COVID-19 Outcome Prediction by Integrating Clinical and Metabolic Data using Machine Learning Algorithms

Area of Science:

• Computational biology
• Medical informatics
• Biomedical data science

Background:

• COVID-19, caused by SARS-CoV-2, has led to millions of deaths globally.
• Machine learning (ML) offers potential for identifying high-risk individuals and predicting disease outcomes.

Purpose of the Study:

• To evaluate ML models for predicting COVID-19 outcomes.
• To compare models using clinical data versus combined clinical and metabolic data.

Main Methods:

• 154 patients were analyzed with two data profiles: basic (clinical/demographic) and extended (clinical/demographic/metabolomic/immunological).
• Feature selection was performed using a genetic algorithm (GA).
• Random forest models were trained and tested for predicting COVID-19 severity stages.

Main Results:

• The extended profile model showed higher utility in early disease stages.
• Clinical data models were more effective for predicting severe illness, critical illness, and death.
• Key predictive variables identified by ML include trimethylamine N-oxide, lipid mediators, and neutrophil/lymphocyte ratio.

Conclusions:

• Machine learning and genetic algorithms provide robust models for predicting COVID-19 outcomes across various severity grades.
• Integration of diverse data types enhances predictive capabilities for different disease stages.