Systematic Review of Doubly Robust Methods for Population-Adjusted Indirect Comparisons
Author(s)
Sarah Ren, PhD1, Jen-Yu Amy Chang, MSc, RPh, PhD1, Nicholas Latimer, MSc, PhD1, Ruth Wong, BSc, MSc, PhD1, Min-Hua Jen, PhD2, Uwe Siebert, MPH, MSc, ScD, MD3, Kate Ren, PhD1.
1University of Sheffield, Sheffield, United Kingdom, 2Eli Lilly, Uxbridge, United Kingdom, 3UMIT TIROL - University for Health Sciences and Technology, Hall in Tirol, Austria.
1University of Sheffield, Sheffield, United Kingdom, 2Eli Lilly, Uxbridge, United Kingdom, 3UMIT TIROL - University for Health Sciences and Technology, Hall in Tirol, Austria.
OBJECTIVES: Population-adjusted indirect comparisons (PAICs), including matching-adjusted indirect comparison (MAIC), simulated treatment comparison (STC), and multilevel network meta-regression (ML-NMR), are increasingly used in health technology assessment (HTA) to compare treatments across heterogeneous trial populations. Doubly robust methods have emerged as a promising approach, providing asymptotically unbiased estimates when either the trial assignment or the outcome model is correctly specified. This review aims to explore the development and application of doubly robust estimators within PAICs.
METHODS: We systematically searched PubMed, EMBASE, and Web of Science from inception to May 2025. Two reviewers independently screened studies. Articles were included if they elaborated on doubly robust methods for covariate adjustment in settings where individual patient-level data (IPD) were limited and only aggregate data were available for some studies. Extracted data included study objective, outcome types, method assumptions, and reported benefits or limitations.
RESULTS: Of the 54 articles screened, 14 underwent full-text review and four met the inclusion criteria. Two papers demonstrated that entropy-balancing approaches, such as MAIC, yield consistent estimates under linear outcome models, even when the trial assignment model is misspecified. One study investigated the integration of data-adaptive methods, including machine learning, into doubly robust frameworks, highlighting both potential and challenges. Another proposed a doubly robust estimator for marginal hazard ratios but did not assess its performance either theoretically or through simulation. Across all studies, common limitations included susceptibility to bias when both models were misspecified or when important confounders were unmeasured.
CONCLUSIONS: Although doubly robust methods offer theoretical advantages in PAIC settings, their practical application and methodological development remain limited. Further research is needed to validate and expand their role in HTA.
METHODS: We systematically searched PubMed, EMBASE, and Web of Science from inception to May 2025. Two reviewers independently screened studies. Articles were included if they elaborated on doubly robust methods for covariate adjustment in settings where individual patient-level data (IPD) were limited and only aggregate data were available for some studies. Extracted data included study objective, outcome types, method assumptions, and reported benefits or limitations.
RESULTS: Of the 54 articles screened, 14 underwent full-text review and four met the inclusion criteria. Two papers demonstrated that entropy-balancing approaches, such as MAIC, yield consistent estimates under linear outcome models, even when the trial assignment model is misspecified. One study investigated the integration of data-adaptive methods, including machine learning, into doubly robust frameworks, highlighting both potential and challenges. Another proposed a doubly robust estimator for marginal hazard ratios but did not assess its performance either theoretically or through simulation. Across all studies, common limitations included susceptibility to bias when both models were misspecified or when important confounders were unmeasured.
CONCLUSIONS: Although doubly robust methods offer theoretical advantages in PAIC settings, their practical application and methodological development remain limited. Further research is needed to validate and expand their role in HTA.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
SA90
Topic
Methodological & Statistical Research, Study Approaches
Topic Subcategory
Meta-Analysis & Indirect Comparisons
Disease
No Additional Disease & Conditions/Specialized Treatment Areas