CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Parametric and non-parametric confidence intervals for the maximum of the Youden index and its associated threshold Authors:  Leonidas Bantis - The University of Texas MD AndersonCancer Center (United States) [presenting]
Christos T Nakas - University of Bern (Switzerland)
Benjamin Reiser - University of Haifa (Israel)
Abstract: The area under the receiver operating characteristic (ROC) curve as a measure of the overall accuracy of a given biomarker is commonplace. However, after establishing the utility of a biomarker, clinicians in practice need a decision threshold in order to establish whether intervention or simple monitoring is needed. Unnecessary intervention can be avoided and necessary action can be taken based on an optimal decision threshold. The Youden index can be utilized both for the evaluation of the overall accuracy of a biomarker and the derivation of a clinically interpretable optimized decision threshold which can be used for clinical decision making. We present new methods for the construction of parametric and non-parametric confidence intervals for the Youden index as well as its associated decision threshold. Our parametric methods employ the delta approximation and are based on either the normality assumption or a power transformation to normality. Our non-parametric methods are based on kernels and on restricted cubic splines that are forced to be monotone. We compare our approaches to existing ones through simulations and apply them to serum based markers of a prospective observational study involving diagnosis of late-onset sepsis in neonates.