Is Short-Term Heart Rate Variability Good Enough to Predict Vascular Events in Hypertensive Patients?

Vascular events are the main cause of premature death and disability in the developed countries, where there is great interest in the development of computational tools for their early detection. A very relevant variable for their study is the heart rate, that can be analyzed through heart rate variability (HRV). Furthermore, high blood pressure is an important risk factor for most cardiovascular diseases. In fact, small reductions in blood pressure are known to markedly reduce cardiovascular morbidity and mortality. This study evaluates the predictive value of short-term HRV (STHRV) by developing models based on data mining algorithms to stratify the risk of vascular events from hypertensive patients. For this specific framework, the performance of various machine learning models (Random Forest, Support Vector Machines, Gaussian Naive Bayes, KN Nearest Neighbours and Logistic regression), trained with different time lengths of 5, 30 and 60 minutes of HRV features during sleep stage was compared. The analyzed HRV parameters were associated to time, frequency and nonlinear features. A total of 139 Holter recordings from hypertensive patients of whom 17 developed a vascular event were analyzed. Results indicated that classification models developed using STHRV, with only 5 minutes length, provided similar or even better results than those developed with longer time series. Furthermore, the STHRV models provided a higher sensitivity and a slightly higher F1 score. The best one, based on Support Vector Machines, yielded 88.2% sensitivity and 75% F1 score. Thus, this research suggests the feasibility of STHRV analysis for risk stratification of hypertensive patients to anticipate serious vascular events.