Respiratory Infections Prediction via Wearable Sensors: A Machine Learning Approach

Objective: This study aims to explore the efficacy of wearable devices in monitoring vital signs and detecting early indicators of health abnormalities, with a focus on their potential to predict acute respiratory infections. Methods: Utilizing a specially developed app, "Respiratory Health Research," participants were monitored through smartwatches equipped with sensors for heart rate variability (HRV), oxygen saturation, respiratory rate, and body temperature. A machine learning algorithm was designed to analyze these physiological parameters, trained with data from 201 subjects and validated with data from an additional 272 subjects. Participants were included if they had been monitored for at least three days before symtoms onset. The confirmation of respiratory infection was through either telephonic visit with medical diagnosis or self-reported symptom questionnaires. The algorithm's performance was evaluated based on its sensitivity, specificity, and accuracy in predicting the onset of respiratory infection symptoms within a three-day window. Results: The study trained the prediction algorithm with 201 cases (162 males, 39 females; age range 18-87, mean age 38.2 ± 13.28) and validated it with 272 cases (252 males, 20 females; age range 18-78, mean age 36.4 ± 12.7), all diagnosed with acute respiratory infections. The infections included pneumonia, bronchitis, COVID-19, and upper respiratory tract infections. The algorithm demonstrated a sensitivity of 69.5%, a specificity of 91.3%, and an overall accuracy of 80.4% in predicting the onset of symptoms. Conclusion: Our trend prediction algorithm based on wearable device data shows promising accuracy in early prediction of respiratory infections.