Does COVID-19 Clinical Status Associate with Outcome Severity? An Unsupervised Machine Learning Approach for...

Area of Science:

• Medical Informatics
• Computational Biology
• Epidemiology

Background:

• The COVID-19 pandemic has caused significant mortality and overwhelmed healthcare systems.
• Existing machine learning (ML) applications for COVID-19 often rely on supervised methods, requiring extensive labeled data.
• Current understanding of COVID-19's clinical manifestations and mortality predictors remains incomplete.

Purpose of the Study:

• To apply an unsupervised clustering approach to identify distinct patient groups based on clinical variables.
• To explore correlations between clinical characteristics, symptoms, and outcomes in hospitalized COVID-19 patients without prior assumptions.
• To investigate potential novel insights into COVID-19 patient stratification and mortality risk.

Main Methods:

• Utilized a dataset of 268 hospitalized COVID-19 patients with 40 clinical variables.
• Applied dimensionality reduction techniques: Principal Component Analysis (PCA) for numerical data and Multiple Correspondence Analysis (MCA) for categorical data.
• Employed Gaussian Mixture Models (GMM) with Bayesian Information Criterion (BIC) to determine optimal patient clusters.

Main Results:

• Identified four distinct clusters of patients exhibiting similar clinical profiles.
• A significant finding was a cluster of patients presenting with asymptomatic disease who had a 23.8% mortality rate.
• This highlights a potential disconnect between perceived symptom severity and actual patient outcomes.

Conclusions:

• Unsupervised learning can reveal hidden patterns in complex disease data, such as COVID-19.
• The high mortality in the asymptomatic cluster necessitates a re-evaluation of how COVID-19 severity and risk are assessed.
• Further research is warranted to understand the factors contributing to poor outcomes in seemingly asymptomatic COVID-19 patients.