A new partition testing strategy for multiple endpoints.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Inference

Background:

• Evaluating multiple efficacy endpoints in confirmatory clinical trials presents significant statistical challenges.
• Existing methods often rely on closed or partition testing, focusing on dose-endpoint combinations as intersection hypotheses.
• These methods can be limited by correlation assumptions and lack of consistent inferences, with likelihood ratio tests rarely used due to computational complexity.

Purpose of the Study:

• To generalize the formulation of multiple endpoints problems, including alternative and co-primary endpoints.
• To propose a novel partition testing approach for evaluating multiple endpoints in clinical trials.
• To develop a method that provides consistent inferences without relying on endpoint correlation estimations.

Main Methods:

• Generalized the multiple endpoints problem formulation to accommodate various primary endpoint scenarios.
• Developed a new partition testing strategy utilizing a consonance-adjusted likelihood ratio test.
• The proposed method avoids direct estimation of endpoint correlations or assumptions of independence.

Main Results:

• The new partition testing approach offers consistent inferences for multiple endpoint evaluations.
• The procedure is demonstrated to be conservative, ensuring robust statistical power.
• It bypasses the need for potentially contentious assumptions about endpoint correlation.

Conclusions:

• The proposed consonance-adjusted likelihood ratio test offers a statistically sound and consistent method for evaluating multiple endpoints in clinical trials.
• This approach addresses limitations of current methods by providing reliable inferences without strong correlation assumptions.
• It offers a valuable tool for clinical trial design and analysis, particularly in complex multi-endpoint scenarios.