ROC Analysis and a Realistic Model of Heart Rate Variability

Abstract

We have carried out a pilot study on a standard collection of electrocardiograms from patients who suffer from congestive heart failure, and subjects without cardiac pathology, using receiver-operating-characteristic (ROC) analysis. The scale-dependent wavelet-coefficient standard deviation σ_{wav}(m), a multiresolution-based analysis measure, is found to be superior to two commonly used measures of cardiac dysfunction when the two classes of patients cannot be completely separated. A jittered integrate-and-fire model with a fractal Gaussian-noise kernel provides a realistic simulation of heartbeat sequences for both heart-failure patients and normal subjects.