CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Direct methodology for adjusting ROC curves: A robust approach Authors:  Ana Maria Bianco - Universidad de Buenos Aires (Argentina) [presenting]
Jesica Charaf - Universidad de Buenos Aires (Argentina)
Abstract: Receiver Operating Characteristic (ROC) curve is a graphical tool that became a key tool for evaluating a diagnostic test based on a continuous marker. In practice, several factors, such as age, gender or blood pressure, may improve the discriminatory ability of the marker. When this is the case, it seems wise to assess the possible covariates effects in the ROC analysis to avoid oversimplification. To incorporate this additional information, we follow the direct method, where the effect of the covariates is directly evaluated on the ROC curve by means of a generalized linear model and pseudo-observations. In this framework, the estimator of the ROC curve is obtained through a stepwise procedure. The aim is twofold. On the one hand, we illustrate the instability of the classical method to estimate the conditional ROC curve in presence of outliers, and we also provide a robust alternative. The proposal combines robust estimators of the coefficients of the involved parametric models with an adaptive weighted empirical estimator. Through a Monte Carlo study, we compare the performance of the proposed estimators with that of the classical ones in clean and contaminated samples.