Background:
Inconsistent results from COVID-19 studies raise the issue of patient heterogeneity.
Objective:
The objective of this study was to identify homogeneous subgroups of patients (clusters) using baseline characteristics including inflammatory biomarkers and the extent of lung parenchymal lesions on CT, and to compare their outcomes.
Design:
Retrospective single-center study.
Setting:
Medical ICU of the University Hospital of Clermont-Ferrand, France.
Patients:
All consecutive adult patients aged greater than or equal to 18 years, admitted between March 20, 2020, and August 31, 2021, for COVID-19 pneumonia.
Interventions:
Characteristics at baseline, during ICU stay, and outcomes at day 60 were recorded. On the chest CT performed at admission the extent of lung parenchyma lesions was established by artificial intelligence software.
Measurements and Main Results:
Clusters were determined by hierarchical clustering on principal components using principal component analysis of admission characteristics including plasma interleukin-6, human histocompatibility leukocyte antigen-DR expression rate on blood monocytes (HLA-DR) monocytic-expression rate (mHLA-DR), and the extent of lung parenchymal lesions. Factors associated with day 60 mortality were investigated by univariate survival analysis. Two hundred seventy patients were included. Four clusters were identified and three were fully described. Cluster 1 (obese patients, with moderate hypoxemia, moderate extent of lung parenchymal lesions, no inflammation, and no down-regulation of mHLA-DR) had a better prognosis at day 60 (hazard ratio [HR] = 0.27 [0.15–0.46], p < 0.01), whereas cluster 2 (older patients with comorbidities, moderate extent of lung parenchyma lesions but significant hypoxemia, inflammation, and down-regulation of mHLA-DR) and cluster 3 (patients with severe parenchymal disease, hypoxemia, inflammatory reaction, and down-regulation of mHLA-DR) had an increased risk of mortality (HR = 2.07 [1.37–3.13], p < 0.01 and HR = 1.52 [1–2.32], p = 0.05, respectively). In multivariate analysis, only clusters 1 and 2 were independently associated with day 60 death.
Conclusions:
Three clusters with distinct characteristics and outcomes were identified. Such clusters could facilitate the identification of targeted populations for the next trials.