Machine learning in comparative effective research of pharmacological interventions: a scoping review

Machine learning (ML) is rapidly transforming the evaluation of the comparative effectiveness research (CER) of pharmacological interventions. It offers an attractive data-driven alternative to traditional statistical causal inference methods for estimation of heterogeneous and individualized treatment effects and is particularly well-suited for high-dimensional real-world data. While variety of ML algorithms have been applied to CER, methodological challenges and limitations remain, and recommendations for conducting, reporting, and evaluating such studies have only recently been proposed.

This paper conducts a scoping review of ML approaches used in CER for pharmacological interventions. We searched multiple databases from January 2020 to January 2026, supplemented by reference mining, and included 28 primary studies. We focused on research topics, ML methodologies, reporting, validity, and robustness of the findings, and appraised risk of bias.

Although the use of ML methods in CER of pharmacologic intervention is relatively recent, it is expanding and consolidating around few specific methods, notably doubly robust estimators paired with ensemble learners. However, methodological rigor and reporting consistency remain uneven, yielding a moderate risk of bias. Standardized methods, transparent reporting, and dedicated appraisal tools are needed before these approaches can reliably inform research, clinical practice, and policy.