OBJECTIVES: To develop a prediction model for in-hospital mortality in patients diagnosed with COVID-19 receiving treatment with remdesivir, using machine learning (ML) approaches, and to find the best predictor

METHODS: We analyzed 1,362 hospitalized patients with COVID-19 who had received therapy with remdesivir at three hospitals in Yogyakarta. Several supervised ML algorithms were used to construct a model to analyze and predict COVID-19 mortality based on demographic, clinical, and laboratory data. Using WEKA ML software, the Decision Tree (DT), Random Forest (RF), Logistic Regression, Naïve Bayes, and AdaBoost algorithms were applied to the prediction model based on sensitivity, receiver operating characteristic (ROC), accuracy, and precision.

RESULTS: The best ML was RF (accuracy 82,8%, sensitivity 82,9%, precision 82,8%, and ROC 89,1%). The top five predictors were myocardial infarction (OR 97.351,95% CI [35.418–267.582], p<0.001), neutrophils (OR 5.5099, 95% CI [3.726–8.144], p<0,001), SPO2 (OR 4.506, 95% CI [3.468–5.854], p<0.001), and septic shock (OR 4.041, 95% CI [2.105–7.755], p<0.001).

CONCLUSIONS: This model can accurately predict in-hospital mortality and identify the independent predictors of COVID-19 patients. It can be applied in the case of a future pandemic