Regression discontinuity threshold optimization.

Area of Science:

• Econometrics
• Health Economics
• Machine Learning

Background:

• Clinical and policy decisions often rely on treatment thresholds (e.g., cholesterol levels for statin prescription).
• Existing methods like regression discontinuity estimate treatment effects at specific thresholds but do not optimize them.
• Optimizing these thresholds is crucial for maximizing patient quality-adjusted life years and overall expected outcomes.

Purpose of the Study:

• To develop a framework for optimizing policy thresholds to maximize the cumulative treatment effect on treated individuals.
• To introduce an estimator for the optimal threshold within a constrained optimization setting.
• To extend threshold optimization to include conservative estimates, policy costs, and non-linear relationships.

Main Methods:

• Utilizing a constrained optimization framework to estimate the optimal policy threshold.
• Employing Gaussian process regression for non-linear estimation of treatment effects.
• Extending the methodology to incorporate conservative threshold estimation and policy cost constraints.

Main Results:

• A novel estimator for the optimal policy threshold has been developed.
• The approach effectively uses machine learning for flexible, non-linear estimation.
• The framework allows for the inclusion of cost-benefit analyses and risk-averse decision-making.

Conclusions:

• The proposed method enables the selection of thresholds that optimize treatment benefits while considering practical constraints.
• This approach can be applied to various fields, from healthcare to ride-sharing, for policy optimization.
• The study provides a robust statistical framework for evidence-based threshold setting.