Online Program Home
My Program

Abstract Details

Activity Number: 258 - SPEED: Causal Inference and Related Methodology
Type: Contributed
Date/Time: Monday, July 30, 2018 : 2:00 PM to 2:45 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #332931
Title: Gaussian Process Propensity Scores for Multiple Treatment Regimes
Author(s): Brian Vegetabile* and Daniel L. Gillen and Hal Stern
Companies: UC Irvine and University of California, Irvine and University of California, Irvine
Keywords: Nonparametric Estimation; Multiple Treatments; Observational Study
Abstract:

Adjusting for an estimated propensity score in a binary treatment setting has become a standard procedure to control for confounding within observational studies. Multiple (>2) treatments pose additional issues, such as defining the relevant study population for inference, designing metrics to assess covariate imbalance and selecting a probability model for treatment assignment that balances covariates well amongst the treatment versions. Here, we develop a Gaussian process model to estimate the probability of treatment assignment in the multiple treatment setting. We demonstrate a metric of covariate imbalance that is a function of the weighted distributions of the covariates conditioned upon the observed treatment assignments. This metric depends on the hyperparameters of the Gaussian process through the estimated probability of treatment, leading to an estimation technique that chooses the hyperparameters to minimize covariate imbalance. The utility of the method is demonstrated through simulation studies and through a real world application seeking to estimate the relative effectiveness of multiple intervention arms in a non-randomized setting.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program