Using Subject Level Covariate Information in Bayesian Mixture Models for Basket Trials.

Area of Science:

• Oncology
• Biostatistics
• Clinical Trial Design

Background:

• Basket trials are increasingly vital in precision medicine for evaluating treatments across multiple cancer types.
• They offer efficiency gains and enable testing drugs in rare cancers but face challenges due to treatment efficacy heterogeneity.
• Existing models struggle to incorporate patient-level data effectively for early-phase decision-making.

Purpose of the Study:

• To develop a novel statistical model for optimizing go/no-go decisions in early-phase basket trials.
• To address the challenge of treatment efficacy heterogeneity across different cancer types within a basket trial.
• To leverage subject-level covariate information for more robust treatment evaluation.

Main Methods:

• A Bayesian mixture model, termed the Bayesian Partition Model with Covariates (BPMx), was developed.
• The model incorporates a latent cluster structure where mixture weights are informed by subject-level covariate similarity.
• Bayesian statistical methods were employed for robust estimation and inference.

Main Results:

• The BPMx model demonstrated robust estimation of basket-level mean response in simulation studies.
• The model provided insights into the underlying latent cluster structure within the trial data.
• Application to a published basket trial illustrated the model's practical utility.

Conclusions:

• The BPMx model offers a powerful tool for enhancing decision-making in basket trials, particularly in early phases.
• It effectively handles heterogeneity in treatment response by incorporating subject-level covariates.
• This approach can improve the efficiency and success rate of precision medicine trials, especially for rare cancers.