Advantages of Metabolomics-Based Multivariate Machine Learning to Predict Disease Severity: Example of COVID.

Area of Science:

• Biochemistry
• Computational Biology
• Medical Diagnostics

Background:

• The COVID-19 pandemic overwhelmed healthcare systems, necessitating efficient patient triage.
• Optimizing resource allocation requires accurate early prediction of disease severity.

Purpose of the Study:

• To evaluate if high definition metabolomics and machine learning (ML) improve COVID-19 patient prognostication and triage.
• To compare the predictive performance of metabolomics-enhanced ML models against standard clinical parameters.

Main Methods:

• High resolution mass spectrometry was used to obtain metabolomics profiles from 64 COVID-19 patients.
• Machine learning algorithms were developed integrating clinical data and metabolomics profiles.
• Model performance was assessed using Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• Standard clinical parameters predicted severity (need for mechanical ventilation) with an AUC of 0.85.
• Integrating metabolomics data with ML substantially improved prediction performance (AUC = 0.92).
• Key clinical predictors included SpO2, respiratory rate, Horowitz quotient, and age.

Conclusions:

• Metabolomics combined with ML significantly enhances the accuracy of COVID-19 severity prediction and patient triage.
• This integrated approach can identify novel biological markers for improved diagnostics.
• The technique is clinically deployable and supports the advancement of personalized medicine.