What question did this study set out to answer?

This analysis aims to explore the relationship between s-EASIX trajectories and 30-day mortality in sepsis patients.

February 28, 2026Open Access

Machine learning analysis of s-EASIX for predicting 30-day mortality in sepsis patients from MIMIC-IV

Q: What does this research mean for the field?

Dynamic increasing patterns of the s-EASIX trajectory correlate with elevated 30-day mortality in sepsis patients, indicating that persistent endothelial dysfunction raises the risk of unfavorable prognosis. Novelty: ClaimNovelty.NOVEL_FINDING. Consensus alignment: ConsensusAlignment.CHALLENGES_CONSENSUS.

Key Points

This analysis aims to explore the relationship between s-EASIX trajectories and 30-day mortality in sepsis patients.
Retrospective cohort analysis using MIMIC-IV database
Investigation of s-EASIX trajectory patterns in 8113 sepsis patients
Application of Kaplan–Meier curves and Cox proportional hazards regression
Use of machine learning models for risk prediction and analysis of SHAP values
Identified five classes of s-EASIX trajectories influencing mortality
30-day mortality significantly increased in Class 4 (HR 1.79) and Class 5 (HR 2.53)
LightGBM model achieved an AUC of 0.842 in validation
Dynamic s-EASIX trajectory patterns correlate with elevated mortality risk
The model provided interpretable results for patient risk stratification

Abstract

Endothelial dysfunction is an important risk factor for the progression of sepsis. The simplified endothelial activation and stress index (s-EASIX) serves as an indirect measure of endothelial activation, whose dynamic changes have an unclear association with prognosis in sepsis. Therefore, we conducted this study to investigate the association between clinical subphenotypes indicated by s-EASIX trajectories and 30-d mortality in sepsis. Based on MIMIC-IV v3.1, the association of s-EASIX dynamic trajectories with 30-d mortality in sepsis was investigated in this retrospective cohort analysis. The prognostic value of trajectory patterns was verified by Kaplan–Meier curves, multivariate regression, and subgroup analyses. Machine learning models incorporating s-EASIX were established, and the weights of contribution of key variables to model decision-making were revealed using SHAP values. This study screened 8113 sepsis patients and identified five classes of s-EASIX trajectories. Cox proportional hazards regression revealed that the 30-d mortality significantly rose in Class 4 (Mid-Increasing) (HR 1.79, 95% CI 1.51–2.11) and Class 5 (High-SlowDecline) (HR 2.53, 95% CI 2.06–3.11), and it was comparable between Class 3 (High-FastDecline) and Class 1/2 (Low-Stable and Mid-Stable). The independent prognostic value of trajectory patterns was verified by multivariate regression, and the HR values for high-risk trajectories remained within 2.53–5.95 after adjusting for demographics and confounders. According to model assessment, LightGBM exhibited superior performance in the validation set (AUC 0.842, 95% CI 0.818–0.866), and its predictive reliability was proven by the Brier score (0.014 in the validation set). Moreover, we analyzed the SHAP values and identified the s-EASIX trajectory as the core variable; the model served as an interpretable tool for risk stratification and early intervention in high-risk sepsis patients. The dynamic increasing pattern of the s-EASIX trajectory correlates with the elevation of 30-d mortality in sepsis, suggesting that persistent endothelial dysfunction raises the risk of unfavorable prognosis.

Bookmark

View Full Paper

Cite This Study

Kong et al. (Thu,) studied this question.

synapsesocial.com/papers/69a286240a974eb0d3c00e46 https://doi.org/https://doi.org/10.1038/s41598-026-40400-1

Bookmark

View Full Paper