PS22: Innovations in Trial Design with Ordinal Endpoints

In an increasing number of clinical contexts, patient outcomes are best summarized by an ordinal endpoint. For example, the modified Rankin scale in stroke has seven discrete values defined by severity of symptoms. Death or the number of hospital-free days can also be fruitfully treated as an ordinal endpoint, and more generally, hierarchical composite endpoints such as the Win ratio are often ordinal. We propose a session with three speakers, a clinician and two statisticians, with 25 minutes each. The first presentation will discuss assumptions and commonalities between different approaches of analyzing an ordinal endpoint, including dichotomization, utility-based analyses, proportional odds models, and the Wilcoxon-Mann-Whitney test. Patient-centered utilities are often criticized for arbitrariness; the speaker will discuss that alternative approaches such as dichotomization or proportional odds have implied utilities. A second presentation will discuss a case study of the design of a trial with longitudinal measurements of an ordinal endpoint. Here, the transition rates between states are estimated, and the effect of an experimental therapy is measured with an odds ratio. The presenter will explore whether departures from the assumption of a common odds ratio threaten the validity of conclusions about efficacy. A third presentation, given by an FDA regulatory expert, will discuss the application of ordinal ideas to hierarchical composite endpoints, specifically the win ratio. The presenter will explore the benefits and risks or ordinalizing a win ratio endpoint, and discuss using multiple imputation instead of declaring ties. Treatment effects at different levels of the hierarchy will be combined in terms of the concordance coefficient. Predictions of final analysis outcomes computed at interim analyses will be illustrated.