SBSS Student Paper Competition II

• Activity Number: 578 - SBSS Student Paper Competition II
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jennifer E Starling* and Jared S Murray and Carlos Carvalho and Radek Bukowski and James Scott
• Companies: The University of Texas at Austin and The University of Texas at Austin and The University of Texas at Austin and The University of Texas at Austin and The University of Texas at Austin
• Keywords: Bayesian additive regression tree; ensemble method; Gaussian process; regularization; regression tree

Abstract

BART with Targeted Smoothing: An Analysis of Patient-Specific Stillbirth Risk:

We introduce BART with Targeted Smoothing, a new Bayesian tree- based model for nonparametric regression. The goal is to introduce smoothness over a single target covariate while not requiring smoothness over other covariates. tsBART extends the Bayesian Additive Regression Trees model by parameterizing terminal nodes with smooth functions rather than independent scalars. tsBART captures complex nonlinear relationships and interactions among the predictors and guarantees that the response surface will be smooth in the target covariate. This improves interpretability and regularizes estimates.

We apply tsBART to our motivating example: providing patient-specific estimates of stillbirth risk across gestational age, based on maternal and fetal risk factors. Obstetricians expect stillbirth risk to vary smoothly over gestational age, but not necessarily over other covariates, and tsBART has been designed to reflect this structural knowledge. The results of our analysis show the clear superiority of the tsBART model for quantifying stillbirth risk, thereby providing patients and doctors with better in