Integrating Inflammatory and Endothelial Biomarkers in Machine Learning Models to Predict Sepsis Development in Critically Ill Patients: A Combined Retrospective and Prospective Multicenter Study

Sepsis is a severe, life-threatening condition in which timely recognition improves patient outcomes. Machine learning (ML) methods, usually relying on vital sign data, have shown promise in early sepsis prediction; however, they underrepresent sepsis biomarkers, which are widely used in clinical practice. We aimed to develop and validate interpretable ML models for early sepsis prediction using non-temporal data from the first 24 h of intensive care unit (ICU) admission, and to quantify the incremental value of inflammatory (Interleukin-6 (IL-6), procalcitonin (PCT), C-Reactive Protein (CRP)) and endothelial (soluble Endothelial Protein C Receptor (sEPCR)) biomarkers. A combined cohort of 1257 ICU patients from retrospective and prospective multicenter datasets was analyzed. We evaluated four tree-based models alongside Logistic Regression. We assessed discrimination (Area Under the Receiver Operating Characteristic Curve (AUC); Area Under the Precision-Recall Curve (AUPRC)), calibration, clinical utility (Decision Curve Analysis), and used structured feature ablation and Shapley additive explanations (SHAP) for interpretation. Light Gradient Boosting Machine (LGBM) achieved the best performance on the combined cohort, with an AUC of 0.86 and an AUPRC of 0.62. Moreover, LGBM achieved an AUC of 0.728 and an AUPRC of 0.288 during prospective multicenter validation from the retrospective to the prospective data. Feature ablation demonstrated that inclusion of IL-6, PCT, CRP, and sEPCR significantly improved model performance (AUC: p = 0.003; AUPRC: p = 0.001). SHAP analysis identified IL-6 and PCT as among the most influential predictors, both showing threshold-like nonlinear effects. These findings highlight the critical added value of biologically relevant biomarkers in ML-based clinical decision support.

Authors

Institutions

Publication Details

Journal
Pathogens
Published
2026-11-18
DOI
https://doi.org/10.3390/pathogens15111214
Primary Topic
Sepsis Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Integrating Inflammatory and Endothelial Biomarkers in Machine Learning Models to Predict Sepsis Development in Critically Ill Patients: A Combined Retrospective and Prospective Multicenter Study

Nikolaos S. Lotsios, Charikleia S. Vrettou, Μαρία Θεοδωρακοπούλου, Theofanis Ganitidis et al.
5 links
Pathogens
Sepsis Diagnosis and Treatment
article

Integrating Inflammatory and Endothelial Biomarkers in Machine Learning Models to Predict Sepsis Development in Critically Ill Patients: A Combined Retrospective and Prospective Multicenter Study

Nikolaos S. Lotsios, Charikleia S. Vrettou, Μαρία Θεοδωρακοπούλου, Theofanis Ganitidis, Sophia Pouriki, Ioanna Dimopoulou, Kyriakos Karkoulias, Konstantina Dede, Spyretta Golemati, Chrysi Keskinidou, Olga Kampouropoulou, Sylvia Raftopoulou, Anastasia N. Kotanidou, Alice Georgia Vassiliou, Vassiliki Giannopoulou, Zafeiria Mastora, Apostolos Karalis, Georgios Ioannis Poupouzas, Stavros Papapanagiotou, Ioannis Papapanagiotou, Ioannis Charcharos, Vassilis Issaris, Helen Christopoulou
article en

Abstract

Sepsis is a severe, life-threatening condition in which timely recognition improves patient outcomes. Machine learning (ML) methods, usually relying on vital sign data, have shown promise in early sepsis prediction; however, they underrepresent sepsis biomarkers, which are widely used in clinical practice. We aimed to develop and validate interpretable ML models for early sepsis prediction using non-temporal data from the first 24 h of intensive care unit (ICU) admission, and to quantify the incremental value of inflammatory (Interleukin-6 (IL-6), procalcitonin (PCT), C-Reactive Protein (CRP)) and endothelial (soluble Endothelial Protein C Receptor (sEPCR)) biomarkers. A combined cohort of 1257 ICU patients from retrospective and prospective multicenter datasets was analyzed. We evaluated four tree-based models alongside Logistic Regression. We assessed discrimination (Area Under the Receiver Operating Characteristic Curve (AUC); Area Under the Precision-Recall Curve (AUPRC)), calibration, clinical utility (Decision Curve Analysis), and used structured feature ablation and Shapley additive explanations (SHAP) for interpretation. Light Gradient Boosting Machine (LGBM) achieved the best performance on the combined cohort, with an AUC of 0.86 and an AUPRC of 0.62. Moreover, LGBM achieved an AUC of 0.728 and an AUPRC of 0.288 during prospective multicenter validation from the retrospective to the prospective data. Feature ablation demonstrated that inclusion of IL-6, PCT, CRP, and sEPCR significantly improved model performance (AUC: p = 0.003; AUPRC: p = 0.001). SHAP analysis identified IL-6 and PCT as among the most influential predictors, both showing threshold-like nonlinear effects. These findings highlight the critical added value of biologically relevant biomarkers in ML-based clinical decision support.

PathogensVol. 15(11)
National Technical University of Athens (GR), National and Kapodistrian University of Athens (GR), Space Hellas (Greece) (GR), KAT General Hospital of Attica (GR), Evangelismos Hospital (GR), Sotiria General Hospital (GR)
Good health and well-being
Sepsis Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon