Targeted Smooth Bayesian Causal Forests: An analysis of heterogeneous treatment effects for simultaneous versus interval medical abortion regimens over gestation
share
This article introduces Targeted Smooth Bayesian Causal Forests, or tsbcf, a semi-parametric Bayesian approach for estimating heterogeneous treatment effects which vary smoothly over a single covariate in the observational data setting. The tsbcf method induces smoothness in estimated treamtent effects over the target covariate by parameterizing each tree's terminal nodes with smooth functions. The model allows for separate regularization of treatement effects versus prognostic effect of control covariates; this approach informatively shrinks towards homogeneity while avoiding biased treatment effect estimates. We provide smoothing parameters for prognostic and treatment effects which can be chosen to reflect prior knowledge or tuned in a data-dependent way. We apply tsbcf to early medical abortion outcomes data from British Pregnancy Advisory Service. Our aim is to assess relative effectiveness of simultaneous versus interval administration of mifepristone and misoprostol over the first nine weeks of gestation, where we define successful outcome as complete abortion requiring neither surgical evacuation nor continuing pregnancy. We expect the relative effectiveness of simultaneous administration to vary smoothly over gestational age, but not necessarily other covariates, and our model reflects this. We demonstrate the performance of the tsbcf method on benchmarking experiments. The R package tsbcf implements our method.
READ FULL TEXT