Distributed fusion R-learner of heterogeneous treatment effect using distributed medicaid data

Area of Science:

• Health Informatics
• Statistical Learning
• Privacy-Preserving Data Analysis

Background:

• Data-driven decision-making necessitates accurate estimation of heterogeneous treatment effects.
• Integrating data from multiple sites improves sample size for robust CATE estimation.
• Challenges include treatment effect heterogeneity across sites and data privacy concerns.

Purpose of the Study:

• To develop a method for jointly estimating conditional average treatment effects (CATE) across distributed data sites.
• To address challenges of treatment effect heterogeneity and privacy protection in data integration.
• To enable efficient and private information exchange for improved CATE estimation.

Main Methods:

• Proposed a distributed fusion learning approach, DF R-learner.
• Jointly estimates CATE across sites without pooling individual-participant data.
• Employs a data-driven fusion penalty to combine similar parameters and confidence distributions for efficient, private exchange.

Main Results:

• DF R-learner allows for differing CATE functions across sites.
• Achieves improved estimation by combining similar parameters.
• Demonstrated theoretically and empirically no loss of efficiency compared to centralized data methods.
• Successfully applied to study medication treatment for opioid use disorder using distributed Medicaid data.

Conclusions:

• DF R-learner effectively estimates CATE in a distributed, privacy-preserving manner.
• The method addresses key challenges in real-world data integration.
• Offers a viable solution for leveraging multi-site data while protecting sensitive information.