Abstract 14096: Heart Rate Variability in Post-COVID-19 Recovered Subjects Using Machine Learning

Introduction: A significant proportion of patients recovering from COVID-19 infection experience symptoms attributable to autonomic cardiovascular dysregulation. Heart rate variability (HRV) is a non-invasive marker of cardiovascular dysautonomia. Machine learning (ML) models based on HRV can be used to identify post COVID-19 patients with autonomic dysfunction. Methods: We evaluated HRV and blood pressure (BP) responses to orthostatic stress (3-min active standing) in 92 patients within 30-45 days of recovery from COVID-19 infection and 120 healthy controls. HRV was evaluated based on 12-lead electrocardiogram over a 60 second period during supine paced breathing. Lead II was used to extract ECG features including (a) average RR interval, (b) R wave height, (c) Heart Rate (HR) standard deviation and (d) HRV root mean square [HRV-RMS]. We also assed for (1) orthostatic hypotension (OH; >20/10 mmHg fall in BP) and (2) postural orthostatic tachycardia syndrome (POTS; HR increase >30 bpm without OH). Using ML, eleven candidate features were tested with eight algorithms (logistic regression, RandomForests, CatBoost, XGBoost, Extra-tree classifier, Multiple Perceptron (ANN), Support Vector Machines and AdaBoost Classifier) to distinguish between COVID-19 recovered and healthy controls. Results: HRV was significantly lower in post COVID-19 recovered subjects as compared to healthy controls (6.25+4.9 ms vs 9.8+8.9 ms; P<0.001). OH was reported in 12 patients (13.1%) while two patients (2.2%) had POTS. Patients with OH had a significantly lower HRV as compared to those without OH (3.29+3.16 ms vs 6.69+5.01 ms; P=0.025). Accuracy of various ML models varied between 67-80% with multiple perceptron being top model [80% weighted accuracy, AUC: 79.8%, Matthews’s correlation coefficient: 0.59]. Permutation importance feature ranking showed HRV, Average RR and HR to be top feature that distinguish between COVID-19 recovered and healthy controls. Conclusions: A significant proportion of COVID-19 recovered patients experienced autonomic dysfunction as evident by lower HRV and presence of OH and POTS. ML model can help in early identification of autonomic dysfunction thereby leading to proper management in these patients