Related Experiment Videos
Critically ill COVID-19 phenotypes identified by unsupervised clustering: A multicenter retrospective ICU study
Y Zerbib1, M Carpentier1, C Van Derbeken1
1Intensive care department, CHU Amiens-Picardie, Amiens, France.
Heart & Lung : the Journal of Critical Care
|July 29, 2026
Summary
This study identified two distinct patient profiles in severe COVID-19, revealing that corticosteroids improved survival and reduced thrombotic events in one cluster.
Area of Science:
- Critical Care Medicine
- Infectious Diseases
- Data Science in Healthcare
Background:
- Critically ill COVID-19 patients are often treated as a single group, despite evidence of distinct phenotypes.
- Unsupervised clustering has shown potential for identifying these different patient profiles.
Purpose of the Study:
- To identify distinct patient clusters within severe COVID-19.
- To investigate differential therapeutic responses, particularly to corticosteroids, among these identified clusters.
Main Methods:
- A multicenter retrospective observational study analyzed 329 COVID-19 patients.
- Hierarchical Clustering on Principal Components (HCPC) was used to define patient clusters based on laboratory data.
- Demographic, clinical management, mortality, and treatment data were compared across clusters.
Main Results:
- Two distinct COVID-19 patient clusters were identified by HCPC.
- Cluster 2 patients were older, had more comorbidities, experienced greater organ failure, and had higher ICU mortality.
- Corticosteroid use was associated with improved survival and fewer thrombotic events in Cluster 1, but not in Cluster 2.
Conclusions:
- Two distinct COVID-19 patient profiles with differing severity and outcomes were identified using HCPC.
- Corticosteroids demonstrated a survival benefit and reduced thrombotic events specifically in Cluster 1 patients.
