On Kullback-leibler Divergence as a Measure for Medical Diagnostics and Cut-point Selection Criterion

Presented at ENAR Conference

Recently, Kullback-Leibler divergence measur (KL), which captures the disparity between two distributions, has been considered as an index for determining the diagnostic performance of markers. This study investigates variety of applications of KL divergence in medical diagnostics, including overall measures of rule-in and rule-out potential and proposes an optimization criteria based on KL divergence for cut point selection. Moreover, the paper links the KL divergence with some common Receiver Operating Characteristic (ROC) measures and presents analytically and numerically the relations in situations of one crossing point as well as multiple crossing points. Furthermore, the graphical application and interpretation of KL divergence, which is referred as the information graph, is discussed. A comprehensive data analysis of the Dutch Breast Cancer Data are provided to illustrate the proposed applications.