Estimating Diagnostic Accuracy When Candidate Tests Are Subject to Detection Limits---Application to Diagnosing Early Stage Alzheimer Disease

• Activity Number: 246 - Improved Disease Classification Through Extensions of ROC Curve Estimation and Biomarker Characterization
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Chengjie Xiong* and Jingqin Luo and John Morris
• Companies: Division of Biostatistics, Washington University in St. Louis and Division of Public Health Sciences, Washington University in St. Louis and Knight Alzheimer Disease Research Center, Washington University in St. Louis
• Keywords: Diagnostic accuracy; ROC curves; AUC; Alzheimer disease; detection limits; Cox proportional hazards model

Abstract

In disease diagnosis, individuals are usually assumed healthy or diseased. When diagnostic markers are fully measured, Receiver Operating Characteristic (ROC) curves are used to assess diagnostic accuracy. However, full measurements of diagnostic markers may not be available due to various practical/technical limitations. For example, in the diagnosis of Alzheimer disease (AD) using cerebrospinal fluid (CSF) biomarkers, the Roche Elecsys® immunoassays have a measuring range for multiple molecular concentrations. We propose a new statistical method for estimating the diagnostic accuracy when a marker is subject to detection limits by dividing the entire study sample into two sub-samples by a threshold. We propose a family of estimators to the area under the ROC curve (AUC) by combining a conditional nonparametric estimator and another conditional semi-parametric estimator derived from a Cox’s proportional hazards model. We derive the variance to these estimators, assess their performance through a simulation study, and recommend the optimum thresholds. Finally, we apply the method to assess the ability of CSF biomarkers and cognitive tests in diagnosing early stage AD dementia.