From the paper: Mahdavi et al., A machine learning based exploration of COVID-19mortality risk
Early prediction of patient mortality risks during a pandemic can decrease mortality by assuring efficientresource allocation and treatment planning. This study aimed to develop and compare prognosis predic-tion machine learning models based on invasive laboratory and noninvasive clinical and demographicdata from patients’ day of admission. Three Support Vector Machine (SVM) models were developedand compared using invasive, non-invasive, and both groups. The results suggested that non-invasivefeatures could provide mortality predictions that are similar to the invasive and roughly on par withthe joint model. Feature inspection results from SVM-RFE and sparsity SVM displayed that comparedwith the invasive mode, the non-invasive model can provide better performances with fewer numberof features, pointing to the presence of high predictive information contents in several non-invasivefeatures, including SPO2, age, and cardiovascular disorders. Furthermore, while the invasive modelwas able to provide better mortality predictions for the imminent future, non-invasive features displayedbetter performance for more distant expiration intervals. Early mortality prediction using non-invasivemodels can give us insights as to where and with whom to intervene. Combined with novel technologies,such as wireless wearable devices, these models can create powerful frameworks for various medicalassignments and patient traige.