Clinical severity prediction of COVID-19 admitted patients in Spain: SEMI and REDISSEC cohorts

This report addresses, from a machine learning perspective, a multi-class classification problem to predict the first deterioration level of a COVID-19 positive patient at the time of hospital admission. Socio-demographic features, laboratory tests and other measures are taken into account to learn the models. Our output is divided into 4 categories ranging from healthy patients, followed by patients requiring some form of ventilation (divided in 2 cate-gories) and finally patients expected to die. The study is conducted thanks to data provided by Sociedad Española de Medicina Interna (SEMI) and Red de Investigación en Servicios de Salud de Enfermedades Crónicas (REDISSEC). Results show that logistic regression is the best method for identifying patients with clinical deterioration.