Automatic Prediction of Vascular Events by Heart Rate Variability Analysis in Hypertensive Patients

Abstract

This paper proposes an automatic classifier for risk assessment of developing vascular events in hypertensive patients. The proposed classifier separates lower-risk patients from higher-risk ones, using linear and nonlinear Heart Rate Variability (HRV) measures. Higher risk patients were those having a clinical vascular event (e.g. myocardial infarction, syncope, stroke or transient ischemic attack) within one year after the Holter recording. A database of Holter recordings with clinical data of patients followed up for at least 12 months were collected a hoc. 17 out of 142 patients had one of the following vascular events: 11 myocardial infarctions, 3 strokes, 2 syncopal events. An ensemble tree-based algorithm suitable for imbalanced dataset, called Rusboost, was adopted to develop the classifier. The proposed classifier achieved sensitivity and specificity rate of 71% and 66%, respectively, in identifying higher risk patients. The abnormal HRV revealed by the proposed classifier represented a strong risk factor of vascular events within one year among hypertensive patients (odds ratio higher than 4). Finally, the proposed system outperformed the classification based on carotid intima media thickness measurement, which is a proven power predictor of future vascular events.