A0447
Title: Robust integration of external data in randomized trials
Authors: Guanbo Wang - Harvard University (United States) [presenting]
Abstract: Randomized trials have low efficiency in estimating the average treatment effect when their sample sizes are limited. One approach to improving efficiency is incorporating external data, such as data from other trials or trial emulation, into the estimation process. When the external data are compatible with the trial data and the nuisance models are correctly specified, their integration can yield consistent estimates and enhance efficiency. However, they are usually incompatible in practice. A consistent and asymptotically normal estimator is developed that can leverage external data even if the data are incompatible and nuisance estimates via non-parametric models do not converge to their true values. More importantly, the efficiency of the estimator is no lower than that of the estimator that uses the trial data only. Therefore, investigators can use external data to improve the trial's efficiency without concern for bias. The finite-sample efficiency gains of the method are demonstrated through small sample-sized simulation studies, and the methods are illustrated using data from two trials of paliperidone extended-release for schizophrenia.
