“ROC Curve/Surface, Floor/Ceiling Effect, Survival Model, and Alzheimer Disease”
Abstract: In medical diagnosis, subjects are usually assumed to be one of two basic types —– healthy and diseased (the gold standard). Often times, candidate diagnostic markers are much cheaper and less invasive, but must be compared to the gold standard in their sensitivity and specificity to accurately diagnose the diseases. When candidate diagnostic markers are fully measured, Receiver Operating Characteristic (ROC) curves and surfaces have been the standard approaches for measuring the diagnostic accuracy. However, measurements of diagnostic markers may not be available above or below some limits due to various practical and technical limitations. For example, in the diagnosis of Alzheimer disease (AD), the Roche Elecsys immunoassays have a measuring range of 200–1700 pg/mL for cerebrospinal fluid (CSF) Aβ42 concentration, and a measuring range of 80–1300 pg/mL and 8–120 pg/mL, respectively, for CSF total tau and phosphorylated tau concentration. Some cognitive tests, such as Trailmaking A and B, are also subject to floor and/or ceiling effects. This talk presents a statistical methodology for estimating the diagnostic accuracy when a diagnostic marker is subject to detection limits by dividing the fully observed measurements into two parts with a threshold. We propose a family of estimators to the area under ROC curve (AUC) or the volume under ROC surface (VUS) by combining a conditional nonparametric estimator and another conditional semi-parametric estimator derived from a Cox’s proportional hazards model. We further assess the performance of the proposed estimators as a function of possible thresholds, and recommend the optimum choice. Finally, we apply the proposed methodology to assess the ability of CSF biomarkers and Trailmaking A and B in diagnosing early stage AD.
For inquiries contact Chengjie Xiong.