Integrating decision modeling and machine learning to inform treatment stratification

Area of Science:

• Decision Analysis
• Health Economics
• Machine Learning in Healthcare

Background:

• Traditional "one size fits all" (OSFA) treatment approaches are being replaced by stratified decision-making.
• Understanding patient covariate effects on treatment effectiveness and cost-effectiveness is crucial for stratification.
• Machine learning (ML) methods can identify outcome heterogeneity without pre-specifying subgroups.

Purpose of the Study:

• To propose a method integrating ML estimates with decision modeling for long-term, policy-relevant outcomes.
• To develop a novel policy tree implementation for defining subgroups based on decision model outputs.
• To evaluate the impact of ML integration and subgroup stratification on treatment decision-making.

Main Methods:

• Integration of ML-based survival time estimates into a decision modeling framework.
• Implementation of policy tree algorithms to define patient subgroups using decision model outputs.
• Application to the Systolic Blood Pressure Intervention Trial (SPRINT) data, comparing standard vs. intensive blood pressure targets.

Main Results:

• Integrating ML into decision models can alter the estimated incremental net health benefit (INHB) for OSFA policies.
• Stratifying treatment decisions using ML-defined subgroups, identified via a tree-based algorithm, can enhance INHB estimates.
• The SPRINT trial data demonstrated potential benefits of personalized treatment strategies over OSFA.

Conclusions:

• Combining ML with decision modeling offers a powerful approach for personalized treatment strategies.
• Subgroup identification through ML-driven policy trees can optimize treatment decisions and improve health economic outcomes.
• This integrated methodology supports a move towards more effective and cost-efficient stratified healthcare.