Cumulative Logit Ordinal Regression with Proportional Odds under Nonignorable Missing Response -- Application to Phase III Trial

TL;DR

This paper develops an EM-based method for ordinal regression with proportional odds in clinical trials with nonignorable missing data, demonstrated through simulations and a psoriasis Phase III study.

Contribution

It introduces a novel EM algorithm for nonignorable missing responses in ordinal regression models, addressing a gap in clinical trial analysis.

Findings

Method performs well in simulations with finite sample properties

Applied successfully to a Phase III psoriasis trial

Provides a SAS implementation for practical use

Abstract

Missing data are inevitable in clinical trials, and trials that produce categorical ordinal responses are not exempted from this. Typically, missing values in the data occur due to different missing mechanisms, such as missing completely at random, missing at random, and missing not at random. Under a specific missing data regime, when the conditional distribution of the missing data is dependent on the ordinal response variable itself along with other predictor variables, then the missing data mechanism is called nonignorable. In this article we propose an expectation maximization based algorithm for fitting a proportional odds regression model when the missing responses are nonignorable. We report results from an extensive simulation study to illustrate the methodology and its finite sample properties. We also apply the proposed method to a recently completed Phase III psoriasis study using an investigational compound. The corresponding SAS program is provided.