2018/04/20 by Evan T. R. Rosenman, Art B. Owen, Rosenman, Evan +5 · 2 citations
Mathematics · #Advanced Causal Inference Techniques #Statistical Methods and Inference #Statistical Methods in Clinical Trials
paper · pdf · doi:10.48550/arxiv.1804.07863
This project considers how one might augment a limited amount of data from\nrandomized controlled trial (RCT) with more plentiful data from an\nobservational database (ODB), in order to estimate a causal effect. In our\nmotivating setting, the ODB has better external validity, while the RCT has\ngenuine randomization. We work with strata defined by the propensity score in\nthe ODB. Subjects from the RCT are placed in strata defined by the propensity\nthey would have had, had they been in the ODB. Our first method simply spikes\nthe RCT data into their corresponding ODB strata. Our second method takes a\ndata-driven convex combination of the ODB and RCT treatment effect estimates\nwithin each stratum. Using the delta method and simulations we show that the\nspike-in method works best when the RCT covariates are drawn from the same\ndistribution as in the ODB. Our convex combination method is more robust than\nthe spike-in to covariate-based inclusion criteria that bias the RCT data. We\napply our methods to data from the Women's Health Initiative, a study of\nthousands of postmenopausal women which has both observational and experimental\ndata on hormone therapy (HT). Using half of the RCT to define a gold standard,\nwe find that a version of the spiked-in estimate yields stable estimates of the\ncausal impact of HT on coronary heart disease.\n