Title: Adaptive Combination of Randomized and Observational Data

Abstract: Data from both a randomized trial and an observational study are sometimes simultaneously available for evaluating the effect of an intervention. The randomized data typically allows for reliable estimation of average treatment effects but may be limited in sample size and patient heterogeneity for estimating conditional average treatment effects for a broad range of patients. Estimates from the observational study can potentially compensate for these limitations, but there may be concerns about whether confounding and treatment effect heterogeneity have been adequately addressed. We propose an approach for combining conditional treatment effect estimators from each source such that it aggressively weights toward the randomized estimator when bias in the observational estimator is detected. This allows the combination to be consistent for a conditional causal effect, regardless of whether assumptions required for consistent estimation in the observational study are satisfied. When the bias is negligible, the estimators from each source are combined for optimal efficiency. We show the problem can be formulated as a penalized least squares problem and consider its asymptotic properties. Simulations demonstrate the robustness and efficiency of the method in finite samples, in scenarios with bias or no bias in the observational estimator. We illustrate the method by estimating the effects of hormone replacement therapy on the risk of developing coronary heart disease in data from the Women's Health Initiative.