Dynamic evaluation of a COVID-19 death prediction model using Extreme Gradient Boosting Predictive Model

Area of Science:

• Infectious Diseases
• Medical Informatics
• Biostatistics

Background:

• The COVID-19 pandemic presents evolving challenges due to new variants and varying vaccination rates.
• Accurate prediction of severe COVID-19 outcomes is crucial for effective patient management.
• Existing models require updates to account for temporal changes in disease dynamics.

Purpose of the Study:

• To develop and validate a predictive model for COVID-19 mortality in hospitalized patients.
• To identify key clinical and laboratory predictors of COVID-19 mortality.
• To incorporate dynamic pandemic factors into disease severity assessment.

Main Methods:

• Utilized the Extreme Gradient Boost (XGBoost) machine learning model.
• Integrated data from electronic medical records, vaccination databases, and SARS reports.
• Correlated model predictions with laboratory results, vaccination status, comorbidities, and clinical signs/symptoms.

Main Results:

• The XGBoost model achieved a high predictive performance with an Area Under the Curve (AUC) of 96.4%.
• Significant predictors included body temperature, blood pressure, respiratory rate, heart rate, urea, magnesium, sodium, and C-reactive protein.
• The model demonstrated effectiveness in predicting mortality based on admission data.

Conclusions:

• XGBoost is a robust tool for predicting COVID-19 mortality in hospitalized individuals.
• Key clinical and laboratory variables are vital for accurate mortality prediction.
• The model provides a valuable approach for assessing disease severity in the evolving pandemic context.