Predicting the COVID-19 mortality among Iranian patients using tree-based models: A cross-sectional study

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• COVID-19 poses a significant mortality risk for hospitalized patients.
• Accurate prediction of mortality is crucial for effective patient management and resource allocation.
• Machine learning offers potential for developing predictive models for infectious disease outcomes.

Purpose of the Study:

• To evaluate the efficacy of various machine learning models in predicting COVID-19 mortality.
• To identify key clinical features associated with COVID-19 mortality in hospitalized individuals.
• To compare the predictive performance of different machine learning algorithms.

Main Methods:

• Utilized a dataset of 44,112 hospitalized COVID-19 patients from March 2020 to August 2021.
• Employed Random Forest-Recursive Feature Elimination for feature selection, identifying critical predictors.
• Developed and compared Decision Tree, Random Forest, LightGBM, and XGBoost models using performance metrics including ROC-AUC.

Main Results:

• Key predictors for COVID-19 mortality included age, sex, hypertension, malignancy, pneumonia, cardiac issues, cough, dyspnea, and respiratory diseases.
• XGBoost and LightGBM models demonstrated superior performance, achieving an ROC-AUC of 0.83.
• The best-performing models achieved a sensitivity of 0.77 in predicting mortality.

Conclusions:

• XGBoost, LightGBM, and Random Forest models exhibit high predictive accuracy for COVID-19 mortality.
• These machine learning models show promise for application in clinical settings to aid in patient care.
• External validation is recommended to confirm the generalizability and reliability of these predictive models.