Uso de regressão logística para avaliação dos parâmetros de métodos lineares e não lineares aplicados na variabilidade da frequência cardíaca

Abstract

With the increase in cardiovascular diseases and the need for non-invasive and specific tools for diagnosis and prognosis, Heart Rate Variability is an alternative that serves to analyze RR intervals, which is an interval identified from the Electrocardiogram, having a direct relationship whit the Autonomic Nervous System, which may be a marker for heart disease. For time series analysis of RR intervals, there are a variety of parameters that can be obtained from linear (time and frequency domain) and non-linear analysis methods. Thus, establishing possible predictors of cardiac events can be a challenging task, and a multivariate approach, such as logistic regression, is a appropriate for the elaboration of a prediction model. The objective of this work is to analyze the time series of RR intervals in normal, infarcted and ventricular tachycardia individuals and, through this analysis, to verify it from the RR intervals it is possible to separate the individuals according to the cardiac event that occurred. RR intervals, QT intervals and TT intervals were analyzed, where these data were obtained from the Physionet database. The analyzes were performed using the software Kubios and GraphPad Prism 5. The results showed statistical differences between all groups, as in the time domain where there were differences in all parameters between normal vs. normal groups, infarctions, in the frequency domain the highest prevalence was between normal vs. ventricular tachycardia and in the nonlinear method was identified in the comparison of normal vs. infarcted and infarcted vs. tachycardia, while the parameters of the frequency domain, QT intervals and Tp-e intervals established a statistical difference between all groups. In addition, logistic regression was used as a complementary analysis and, among the prediction models developed, it was found that the combination of four variables has an average F1 score of 78.4% for the accuracy of cardiac events. Time domain and frequency domain parameters were highlighted as possible predictors in the logistic regression model. The work that, from the analyzes carried out, it is possible to differentiate the predictive parameters of the compared groups, being the heart rate variability a physiological phenomenon that can suggest, from its analysis, a diagnosis and a personalized prognosis, from computational methods, contributing to medial practice.