Bayesian transition models for ordinal longitudinal outcomes

Ordinal longitudinal data on patient health status have been widely collected as an outcome in COVID-19 clinical trials (e.g., WHO Clinical Progression Scale). However, analyses often discard information by collapsing the ordinal trajectories into time-to-recovery or free-days summaries. We introduce the ordinal transition model for ordinal longitudinal outcomes, which is an extension of the proportional odds model for ordinal outcomes to longitudinal outcomes using transition modeling. Using the ACTT-1 clinical trial as a case study, we outline key considerations for analyzing ordinal longitudinal data and demonstrate ordinal transition model fitting. Finally, we present simulation results evaluating the power of the ordinal transition model to detect a treatment effect compared to other commonly used analysis methods. When assumptions are correctly specified, the ordinal transition model can leverage the additional information contained in the ordinal and longitudinal components of the outcomes to significantly increase statistical power, enabling trials to enroll fewer participants and reach conclusions more quickly.