Counting the wins: extensions and applications of win statistics for time-to-event outcomes in randomized controlled trials
Embargo Date
2028-03-05
OA Version
Citation
Abstract
This dissertation develops and evaluates extensions of win statistics for analyzing time-to-event outcomes in randomized controlled trials (RCTs). Win statistics, including the win ratio (WR), provide an alternative interpretation to the conventional hazard ratio (HR). Although win statistics, and particularly the WR, have been increasingly adopted in clinical trials, further methodological development and broader clinical applications are needed. Chapter 1 addresses settings of non-proportional hazards by introducing a novel nonparametric approach that partitions follow-up time into disjoint intervals to estimate the WR. This approach leverages a meta-analytic framework to detect and quantify treatment effect heterogeneity in contexts such as immuno-oncology and cardiology trials. Chapter 2 considers the challenge of censoring in generalized pairwise comparisons (GPC), where censored pairs often do not contribute information to the test statistic. We propose a method based on pseudo-observations to address this issue and compare its performance with existing GPC methods. Chapter 3 extends Kraemer’s composite-moderator framework to time-to-event outcomes using the expected win time against reference (EWTR). By combining multiple baseline moderators into a composite measure, we evaluate conditions under which personalized treatment recommendations (PTRs) outperform single moderators. Extensive simulation studies, along with illustrations using reconstructed and real clinical trial datasets, were conducted to assess these methods. Collectively, this work advances methodological tools for analyzing time-to-event data in RCTs by addressing three key challenges: non-proportional hazards, censoring, and treatment effect heterogeneity. These contributions aim to improve the detection of treatment effect heterogeneity and treatment effect estimation and support the development of personalized treatment strategies.
Description
2026