Testing for treatment differences with dropouts present in clinical trials--a composite approach

Area of Science:

• Clinical Trials Analysis
• Biostatistics
• Longitudinal Data Analysis

Background:

• Missing data due to patient dropout is a significant issue in clinical trial analysis.
• Traditional methods may not accurately reflect clinical relevance when analyzing continuous outcomes.
• Focusing solely on hypothetical complete-data means can be misleading.

Purpose of the Study:

• To propose a clinically relevant analytical approach for continuous outcomes in clinical trials with missing data.
• To shift focus from hypothetical complete-data means to the conditional mean of study completers and dropout probabilities.
• To develop and evaluate multiple testing procedures for this new analytical framework.

Main Methods:

• Utilized a pattern-mixture modeling approach to factor the likelihood function.
• Directed analysis towards multiple testings of a composite hypothesis.
• The composite hypothesis incorporates dropout probability and the conditional mean of completers.
• Reviewed three closed step-down multiple-testing procedures.

Main Results:

• The proposed approach provides a more clinically interpretable analysis of continuous outcomes in the presence of missing data.
• Demonstrated the application of pattern-mixture models and multiple testing procedures.
• Illustrative data from multiple clinical trials confirmed the utility of the approach.

Conclusions:

• The conditional mean of completers combined with dropout probability offers superior clinical relevance for analyzing continuous outcomes in clinical trials.
• The proposed pattern-mixture modeling and multiple testing framework effectively addresses missing data challenges.
• This methodology enhances the interpretability and robustness of clinical trial results.