Early Triage of COVID-19 patients exploiting Data-Driven Strategies and Machine Learning Techniques

Abstract

Since the first advent of SARS-CoV-2 in December 2019, Coronavirus disease (COVID-19) is still affecting the world. In the pandemic situation of the novel infectious disease, early detection of COVID-19 infection and severity for febrile respiratory patients is critical for efficient management of the medical system delivery system with limited medical personnel and facilities. Thus, we propose early triage exploiting data-driven strategical methods and machine learning techniques using the data of 5,628 admitted patients provided by Korea Central Disease Control Headquarters and 50 confirmed cases in Korea University Ansan Hospital. We proved validity of our data-driven strategies with machine learning models accuracy by doing 200 experiments and find out the features that affect COVID-19 through various feature selection in each medical inspection step. As a result, Stage 5 shows the results of blood test could affect to classify critical and severe cases obtaining precision of 0.2, 0.03 higher than without blood test results. But Stage 3 without blood test results achieved the highest accuracy of 0.88 showing possibility of early triage system without blood test. In conclusion, our triage system, based on data-driven strategies and machine learning techniques, can help in early detection and triage of COVID-19 patients.