A machine learning approach for predicting mortality in unvaccinated COVID-19 patients using a novel web application

Objectives: During a medical emergency of epidemic proportions, triaging and resource allocation pose a significant challenge. In COVID-19, early identification of high-risk patients was crucial. Several studies have shown that machine learning (ML) enhances predictive accuracy in these patients. Using data from unvaccinated COVID-19 patients, we developed and validated ML models to predict mortality and compared them with traditional statistical methods used in our previous study. Finally, we present a novel web application to predict mortality. Materials and Methods: We conducted a retrospective study of 401 COVID-19 patients admitted between July and December 2020. ML models, including support vector machine (SVM), Random Forest, and XGBoost, were developed using demographic, clinical, and laboratory data. Statistical analysis: Models were evaluated using metrics such as accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). Feature reduction was performed to enhance clinical relevance. A web application was developed for real-time mortality risk predictions. Results: The SVM model performed the best. After feature reduction to five key predictors (clinical severity, IL-6, lactate dehydrogenase- LDH, neutrophil percentage, and neutrophil-to-lymphocyte ratio- NLR), the AUC improved from 0.82 to 0.83, with a sensitivity of 0.83 and specificity of 0.78. This outperformed our previous study results, where individual AUCs ranged from 0.614 to 0.710. The final model achieved a positive predictive value of 0.63 and a negative predictive value of 0.91, indicating high reliability in identifying both high-risk patients and those likely to survive. Conclusions: ML models, particularly SVM, significantly improved COVID-19 mortality prediction. This approach could enhance early risk identification and resource allocation in clinical settings.