Background: Cholesterol metabolism is dysregulated in sepsis contributing to patient heterogeneity. Subphenotypes displaying lower lipoprotein levels and higher mortality (HYPO) or higher lipoprotein levels and lower mortality (NORMO), were described. We developed a simplified clinical algorithm for bedside subphenotype recognition.
Methods: We analyzed data from four prospective studies (internal dataset), focusing on HYPO and NORMO subphenotypes. A 1,000-tree Random Forest classifier and logistic regression models were built, using clinical features to predict subphenotypes. Performance was evaluated by comparing predictions to actual subphenotypes derived from a machine learning model. The model was applied to an external dataset of 281 patients from three French studies.
Results: The internal cohort consisted of 386 patients (median age 63, 46% female). Four clinical features [hepatic SOFA, cardiovascular SOFA, low (LDL-C) and high-density lipoprotein cholesterol (HDL-C)] predicted HYPO vs. NORMO subphenotypes with an AUC of 0.86, a sensitivity of 0.771 and a specificity of 0.779. In the internal dataset, 28-day mortality for HYPO vs. NORMO patients was 26% vs. 15%, and in the external cohort, 30% vs. 10%. HYPO internal vs. external dataset LDL-C levels were similar (p = 0.99), but HDL-C (p = 0.02) levels were different. Median NORMO internal vs. external dataset LDL-C (p = 0.99) and HDL-C (p = 0.12) levels were similar. HYPO patients had lower LDL-C, HDL-C and total cholesterol than NORMO patients in both internal and external datasets.
Conclusions: Our simplified clinical data algorithm may allow for bedside recognition of septic patients displaying lipid dysregulation subphenotypes. External validation is needed to verify these results.
Keywords: cholesterol; lipid dysregulation; lipids; lipoproteins; sepsis; subphenotyping.
Copyright © 2025 The Author(s). Published by Wolters Kluwer Health, Inc. on behalf of the Shock Society.