A COVID-19 surveillance platform to monitor risk of infection based on a machine learning model

Abstract

Objective: To develop a platform for daily survey of COVID-19 signs and symptoms in health employees to indicate the need of additional individual diagnostic procedures and to assist institutional planning to prevent the spread of the virus and sustain the hospital operations during the pandemic. Methods: We used information from a recent meta-analysis to simulate datasets of patients with different signs, symptoms and comorbidities to evaluate machine-learning algorithms for each dataset classification. The best performing model identifying COVID-19 from other similar conditions including H1N1 and seasonal influenza was selected as the base model for developing a platform for risk assessment. Results and Conclusion: The platform was deployed for surveillance of 4,200 collaborators from a tertiary hospital on a voluntary basis, but it can be readily adapted for other environments or populational surveillance to assist public authorities devising strategies to prevent the spread of the virus.