Improving trial generalizability using observational studies.

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Randomized controlled trials (RCTs) provide robust evidence but may lack generalizability.
• Observational studies (OSs) reflect real-world populations but are prone to bias.
• Combining RCT and OS data can yield more reliable and generalizable treatment effect estimates.

Purpose of the Study:

• To develop a novel statistical framework for integrating RCT and OS data.
• To enhance the generalizability of treatment effect estimates from clinical trials.
• To improve the precision and robustness of causal inference in comparative effectiveness research.

Main Methods:

• Proposed a calibration weighting estimator to balance covariates between RCT and OS datasets.
• Developed a doubly robust augmented calibration weighting estimator based on semiparametric efficiency theory.
• Utilized a nonparametric sieve method for robust approximation of nuisance functions and data-adaptive predictor selection.

Main Results:

• The proposed estimators demonstrated improved generalizability of treatment effect estimates.
• The doubly robust estimator achieved the semiparametric efficiency bound under stated assumptions.
• Simulation studies and a real-world application confirmed the finite sample performance of the methods.

Conclusions:

• The developed calibration weighting methods effectively integrate RCT and OS data.
• The approach enhances the external validity of findings from randomized controlled trials.
• This methodology offers a powerful tool for estimating treatment effects in diverse populations, as demonstrated in the adjuvant chemotherapy example.