Phenotype clustering of hospitalized high-risk patients with COVID-19 - a machine learning approach within the...

Area of Science:

• Cardiology
• Infectious Diseases
• Machine Learning

Background:

• COVID-19 patients with cardiovascular disease (CV) or risk factors (RF) have varied prognoses.
• Predicting outcomes in this heterogeneous group requires advanced analytical methods.

Purpose of the Study:

• To phenotype hospitalized COVID-19 patients with CV disease/RF using unsupervised machine learning (ML).
• To identify distinct patient clusters within the PCHF-COVICAV registry for outcome prediction.

Main Methods:

• Unsupervised ML (K-medoids algorithm) applied to 458 patients from the PCHF-COVICAV registry.
• Analysis incorporated 46 baseline variables including demographics, clinical status, comorbidities, and biochemical data.

Main Results:

• Three distinct patient clusters were identified.
• Cluster 1: Predominantly women with fewer comorbidities/RF.
• Cluster 2: Predominantly men with non-CV conditions and milder symptoms.
• Cluster 3: Predominantly men with severe CV disease, high RF, inflammation, organ dysfunction, and highest 6-month mortality.

Conclusions:

• ML successfully delineated three clinical clusters among COVID-19 patients with CV disease/RF.
• A specific cluster of males with severe CV disease, heart failure, multiple RF, and inflammation exhibited a significantly poorer prognosis.