Predicting COVID-19 Outcomes: Machine Learning Predictions Across Diverse Datasets

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• COVID-19 has caused a global health crisis.
• Predicting patient mortality is crucial for managing the pandemic.
• Machine learning (ML) offers potential for prognostic prediction.

Purpose of the Study:

• To predict mortality risk in COVID-19 patients using ML algorithms.
• To evaluate the effectiveness of different datasets for prediction.
• To identify key predictors of mortality.

Main Methods:

• Retrospective analysis of 404 COVID-19 patients.
• Utilized fever, oxygen saturation, lab results, CT findings, and comorbidities.
• Applied various ML models including gradient boosting, random forest, and XGBoost.
• Used Synthetic Minority Oversampling Technique for data balancing.

Main Results:

• The gradient boosting model achieved 98.4% accuracy in mortality prediction.
• A dataset combining CT parenchyma score, vessel diameters, and lab results was most effective.
• Specific CT and laboratory findings were identified as strong predictors.

Conclusions:

• Patient prognosis in COVID-19 can be accurately predicted.
• Thorax CT scans and laboratory findings are valuable prognostic indicators.
• ML models provide a powerful tool for COVID-19 mortality risk assessment.