CAPITAL: Optimal subgroup identification via constrained policy tree search.

Area of Science:

• Biostatistics
• Health Informatics
• Personalized Medicine

Background:

• Personalized medicine aims to tailor treatments to individual patient characteristics.
• Identifying patient subgroups with superior treatment response is crucial.
• Current methods often prioritize treatment effect over subgroup size.

Purpose of the Study:

• To develop an optimal subgroup selection rule (SSR) that maximizes patient numbers benefiting from treatment.
• To achieve a pre-specified clinically meaningful mean outcome, such as average treatment effect.
• To introduce a flexible method for personalized medicine subgroup identification.

Main Methods:

• Derived two theoretical forms of the optimal SSR based on treatment-covariate interaction.
• Proposed the Constrained Policy Tree search algorithm (CAPITAL) for optimal SSR identification.
• Developed a method flexible for multiple constraints and time-to-event data (restricted mean survival time).

Main Results:

• The optimal SSR effectively maximizes the number of patients receiving beneficial treatment.
• The CAPITAL algorithm successfully identifies interpretable decision trees for subgroup selection.
• The method demonstrated validity and utility in simulations and real-world data applications.

Conclusions:

• The proposed optimal SSR and CAPITAL algorithm offer a clinically meaningful approach to personalized medicine.
• This method enhances the identification of patient subgroups who benefit most from targeted therapies.
• The approach is adaptable for various clinical outcomes and constraints.