Decoding Clinical Biomarker Space of COVID-19: Exploring Matrix Factorization-based Feature Selection Methods.

Area of Science:

• Biomedical Informatics
• Clinical Medicine
• Data Science

Background:

• Effective patient triage is crucial for managing complex diseases like COVID-19 during pandemics.
• Current methods relying on clinical presentation and patient characteristics need more accurate biomarkers for optimal decision-making.
• There is a critical need for improved clinical indicators to predict prognosis and morbidity in COVID-19 patients.

Approach:

• Developed a machine learning model integrating Feature Selection and Prognosis Classification schemes.
• Employed Matrix Factorization (MF)-based methods for feature selection from blood test data.
• Utilized the Random Forest algorithm for classifying patient prognosis based on selected clinical indicators.

Key Points:

• Identified specific clinical indicators from blood tests that predict poor prognosis in COVID-19 patients.
• Arterial Blood Gas (ABG) O2 Saturation emerged as a significant predictor of patient outcomes.
• C-Reactive Protein (CRP) was also identified as a key biomarker for determining poor prognosis.

Conclusions:

• The machine learning approach successfully identified critical biomarkers for COVID-19 prognosis.
• Arterial Blood Gas O2 Saturation and C-Reactive Protein are vital for accurate patient triage.
• This work facilitates the development of quantitative clinical management systems for COVID-19 and similar diseases.