Transporting Results from a Trial to an External Target Population When Trial Participation Impacts Adherence

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Randomized clinical trials (RCTs) are crucial for treatment guidelines but often lack generalizability to real-world populations.
• Differences in covariates and treatment-outcome mediators, such as adherence, limit the applicability of RCT findings.
• Existing methods for addressing covariate differences are established, but methods for mediator differences, particularly adherence, are limited.

Purpose of the Study:

• To develop a sensitivity analysis framework for generalizing RCT results to real-world populations when adherence data is missing.
• To propose methods for estimating mean potential outcomes in target populations by accounting for adherence differences between trial and real-world settings.
• To apply the developed approach to transport treatment effectiveness data for opioid use disorder medications.

Main Methods:

• Proposed a sensitivity analysis incorporating a parameter for relative adherence differences between trial and target populations.
• Discussed methods for specifying the sensitivity parameter using external knowledge, including range setting and Monte Carlo sampling.
• Introduced two estimators for mean counterfactual outcomes: a plug-in estimator and a double-robust one-step estimator utilizing machine learning.

Main Results:

• The proposed sensitivity analysis allows for estimation of treatment effects in real-world populations even with missing adherence data.
• The double-robust estimator offers flexibility by supporting machine learning for nuisance parameter estimation.
• The approach was successfully applied to transport relapse risk data for opioid use disorder treatments.

Conclusions:

• The developed sensitivity analysis framework enhances the generalizability of randomized clinical trial findings to real-world populations.
• The proposed estimators provide robust methods for addressing adherence differences and estimating treatment effects.
• This work has significant implications for informing treatment guidelines and clinical decision-making in diverse populations.