Identifying and Validating Prognostic Hyper-Inflammatory and Hypo-Inflammatory COVID-19 Clinical Phenotypes Using...

Area of Science:

• Infectious Diseases
• Immunology
• Data Science

Background:

• COVID-19 presents with significant clinical and biological heterogeneity.
• Understanding diverse disease trajectories requires identifying distinct patient phenotypes.
• Phenotyping can improve clinical practice and trial design for COVID-19.

Purpose of the Study:

• To identify distinct COVID-19 subphenotypes using clinical data.
• To develop a machine learning model for accurate subphenotype classification.
• To determine key clinical variables for predicting COVID-19 subphenotypes and outcomes.

Main Methods:

• Employed k-prototypes clustering on 50 clinical variables from 1376 adult COVID-19 patients.
• Utilized machine learning algorithms to identify key classifier variables for phenotype recognition.
• Applied the AdaBoost model for subphenotype prediction and performance evaluation.

Main Results:

• Identified two distinct subphenotypes: Hypo-inflammatory (59.9%) and Hyper-inflammatory (40.1%).
• Hypo-inflammatory patients had lower mortality and shorter hospital stays; Hyper-inflammatory patients were older, male, with higher mortality and organ dysfunction.
• The AdaBoost model achieved high accuracy (0.975) in subphenotype prediction, with "CRP", "IL-2R", and "D-dimer" as key predictors.

Conclusions:

• Two COVID-19 phenotypes were identified, accurately classifiable by machine learning models.
• The identified subphenotypes can guide risk stratification and clinical management strategies.
• Key biomarkers like CRP, IL-2R, and D-dimer are crucial for predicting subphenotypes and patient outcomes.