, 2024
This study aimed to determine the prevalence of ordinal, binary, and numerical composite endpoints among coronavirus disease 2019 trials and the potential bias attributable to their use. We systematically reviewed the Cochrane COVID-19 Study Register to assess the prevalence, characteristics, and bias associated with using composite endpoints in coronavirus disease 2019 randomized clinical trials. We compared the effect measure (relative risk) of composite outcomes and that of its most critical component (i. e. death) by estimating the Bias Attributable to Composite Outcomes index [ln(relative risk for the composite outcome)/ln(relative risk for death)]. Composite endpoints accounted for 152 out of 417 primary endpoints in coronavirus disease 2019 randomized trials, being more frequent among studies published in high-impact journals. Ordinal endpoints were the most common (54% of all composites), followed by binary or time-to-event (34%), numerical (11%), and hierarchical (1%). Composites predominated among trials enrolling patients with severe disease when compared to trials with a mild or moderate case mix (odds ratio = 1. 72). Adaptations of the seven-point World Health Organization scale occurred in 40% of the ordinal primary endpoints, which frequently underwent dichotomization for the statistical analyses. Mortality accounted for a median of 24% (interquartile range: 6%-48%) of all events when included in the composite. The median point estimate of the Bias Attributable to Composite Outcomes index was 0. 3 (interquartile range: -0. 1 to 0. 7), being significantly lower than 1 in 5 of 24 comparisons. Composite endpoints were used in a significant proportion of coronavirus disease 2019 trials, especially those involving severely ill patients. This is likely due to the higher anticipated rates of competing events, such as death, in such studies. Ordinal composites were common but often not fully appreciated, reducing the potential gains in information and statistical efficiency. For studies with binary composites, death was the most frequent component, and, unexpectedly, composite outcome estimates were often closer to the null when compared to those for mortality death. Numerical composites were less common, and only two trials used hierarchical endpoints. These newer approaches may offer advantages over traditional binary and ordinal composites; however, their potential benefits warrant further scrutiny. Composite endpoints accounted for more than a third of coronavirus disease 2019 trials’ primary endpoints; their use was more common among studies that included patients with severe disease and their point effect estimates tended to underestimate those for mortality.
Concepts | Keywords |
---|---|
Competing | competing risks |
Coronavirus | Composite endpoints |
Death | endpoint selection |
Efficiency | outcome measures |
Journals | randomized controlled trials |
Semantics
Type | Source | Name |
---|---|---|
disease | MESH | COVID-19 |
disease | MESH | death |