Cumulative Logit Ordinal Regression With Proportional Odds Under Nonignorable Missing Responses-Application to Phase...

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Modeling

Background:

• Missing data are common in clinical trials, particularly those with ordinal responses.
• Missing data can arise from various mechanisms, including nonignorable missingness where data depend on the response itself.
• Accurate analysis requires methods that appropriately handle these nonignorable missing data patterns.

Purpose of the Study:

• To propose an expectation-maximization (EM) based algorithm for proportional odds regression models.
• To address scenarios with nonignorable missing ordinal responses in clinical trial data.
• To provide a robust statistical method for analyzing incomplete categorical data.

Main Methods:

• Development of an expectation-maximization (EM) algorithm tailored for proportional odds models.
• Simulation studies to evaluate the methodology's performance and finite sample properties.
• Application of the proposed method to real-world clinical trial data from a Phase III psoriasis study.

Main Results:

• The proposed EM algorithm demonstrates effectiveness in fitting proportional odds regression models with nonignorable missing data.
• Simulation results confirm the methodology's validity and finite sample performance.
• Successful application to a Phase III clinical trial highlights practical utility.

Conclusions:

• The developed EM algorithm provides a viable solution for analyzing ordinal response data with nonignorable missingness in clinical trials.
• This method enhances the reliability of statistical inferences when dealing with complex missing data.
• The study offers a practical tool, including a SAS program, for researchers.