Improved inference for MCP-Mod approach using time-to-event endpoints with small sample sizes

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• The Multiple Comparison Procedures with Modeling Techniques (MCP-Mod) framework is approved for phase II studies.
• Current MCP-Mod relies on Maximum Likelihood (ML) estimators with asymptotic properties unsuitable for small sample sizes.

Purpose of the Study:

• To derive improved ML estimators and covariance matrix corrections for the censored Weibull regression model.
• To evaluate the performance of improved ML estimators within standard regression and MCP-Mod frameworks.

Main Methods:

• Developed corrective and preventive approaches for improved ML estimators.
• Conducted two simulation studies comparing ML and improved ML estimators.
• Evaluated estimators and covariance matrices in regression and MCP-Mod settings.

Main Results:

• Improved ML estimators demonstrated reduced bias compared to standard ML estimators.
• Wald-type statistics derived from improved ML estimators effectively controlled type I error.
• Type I error control was maintained without power loss across both tested frameworks.

Conclusions:

• Improved ML estimators are recommended for the MCP-Mod approach, especially for small sample sizes (5-25 subjects per dose).
• These estimators ensure type I error is controlled at the nominal level in phase II clinical trials using MCP-Mod.