Robust exchangeability designs for early phase clinical trials with multiple strata.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacometrics

Background:

• Clinical trials with multiple strata are essential for drug development, especially in rare diseases.
• Sparse data in early-phase trials complicates statistical inference and decision-making.
• Standard hierarchical models with exchangeable strata parameters risk excessive information borrowing for extreme strata.

Purpose of the Study:

• To introduce the exchangeability-nonexchangeability (EXNEX) approach, a robust mixture extension for multi-strata clinical trials.
• To address the limitations of the standard exchangeability (EX) assumption in hierarchical modeling.
• To enhance statistical inference and decision-making in early-phase trials with sparse data.

Main Methods:

• Proposed the exchangeability-nonexchangeability (EXNEX) approach, a Bayesian mixture model.
• Allowed stratum-specific parameters to be either exchangeable or nonexchangeable with others.
• Utilized standard Bayesian software for computations, with careful model specification and prior selection.

Main Results:

• EXNEX demonstrated tempered information borrowing for extreme strata in case studies.
• Frequentist operating characteristics showed good performance for estimation (bias, MSE) and testing (reduced Type I error inflation).
• The approach proved effective in phase I and II trials with three and four strata.

Conclusions:

• The EXNEX approach offers a robust alternative to standard exchangeability assumptions in multi-strata trials.
• It provides improved statistical inference and decision-making, particularly in early-phase drug development with sparse data.
• EXNEX balances information borrowing, mitigating risks associated with extreme strata.