Automated assessment of pulmonary patients using heart rate variability from everyday wearables

Everyday wearables with enhanced computational capacity, good quality sensors, and machine learning/artificial intelligence enabled algorithms have the potential to play a key role not only in the fitness and wellness sector but also in the field of disease diagnostics and monitoring. The major challenges are a limited number of sensors, reliable data collection and processing, and deployment of computationally efficient algorithms. In this multi-cohort study, we have used everyday wearables such as chest band and smartwatch to investigate the heart rate variability (HRV) of 131 subjects which include 40 healthy controls, 69 asthma patients, 9 COPD patients and 13 patients with a co-morbidity of asthma and COPD. We aimed at a comprehensive investigation by exploring a total of 58 features including time domain, frequency domain, non-linear and entropy measures of HRV to identify the HRV indices that provide significant discriminatory information for classification between healthy and pulmonary patients. Feature ranking has been done by the area under the receiver operating characteristics curve. Classification of patients with the 15 top ranked features using data from the chest band heart rate sensor as well as estimated HRV parameters from smartwatch PPG signal have been investigated separately. Using the chest band data, a classification accuracy of 82.07%, precision of 83.13%, recall of 81.53% and F-1 score of 81.7% have been achieved; whereas using the smartwatch data, a classification accuracy of 80%, precision of 79.9%, recall of 80% and F-1 score of 79.94% have been achieved for the test set with an AdaBoost classifier. Heart rate variability metrics also showed significant correlation with disease severity and impact on health-related quality of life as measured by pulmonary function test, Asthma Symptoms Utility Index and COPD assessment test score respectively. The results indicate that HRV analysis using everyday wearables may be helpful in the assessment and management of asthma and COPD.