Multivariate Analysis of Heart Rate Variability as a Predictor of Cardiac Events

Heart Rate Variability (HRV) has been widely studied as a non-invasive tool for assessing
autonomic nervous system function and its relationship with cardiovascular diseases. Reduced
HRV has been linked to increased risk of post-myocardial infarction mortality and sudden cardiac
death [1]. This study aims to identify HRV parameters that can serve as predictors of cardiac
events by differentiating between healthy individuals, patients with myocardial infarction, and
individuals with ventricular tachycardia.
The research used electrocardiographic data from the Physionet database, analyzing 178 RR
interval time series distributed among 75 healthy individuals, 74 myocardial infarction patients,
and 29 individuals with ventricular tachycardia. Time-domain, frequency-domain, and non-linear
methods were applied to extract HRV features, and statistical analyses were performed to determine
significant differences among the groups. The prediction model is applied to perform predictive
analysis using these variables and to characterize the studied groups by calculating the F1 score
for each variable (or their combinations). In this study, a multimodal logistic regression approach
was employed (suitable for more than two groups), enabling the assessment of each variable’s
influence on the outcome. In this context, evaluating the contribution of individual variables or
their combinations across different methods represents a key contribution of this work, as it offers
valuable insight into the predictive capabilities of HRV analysis methods in diagnostic applications.
The results revealed that the parameters of both linear (time and frequency domain) [3], [4]
and nonlinear methods showed significant statistical differences between the groups [2]. The time
domain and non-linear parameters distinguished healthy individuals from patients with myocardial
infarction, as well as patients with myocardial infarction from those with ventricular tachycardia.
However, the parameters in the frequency domain exhibited statistical differences between all
groups, suggesting their potential for a broader classification of cardiac conditions.
To improve predictive precision, a logistic regression model was developed, considering individual
HRV parameters and combinations of multiple variables. The best performing model, which combined four HRV parameters, achieved a mean F1 score of 78.4% to predict cardiac events. The
study underscores the importance of integrating different HRV features into predictive models and
highlights HRV as a valuable tool to assess cardiac risk.
These findings suggest that HRV analysis, when applied using a multivariate approach, could
provide clinicians with a reliable method for the early identification of individuals at risk of severe
cardiac events, potentially improving preventive and therapeutic strategies. Future research should
explore the integration of HRV with machine learning techniques to refine predictive accuracy and
clinical applicability.