Causal Inference
Combining information from trial participants and non-participants in registry-based trials Camila Olarte Parra* Camila Olarte Parra Anthony Matthews Conor Macdonald Anita Berglund Issa J. Dahabreh
The strengths of randomized trials for estimating treatment effects are well understood, but trials often enroll selected populations and have high cost. These drawbacks can be overcome by designing trials nested in registries and using the registry infrastructure to collect information on both trial participants and non-participants. Here, we discuss the identifiability conditions that allow combining information from trial participants and non-participants in registry-based trials to estimate the effects of the randomized treatments, taking into account selection into the trial, assignment into treatment groups, adherence to the assigned treatment, and loss to follow up. We propose estimators that jointly use the randomized and observational data in registry-based trials under different identifiability conditions and we illustrate the use of the estimators using data from a major cardiovascular trial nested in a large national registry.