Investigating Synthetic Controls with Randomized Clinical Trial Data in Rheumatoid Arthritis Studies

Abstract

The cost of clinical research for new drug development has been increasing rapidly. An effective approach to reduce the cost of clinical trials is to use a synthetic control arm to substitute a concurrent control arm. Synthetic control arms are usually created with propensity-score-based methods from historical or external patient-level control data. Although there is much literature discussing how to create synthetic control arms, little is known about how synthetic control arms perform compared to concurrent control arms in real clinical trials. In this paper, we take a real randomized controlled clinical trial and create a synthetic control arm for it using propensity-score-based methods from the control data in other randomized clinical trials. The goal is to demonstrate validity of using synthetic control arms by comparing the performance of synthetic control arms to the concurrent control arm. Four propensity-score-based methods, stratification, matching, inverse probability of treatment weighting, and covariate adjustment are applied to create the synthetic control group. Our results show that the synthetic control arm created with the stratification or matching method could provide an estimate of treatment effect that is as accurate as that of a real randomized clinical trial. This suggests a good opportunity to expedite drug development with reduced cost. We encourage use of these methods in clinical research for drug development when patient-level control data from comparable historical randomized clinical trials are available.