Win Ratio Analysis for Composite Outcomes in Observational Cardiovascular Studies: An Application to the Multi-Ethnic Study of Atherosclerosis

Abstract

The win ratio is a method for analyzing composite hierarchical outcomes that respect a pre-specified clinical priority among component events. Despite growing adoption in cardiovascular trials, its use in observational studies remains virtually unexplored. This thesis evaluates the win ratio and several of its extensions for use with observational data through two investigations. In a simulation study, we compare the hierarchical analysis of the win ratio to the time-to-first-event analysis of the Cox model under varying levels of unmeasured confounding, censoring, and differential component effects. We tested the performance of various inverse-probability weighted win ratio estimators with regards to bias, coverage, type I error, and power. We also show that the win ratio and Cox model reach different conclusions depending on where the treatment effect is concentrated in the hierarchical outcome. In the applied analysis, we compare cardiovascular outcomes across sex and racial/ethnic subgroups in the Multi-Ethnic Study of Atherosclerosis (MESA) cohort using the win ratio and related methods. We construct hierarchical composites outcome endpoints of increasing complexity that capture a broad spectrum of cardiovascular disease, illustrating an example where incorporating baseline coronary artery calcium as a tie-breaking component reverses the direction of the win ratio effect. We provide one of the first applications of the win ratio to demographic comparisons in a large observational cohort study.