Randomization-Based Inference for MCP-Mod

Area of Science:

• Pharmacometrics and Biostatistics
• Clinical Trial Design and Analysis

Background:

• Dose selection is crucial for drug efficacy and patient safety in pharmaceutical development.
• The Generalized Multiple Comparison Procedures and Modeling (MCP-Mod) approach is standard for Phase II dose-response analysis.
• MCP-Mod faces challenges with small sample sizes and binary endpoints, particularly complete separation in logistic regression.

Purpose of the Study:

• To introduce penalized maximum likelihood estimation (MLE) and randomization-based inference to address MCP-Mod limitations in small samples.
• To evaluate the performance of these novel methods compared to standard approaches in dose-finding analyses.
• To demonstrate the applicability of these methods in pharmacometric settings.

Main Methods:

• Implementation of penalized maximum likelihood estimation (MLE) to overcome issues like complete separation.
• Application of randomization-based inference for exact finite sample statistical inference.
• Simulation studies to compare the power and type-I error rates of proposed methods against standard MCP-Mod.

Main Results:

• Randomization-based tests significantly enhance statistical power in small to medium sample sizes while controlling type-I error rates.
• Residual-based randomization tests using penalized MLEs improve computational efficiency and outperform standard randomization methods.
• The proposed methods are effective in pharmacometric settings, demonstrating their practical utility.

Conclusions:

• Penalized MLE and randomization-based inference provide robust solutions for dose-finding analyses within the MCP-Mod framework, especially in small samples.
• These methods offer improved statistical power and computational efficiency compared to traditional approaches.
• The study highlights the potential of randomization-based inference for analyzing dose-finding trials with limited data.