B0724
Title: Assessment of diagnostic markers subject to limit of detection
Authors: Maria del Carmen Pardo - Complutense University of Madrid (Spain) [presenting]
Christos T Nakas - University of Bern (Switzerland)
Alba Franco-Pereira - Universidad Complutense de Madrid (Spain)
Abstract: Diagnostic accuracy of potential biomarkers is routinely assessed using ROC curve methodology. The area under the ROC curve (AUC) is the most widely used index of diagnostic accuracy. Frequently, marker measurements are subject to limits of detection (LOD) due to laboratory instrument precision. As a consequence, measurements below a certain level are undetectable. Various strategies have been proposed for the assessment of diagnostic markers that are subject to LOD since, simply ignoring observations below the LOD leads to negatively biased estimates of the AUC. We propose a nonparametric approach based on two-sample likelihood ratio tests for right censored data for the assessment of the diagnostic accuracy of a biomarker subject to LOD. We compare with existing methods through a large simulation study. We illustrate our methods on a potential biomarker associated with the non-dipping phenomenon for patients with sleep apnoea syndrome.
