Hierarchical Bayesian clustering design of multiple biomarker subgroups (HCOMBS)

Area of Science:

• Clinical trial methodology
• Biostatistics
• Oncology research

Background:

• The Food and Drug Administration (FDA) now accepts master protocol designs, shifting oncology research towards molecular subtype-focused studies.
• New methodologies are needed for biomarker-driven clinical trials in small patient groups, assessing treatment-marker interactions, and managing false positives.

Purpose of the Study:

• Introduce Hierarchical Bayesian Clustering Design of Multiple Biomarker Subgroups (HCOMBS), a novel two-stage umbrella Phase II design.
• Address the need for efficient clinical trial designs in precision oncology, reducing sample size and improving statistical power.

Main Methods:

• HCOMBS employs effect size clustering and information borrowing across multiple biomarker-treatment pairs.
• A two-stage umbrella design integrates clustering and hypothesis testing for simultaneous analysis.
• Simulations compared HCOMBS against Simon's Optimal two-stage design.

Main Results:

• HCOMBS demonstrated a reduced participant requirement per treatment arm compared to Simon's design.
• The family-wise error rate was well-controlled, and marginal power was desirable.
• The methodology was successfully applied to the alliance brain metastases umbrella trial.

Conclusions:

• HCOMBS offers an efficient statistical approach for identifying novel marker-based cancer therapies.
• The design effectively manages operating characteristics in multi-arm settings, supporting precision medicine initiatives.
• This methodology facilitates adaptive clinical trial designs for molecularly defined cancer subgroups.