ABSTRACT
The coronavirus disease 2019 (COVID-19) pandemic is challenging healthcare systems worldwide. The prediction of disease prognosis has a critical role in confronting the burden of COVID-19. We aimed to investigate the feasibility of predicting COVID-19 patient outcomes and disease severity based on clinical and hematological parameters using machine learning techniques. This multicenter retrospective study analyzed records of 485 patients with COVID-19, including demographic information, symptoms, hematological variables, treatment information, and clinical outcomes. Different machine learning approaches, including random forest, multilayer perceptron, and support vector machine, were examined in this study. All models showed a comparable performance, yielding the best area under the curve of 0.96, in predicting the severity of disease and clinical outcome. We also identified the most relevant features in predicting COVID-19 patient outcomes, and we concluded that hematological parameters (neutrophils, lymphocytes, D-dimer, and monocytes) are the most predictive features of severity and patient outcome.
