COVID-19 Severity Prediction in Patients Based on Anomaly Detection Approach

Ikramov, A.; Adilova, F.; Anvarov, K.; Khadjibaev, A.

Background: As the course of COVID-19 varies dramatically, one of the critical challenges is detecting high-risk patients for early prevention and treatment. Lockdown and treatment regimen in Uzbekistan in Spring, 2020 made it possible to collect a significant amount of patient data, including asymptomatic, mild, and severe cases. Based on the data, we propose several models for the early prediction of severe COVID-19. Results: We compared supervised learning and anomaly detection algorithms for the task of severe illness prediction. We analyzed the performance and evaluated risk factors based on the proposed models. The best performing model achieves an F1 score of 0.928 and a C-index of 0.965 on the test set. We evaluated the robustness of the model and tested it on an external dataset. We used different techniques to build model interpretability. Conclusions: Machine learning methods can help in detecting patients that have a high risk of developing a severe course so that medical experts can start proper treatment earlier and, as a result, decrease the number of critical cases. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.