Estimating Heterogeneous Treatment Effects With Real-World Health Data: A Scoping Review of Machine Learning Methods

Area of Science:

• Health Economics
• Biostatistics
• Machine Learning

Background:

• Heterogeneous treatment effects (HTEs) describe individual variations in treatment response.
• Estimating HTEs optimizes patient care and resource allocation.
• Machine learning (ML) offers tools to estimate HTEs using real-world data (RWD) when randomized controlled trials are impractical.

Purpose of the Study:

• To categorize ML approaches for HTE estimation in RWD.
• To evaluate the methodological quality of studies applying ML for HTEs.
• To provide guidance for health economics research.

Main Methods:

• A scoping review following PRISMA-ScR guidelines was conducted.
• Searched major databases (PubMed, Scopus, Web of Science, EBSCO, MEDLINE) for studies from 2014-2025.
• Assessed methodological quality using a standardized checklist.

Main Results:

• 74 studies met inclusion criteria, categorized into prediction-only (n=8), outcome modeling (n=9), and customized conditional average treatment effect (CATE) estimation (n=58).
• Most ML innovations for HTEs originated outside health economics.
• Methodological quality was inconsistent, with limited uptake of advanced methods in health economics.

Conclusions:

• ML methods are increasingly used for HTE estimation with RWD, with tree-based models and CATE approaches gaining traction.
• Improved evaluation standards and transparent reporting are crucial for reliable application in health economics.
• Further research is needed to bridge the gap between ML advancements and health economics practice.