Pervasive intelligent models to predict the outcome of COVID-19 patients

Abstract

Nowadays, there is an increasing need to understand the behavior of COVID-19. After the Directorate-General of Health of Portugal made available the infected patient’s data, it became possible to analyze it and gather some conclusions, obtaining a better understanding of the matter. In this context, the project developed—ioCOVID19—Intelligent Decision Support Platform aims to identify patterns and develop intelligent models to predict and support clinical decisions. This article explores which typologies are associated with different outcomes to help clinicians fight the virus with a decision support system. So, to achieve this purpose, classification algorithms were used, and one target was studied—Patients outcome, that is, to predict if the patient will die or recover. Regarding the obtained results, the model that stood out is composed of scenario s4 (composed of all comorbidities, symptoms, and age), the decision tree algorithm, and the oversampling sampling method. The obtained results by the studied metrics were (in order of importance): Sensitivity of 95.20%, Accuracy of 90.67%, and Specificity of 86.08%. The models were deployed as a service, and they are part of a clinical decision support system that is available for authorized users anywhere and anytime.