Hybrid Modeling for Predicting Inpatient Treatment Outcome: COVID-19 Case

This study presents two methods to support the treatment process of inpatients with COVID-19. The first method is designed to predict treatment outcomes; this method is based on machine learning models and probabilistic graph models of patient clustering. The method demonstrates high quality in terms of predictive models, and the structure of the graph model is supported by knowledge from practical medicine and other studies. The method is used as a basis for finding the optimal intervention plan for severe patients. This plan is a set of interventions for patients that are optimal in terms of minimizing the probability of mortality. We tested the method for critically ill patients (item 4.5) and for 30% of all patients with lethal outcomes the methods found an intervention plan that leads to recovery as a treatment outcome as predicted. Both methods show high quality, and after validation by physicians, this method can be used as part of a decision support system for medical professionals working with COVID-19 patients.