Heart rate variability as a biomarker in patients with Chronic Chagas Cardiomyopathy with or without concomitant digestive involvement and its relationship with the Rassi score

Abstract

Background: Dysautonomia plays an ancillary role in the pathogenesis of Chronic Chagas Cardiomyopathy (CCC), but is the key factor causing digestive organic involvement. We investigated the ability of heart rate variability (HRV) for death risk stratification in CCC and compared alterations of HRV in patients with isolated CCC and in those with the mixed form (CCC + digestive involvement). Thirty-one patients with CCC were classified into three risk groups (low, intermediate and high) according to their Rassi score. A single-lead ECG was recorded for a period of 10-20 min, RR series were generated and 31 HRV indices were calculated. The HRV was compared among the three risk groups and regarding the associated digestive involvement. Four machine learning models were created to predict the risk class of patients.

Results: Phase entropy is decreased and the percentage of inflection points is increased in patients from the high-, compared to the low-risk group. Fourteen patients had the mixed form, showing decreased triangular interpolation of the RR histogram and absolute power at the low-frequency band. The best predictive risk model was obtained by the support vector machine algorithm (overall F1-score of 0.61).

Conclusions: The mixed form of Chagas' disease showed a decrease in the slow HRV components. The worst prognosis in CCC is associated with increased heart rate fragmentation. The combination of HRV indices enhanced the accuracy of risk stratification. In patients with the mixed form of Chagas disease, a higher degree of sympathetic autonomic denervation may be associated with parasympathetic impairment.

Keywords: Autonomic nervous system; Chagas disease; Heart rate variability; Machine learning; Rassi score.

Fig. 1

Feature selection using the four machine learning algorithms. S: selected feature; KNN: k-nearest neighbors; SVM: support vector machine; MLP: multilayer perceptron; RF: random forest. For the description of HRV indices (columns), please see the text

Fig. 2

Confusion matrices for the best classification models (same shown in Table 4). KNN: k-nearest neighbors; SVM: support vector machine; MLP: multilayer perceptron; RF: random forest. Low: low-risk group; Intermediate: intermediate-risk group; High: high-risk group