A0218
Title: A comprehensive overview of ROC curve estimation: Applications in medical research
Authors: Musie Ghebremichael - Harvard University (United States) [presenting]
Abstract: The receiver-operating characteristic (ROC) curve is widely used in many fields, including medicine. Due to its widespread usefulness, various ROC estimation and testing methods have been continuously evolving. Each method makes different assumptions about the underlying distribution of the data, yields a curve with different properties, and produces different estimates of the subsequent area under the ROC curve. These often leave researchers with a complex problem of deciding which method to use. The properties and performances of parametric, nonparametric, semiparametric, Bayesian, and placement value-based ROC curves are investigated. Extensive simulation studies were carried out across various distribution types, sample sizes, and degrees of overlap between the diseased and non-diseased population distributions. Additionally, real-world data from pediatric HIV/AIDS research were analyzed to illustrate and compare the methods. Findings demonstrate that different ROC estimation methods can produce significantly varying results, highlighting the necessity for thorough validation of chosen approaches. With the growing focus on biomarker discovery and the advent of high-throughput assays and diagnostic tools, the ROC will remain an important analytic tool in medical research. A practical guide is provided for statisticians and researchers working in medical fields to select ROC inference methods appropriate to their specific data.
