Strategies and Considerations for Receiver Operating Characteristic (ROC) Regression Modeling

Abstract

The aim of this thesis is to present and compare approaches for estimating the Receiver Operating Characteristic (ROC) curve. Primarily, our focus is on parametric and semiparametric methods. We also present and apply a semiparametric ROC regression technique to evaluate factors predictive of ROC performance. Our anchoring example is based on a data set involving immunoglobulin-G (IgG) responses to SARS-CoV-2 vaccination. The goal of the original study was to compare humoral (antibody) responses to two vaccine doses between healthy controls and solid organ transplant (SOT) recipients. We utilize semiparametric ROC regression to investigate how time post-transplantation impacts the performance of the ROC curve, the idea being that time post-transplantation serves as a reasonable proxy for extent of immunosuppression. We identified that those less proximal to their time of transplant tended to have IgG responses that more closely resembled those of healthy controls. This result is similar to results identified in the original study, which focused on a comparison of group means.