Multidimensional Machine Learning for Assessing Parameters Associated With COVID-19 in Vietnam: Validation Study

Area of Science:

• Medical Informatics
• Machine Learning
• Epidemiology

Background:

• Machine learning (ML) utilizes artificial intelligence algorithms to analyze large datasets, identify patterns, and perform autonomous tasks.
• Emerging diseases like COVID-19 generate substantial data, making ML a valuable tool for analysis and prediction.
• Quantifying and modeling relevant parameters is crucial for understanding disease dynamics and severity.

Purpose of the Study:

• To determine the preclinical characteristics of COVID-19.
• To establish cumulative cutoff values and risk ratios (RRs) for disease severity.
• To identify factors associated with COVID-19 severity using unidimensional and multidimensional analyses in 2173 SARS-CoV-2 patients.

Main Methods:

• Analysis of 2173 patients categorized into mild, moderate, and severe COVID-19 groups.
• Utilized correlation tests, relative risk, and RR to identify significant parameters.
• Employed hierarchical cluster analysis, k-means, and network analysis for parameter classification and visualization.

Main Results:

• COVID-19 severity correlated significantly with age, chest x-ray scoring index, neutrophil percentage and quantity, C-reactive protein, and lymphocyte ratio.
• Albumin showed a protective effect (negative correlation) in moderate-severe cases.
• Network analysis revealed ferritin and age as primary correlates of severity in mild-moderate cases, while ferritin, fibrinogen, and albumin were key in moderate-severe cases.

Conclusions:

• This multidimensional ML study identified key clinical markers for COVID-19 severity in a Vietnamese population.
• Findings provide potential reference markers for improved surveillance and diagnostic management of COVID-19.
• The study highlights the utility of machine learning in dissecting complex disease patterns.