Di Zhang of the Department of Biostatistics defends her dissertation on “Inference on Win Ratio for Clustered Semi-Competing Risk Data”.
Committee Chairperson: Jong H. Jeong, PhD, Department of Biostatistics
Ying Ding, PhD, Department of Biostatistics
Chaeryon Kang, PhD, Department of Biostatistics
Stephen R. Wisniewski, PhD, Department of Epidemiology
Graduate faculty of the University and all other interested parties are invited to attend
Composite endpoints are commonly used in public health with an anticipation that clinically relevant endpoints as a whole would yield meaningful treatment benefits. The traditional way to analyze composite endpoints weights each endpoint equally. This can lead to difficulties in interpreting study results when the components have different clinical importance. The win ratio statistic was proposed recently to resolve this issue by prioritizing the important endpoints through sequential comparisons. The statistical method developments for the win ratio were only in randomized controlled trial settings with independent subjects and no potential confounders. Considering the increasing popularities of research using real-world evidence, we developed statistical frameworks of the win ratio in cluster randomized trial and observational study settings. Throughout the dissertation, we focus on composite endpoints of semi-competing risk structure and two comparison arms, though the proposed techniques could be extended to other types of composite endpoints and multiple comparison arms.
Firstly, we propose to model the win ratio of cluster-randomized data non-parametrically using bivariate clustered U-statistics. The proposed method accounts for the potential dependence among subjects within the same cluster. The asymptotic joint distribution of the joint clustered U-statistics is derived. The asymptotic variance and covariance estimators are constructed and evaluated. Several simulation studies are conducted to assess the type I error probabilities and powers of the test statistic. Then the proposed method is illustrated using a multi-center breast cancer clinical trial.
Secondly, the causal inference for the win ratio in observational studies with independent subjects is developed. We propose to use a combination of propensity score analysis with inverse probability weights and U-statistics. The causal estimand of the proposed estimator is average superiority effect, which is based on the average over marginal distributions of potential outcomes for comparison groups. The asymptotic properties of the proposed test statistic are studied. The asymptotic variance is derived and evaluated through simulation studies.
Lastly, based on the causal inference procedure developed in the second part, we propose a weighted stratified win ratio estimator based on calibrated weights for cluster-correlated data from observational studies. The calibration technique used in the weight estimation creates a good balance of covariates and cluster effects between arms in the overall sample. Additionally, it is robust against misspecified distributional assumptions. The asymptotic properties of the proposed estimator are derived, and the finite sample performance of the estimator is evaluated through simulation studies. The proposed method is applied to an observational study on children with traumatic brain injury (ADAPT trial), using sites or regions as clusters.
Our work has important implications to public health, providing new analytical tools to assess the intervention benefits using informative endpoints, to promote public health and transform health care.