Robust posterior inference for Youden's index cutoff

TL;DR

This paper introduces a robust Gibbs posterior method for directly estimating Youden's index cutoff, allowing for easier incorporation of prior knowledge and improved performance over existing methods without relying on potentially misspecified data models.

Contribution

It proposes a novel Gibbs posterior approach for direct inference on the cutoff, enhancing robustness and flexibility compared to traditional indirect estimation methods.

Findings

The method is robust to data distribution misspecification.

It performs well in simulations against Bayesian and bootstrap methods.

Real data applications demonstrate flexibility and prior incorporation.

Abstract

Youden's index cutoff is a classifier mapping a patient's diagnostic test outcome and available covariate information to a diagnostic category. Typically the cutoff is estimated indirectly by first modeling the conditional distributions of test outcomes given diagnosis and then choosing the optimal cutoff for the estimated distributions. Here we present a Gibbs posterior distribution for direct inference on the cutoff. Our approach makes incorporating prior information about the cutoff much easier compared to existing methods, and does so without specifying probability models for the data, which may be misspecified. The proposed Gibbs posterior distribution is robust with respect to data distributions, is supported by large-sample theory, and performs well in simulations compared to alternative Bayesian and bootstrap-based methods. In addition, two real data sets are examined which illustrate the flexibility of the Gibbs posterior approach and its ability to utilize direct prior information about the cutoff.