Predict In-Hospital Mortality in COVID-19 Patients with a Machine Learning Webpage Tool Based on Clinical Laboratory Data: An Exploratory Research

Background The death risk induced by COVID-19 posed tremendous pressure on medical services, resulting in a shortage of critical care resources and a heavy disease burden. Developing predictive algorithms based on local patient data could be more effective for supporting decisions on the level of healthcare required. Methods Laboratory test results of the COVID-19 patients were collected. Five machine learning (ML) approaches were applied to develop a predictive algorithm for in-hospital mortality. Results Our cohort enrolled 602 patients with COVID-19 and 40 laboratory tests after data preprocessing. The RF-based model was chosen as the predictive algorithm, namely MOP@COVID. MOP@COVID performed well in the training set and validation set. MOP@COVID yielded a sensitivity of 0.818, a specificity of 0.987, an accuracy of 0.973, and an AUC of 0.958 in the external validation cohort. A webpage tool of MOP@COVID was developed to facilitate clinical application. Conclusions The MOP@COVID used routine laboratory test results at hospital admission and could predict the risk of in-hospital mortality in COVID-19 patients. With the webpage tool, MOP@COVID could provide helpful information to clinical doctors and healthcare providers in rural areas.