AI-assisted monitoring of COVID-19 community isolation in Thailand

Abstract

By minimizing human movement and contact, community isolation is an effective containment measure for the COVID-19 pandemic, especially against later strains that cause less severe symptoms. Nonetheless, a significant number of patients who enter community isolation with mild symptoms eventually develop severe pneumonias and require hospitalization. Therefore, the ability to foresee severe cases would be indispensable for managing limited medical resources. Here, we developed a proof-of-concept machine learning model, using daily vital signs data from 1,123 community isolation patients in Bangkok, Thailand, that can predict future hospitalization events up to 3 days in advance with an area under the precision-recall curve of 0.95. The model requires simple inputs, including body temperature, pulse rate, peripheral oxygen saturation, and shortness of breath, that the patients can self-perform and report. Hence, our approach can aid clinicians in providing remote, proactive healthcare service in broad settings