Model-Assisted Bayesian Estimators of Transparent Population Level Summary Measures for Ordinal Outcomes in...

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Modeling

Background:

• Ordinal outcomes are common in clinical trials and offer greater statistical efficiency than binary outcomes.
• Traditional analysis using proportional odds models can lack transparency, particularly when assumptions are violated.
• Existing methods may not accurately reflect the treatment effect across all levels of an ordinal outcome.

Purpose of the Study:

• To propose novel, transparent summary measures for ordinal outcomes in randomized controlled trials.
• To develop efficient Bayesian estimators for these population-level summary measures.
• To evaluate the performance of these new measures against traditional proportional odds approaches.

Main Methods:

• Development of 'weighted geometric mean' odds ratios, relative risks, and 'weighted mean' risk differences.
• Application of model-assisted Bayesian estimators using nonproportional odds models.
• Utilizing covariate adjustment with marginalization via the Bayesian bootstrap.
• A proposed weighting scheme ensures invariance to outcome ordering.

Main Results:

• Computer simulations demonstrate that the proposed summary measures perform well compared to proportional odds methods.
• The new measures offer enhanced transparency in emphasizing different components of the ordinal outcome.
• Analysis of the COVID-OUT trial revealed evidence of a nonproportional odds treatment effect, highlighting the utility of the proposed methods.

Conclusions:

• The proposed transparent summary measures and Bayesian estimators provide a robust alternative for analyzing ordinal outcomes.
• These methods are particularly valuable when the proportional odds assumption is violated.
• The findings support the adoption of these advanced statistical techniques for improved clinical trial analysis.