A Cross-Validated Targeted Maximum Likelihood Estimator for Data-Adaptive Experiment Selection Applied to the Augmentation of RCT Control Arms with External Data
share
Augmenting the control arm of a randomized controlled trial (RCT) with external data may increase power but at the risk of introducing bias. Existing methods for combining data sources generally rely on stringent assumptions or may have decreased coverage or power in the presence of bias. We present a novel approach to the integration of RCT and real-world data (RWD) based on framing the problem as one of data-adaptive experiment selection. We aim to use the available RCT and RWD to select the experiment with the optimal bias-variance tradeoff, where potential experiments include the RCT with or without different candidate real-world datasets. We then develop an experiment-selector cross-validated targeted maximum likelihood estimator (ES-CVTMLE) to select and analyze the optimal experiment. The ES-CVTMLE incorporates two bias estimates: 1) a function of the difference in conditional mean outcome under control for the RCT and combined experiments and 2) an estimate of the average treatment effect on a negative control outcome (NCO). We define the asymptotic distribution of the estimator under varying amounts of bias and construct confidence intervals by Monte Carlo sampling from the estimated limit distribution. In simulations involving varying magnitudes of bias and a violation of the U-comparability assumption that is necessary for estimating bias using an NCO, the ES-CVTMLE has improved coverage compared to test-then-pool approaches and an NCO-based bias adjustment approach and improved power compared to a Bayesian dynamic borrowing approach. We further demonstrate the ability of the ES-CVTMLE to distinguish biased from unbiased external controls through a re-analysis of the effect of liraglutide on glycemic control from the LEADER trial. This novel method could help analyze hybrid RCT-RWD studies when running an adequately powered RCT is infeasible.
READ FULL TEXT