PhD candidate Lan Shi will present her dissertation on Monday, June 8, 2026, at 2 p.m. Central Time, in the 11th floor large classroom at 2525 West End Avenue and online. Her advisors are Dr. Dandan Liu and Dr. Bryan Blette. All are invited and encouraged to attend. For virtual access, contact the department.
Statistical Advancements in Power and Sample Size Determination for Clinical Trials with Ordinal Outcomes
Power and sample size determination is an essential component of clinical trial design. For trials with ordinal outcomes, this is commonly performed using Whitehead鈥檚 method. However, the expanding use of ordinal outcomes to capture complex and evolving patient states introduces methodological challenges that are not fully addressed by standard approaches. One keychallenge in trial design concerns the increasing use of ordinal composite endpoints constructed from correlated binary components. The WHO COVID-19 Ordinal Scale for Clinical Improvement is a prominent example; for such endpoints, the corresponding ordinal outcome distribution is often unavailable at the design stage. Another challenge arises in cluster randomized trials (CRTs) involving ordinal outcomes, where existing approaches typically extend Whitehead鈥檚 method using design-effect adjustments based on a univariate intraclass correlation coefficient (ICC). However, this Whitehead design-effect approach lacks formal mathematical justification, and there is no established definition of a univariate ICC for discrete ordinal outcomes. This dissertation addresses these challenges through three projects. In the first project, we propose a vine copula-based framework for power and sample size determination for ordinal composite endpoints constructed from correlated binary components. The framework reconstructs the ordinal composite distribution using marginal component prevalences and assumptions on cross-component dependence, and incorporates this distribution into Whitehead鈥檚 method. In the second project, we develop closed-form power and sample size formulas based on generalized estimating equations (GEE) for three-category ordinal outcomes in CRTs under independence and exchangeable working correlation structures. These methods are compared with the Whitehead design-effect approach using an ICC for the latent continuous outcome and with GEE-based binary power and sample size formulas applied to dichotomized ordinal outcomes. Simulation studies and trial-design examples are used to evaluate and illustrate the proposed methods. In the third project, we develop OrdCompCRTpower, an interactive R Shiny application that implements these methods for both individual randomized and cluster randomized trials with ordinal and ordinal composite outcomes. Collectively, these three projects provide methodological advances and practical tools to support the design of modern clinical trials with ordinal outcomes.
