Identifying the Phenotypes of Sepsis that will Benefit from Red Blood Cell Transfusion Using Unsupervised Cluster Analysis
Related Experiment Video
Updated: Sep 24, 2025
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
Background:
Sepsis is a heterogeneous syndrome. Previous studies have shown controversial results of the effects of red blood cell transfusion (RBC) on the clinical outcomes of septic patients. This study aimed to identify the phenotypes of sepsis that will benefit from RBC transfusion.
Methods:
Clinical data were extracted from the Medical Information Mart for Intensive Care III database. The study population included adult (age ≥ 18 years) septic patients with moderate non-bleeding anemia (hemoglobin ≤ 10 g/dL) within 24 hours after admission to the intensive care unit (ICU) between 2001 and 2012. After data preprocessing, partitioning around medoids function was used for unsupervised cluster analysis. We used Kaplan-Meier survival analysis and multivariable Cox proportional hazard models to explore the relationship between RBC transfusion and mortality.
Results:
In total, 6,821 septic patients with moderate non-bleeding anemia within 24 hours after ICU admission, and 3,874 patients (56.8%) received RBC transfusion during their stay in the ICU. Using unsupervised cluster analysis, we identified three phenotypes of septic patients with moderate non-bleeding anemia: cluster A (n = 1,835) was characterized by advanced age and heart issues; cluster B (n = 3,043) was characterized by mild disease and relatively high hemoglobin levels; and cluster C (n = 1,943) was characterized by severe disease, low mean arterial pressure, bloodstream infection, coagulopathy, high lactate levels, and high mortality. Only for patients in cluster C, RBC transfusion exhibited protective effects in terms of the 14-day [hazard ratio (HR), 0.50; 95% confidence interval (CI), 0.41 - 0.61; p < 0.001], 28-day (HR, 0.61; 95% CI, 0.51 - 0.72; p < 0.001), and 90-day (HR, 0.67; 95% CI, 0.58 - 0.78; p < 0.001) mortality after adjusting the confounding variables.
Conclusions:
Utilizing unsupervised cluster analysis, we identified three phenotypes of septic patients with moderate non-bleeding anemia who had different responses to RBC transfusion. In the future, randomized controlled trials about prognostic outcomes of RBC transfusions can focus on the specific phenotype of sepsis.
