Speaker: Susana Diaz Coto
Geisel School of Medicine at Dartmouth, Dartmouth College. Hanover, New Hampshire. USA.
Abstract: The binary classification problem is a hot topic in Statistics. Its close relationship with the diagnosis and prognosis of the diseases makes it crucial in biomedical research. In this context, it is important to have biomarkers that classify individuals as accurately as possible. The receiver operating-characteristic (ROC) curve is a graphical tool routinely used to assess such classification ability. However, given the different nature of diagnosis and prognosis problems, the ROC curve estimation has been tackled from separate perspectives in each setting. We revisit here some of these estimation methods and will focus on the so-called two-stage Mixed-Subject (sMS) ROC curve estimator. This method fits both scenarios through a general prediction model (first stage) and the empirical cumulative estimator of the distribution function of the considered biomarker on the total population (second stage). Besides, it can handle data with missing or incomplete outcome values. We also introduce the R package sMSROC which implements the sMS estimator and includes tools that may support researchers in their decision making.