A comparison of machine learning algorithms in predicting COVID-19 prognostics

Area of Science:

• Computational biology
• Medical informatics
• Machine learning in healthcare

Background:

• Machine learning (ML) is crucial for developing prognostic and diagnostic models to aid clinical decision-making.
• Predicting intensive care needs, intubation, and mortality risk in COVID-19 patients is vital for resource allocation and patient management.

Purpose of the Study:

• To evaluate eight supervised ML algorithms for predicting critical outcomes in COVID-19 patients.
• To identify key features influencing prognostic predictions.
• To compare the performance of different ML algorithms in COVID-19 prognosis.

Main Methods:

• Utilized two datasets: one with demographics and clinical data (n=11,712), and another including blood test results (n=602).
• Developed and compared eight supervised ML algorithms, including Extra Tree and CatBoost classifiers.
• Assessed model performance using Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• All prognostic models achieved an AUROC exceeding 0.92.
• Extra Tree and CatBoost classifiers demonstrated superior performance with AUROC values over 0.94.
• Key predictive features identified include C-reactive protein, lymphocyte ratio, lactic acid, and serum calcium levels.

Conclusions:

• Supervised ML, particularly tree-based algorithms, offers significant value in predicting COVID-19 prognosis.
• Accurate prognostic models can support clinical decision-making and improve patient outcomes.
• Biomarkers like C-reactive protein and blood cell counts are critical indicators for COVID-19 severity.