Mortality Prediction for COVID-19 Patients: Methods and Potential

The pandemic spread of Coronavirus leads to increased burden on healthcare services worldwide. Experience shows that required medical treatment can reach limits at local clinics and fast and secure clinical assessment of the disease severity becomes vital. Biomarkers are regularly determined for intensive care patients. Machine learning tools can be used to select appropriate biomarkers in order to estimate the state of health and to predict patient mortality risk. Transparent prediction models allow further statements on the properties and development of the biomarkers in connection with specific health conditions of the intensive care patients. In this work, alternative and advanced model approaches (Support Vector Machine, naive Bayes, Fuzzy system) are compared with models proposed in literature. In addition, aspects such as gender of patients and changes in biomarkers over time are included in the modeling. An artificial neural network (SOM) is used for selecting the biomarkers. A statistical analysis of the biomarkers reveals their values and changes in the critical state of the patients. In a model comparison, a Sugeno-type Fuzzy predictor achieved the best results for health assessment and decision support. The Fuzzy system delivers continuous output values instead of binary decisions and thus doubtful cases can be assigned to a rejection class. An extended Fuzzy model takes into account the patient’s gender and the trend in key features over time and thus provides excellent results with an accuracy better than 98% with the training data. However, this could not be finally verified due to the lack of suitable test data. The generation and training of all models was fully automatic with Matlab© tools and without additional adjustment.