Bayesian Semiparametric Approach to Constrained ROC Curves Using Placement Values

• Activity Number: 82 - Statistical Methods for Disease Prevention and Prediction
• Type: Contributed
• Date/Time: Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Soutik Ghosal* and Zhen Chen
• Companies: Eunice Kennedy Shriver National Institue of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD)
• Keywords: ROC curve; Placement values; probability ordering

Abstract

ROC curve is widely used to study the accuracy of a test in discriminating between two different populations. We propose a semi-parametric method to estimate ROC curves of multiple tests using the placement values when a priori constraints exist on the accuracies. Placement value can be considered as a discriminatory measure between two populations and is defined as the standardization of the response from the diseased population relative to that of the reference or healthy population. ROC curve can be interpreted as the CDF of the placement values. If a constrained relationship exists between tests, suitable probability orders are desired to use. Here, we impose stochastic ordering and variability ordering on the distributions of placement values to estimate ROC curves. We adopt a fully Bayesian method to estimate the ROC curves from stochastic and variability ordered placement values using Dirichlet process mixture. We apply our proposed approach in the Scandinavian Fetal Growth Study to investigate the predictive ability of mean abdominal diameter of the fetus at various gestational ages on certain birth outcomes in a cohort of pregnant women of Caucasian origin.