Prognostic Value of the Endothelial Activation and Stress Index (EASIX) in Critically Ill Patients With Ischemic Stroke: A MIMIC‐IV Database Analysis

Figures

• 

  FIGURE 1. Study flowchart and patient selection. Flow diagram illustrating the detailed screening and selection process of critically ill patients with ischemic stroke included in the final analysis. EASIX, endothelial activation and stress index.

• 

  FIGURE 2. Kaplan–Meier cumulative survival analyses according to lnEASIX median and tertiles. (A–C) Kaplan–Meier curves for (A) 28‐day, (B) 90‐day, and (C) 180‐day cumulative all‐cause mortality from admission, stratified by the median lnEASIX value (low vs. high). (D–F) Kaplan–Meier curves for (D) 28‐day, (E) 90‐day, and (F) 180‐day cumulative all‐cause mortality from admission, stratified by lnEASIX tertiles (Tertile 1 [lowest] to Tertile 3 [highest]). All survival times were calculated from the day of ICU admission. Differences in survival probabilities among the stratified groups were assessed using the log‐rank test.

• 

  FIGURE 3. Restricted cubic spline (RCS) analyses of lnEASIX and mortality outcomes in critically ill patients with ischemic stroke. (A) Association between lnEASIX and in‐hospital mortality. (B) Association between lnEASIX and 28‐day all‐cause mortality after hospital discharge. (C) Association between lnEASIX and 90‐day all‐cause mortality after hospital discharge. (D) Association between lnEASIX and 180‐day all‐cause mortality after hospital discharge. Models were adjusted for covariates included in the fully adjusted model.

• 

  FIGURE 4. Receiver operating characteristic (ROC) curves of lnEASIX for mortality prediction. (A) ROC curve of lnEASIX for predicting in‐hospital all‐cause mortality. (B) ROC curve of lnEASIX for predicting 28‐day all‐cause mortality after hospital discharge. (C) ROC curve of lnEASIX for predicting 90‐day all‐cause mortality after hospital discharge. (D) ROC curve of lnEASIX for predicting 180‐day all‐cause mortality after hospital discharge.

• 

  FIGURE 5. Calibration curves for internal validation of the prognostic model predicting 28‐day, 90‐day, and 180‐day survival. (A–C) Calibration plots comparing the nomogram‐predicted probabilities of survival with the actual observed survival (Kaplan–Meier estimates) at (A) 28 days, (B) 90 days, and (C) 180 days after admission. The x‐axis represents the predicted survival probability, while the y‐axis represents the actual survival rate. The red dashed line denotes a theoretically perfect prediction (ideal line, where predicted probability exactly matches actual outcome). The solid black line represents the apparent predictive accuracy of the model. The blue solid line with error bars illustrates the bias‐corrected calibration performance, rigorously calculated using 1000 bootstrap resamples for internal validation. The close alignment of the bias‐corrected lines with the ideal 45‐degree reference lines across all three timeframes indicates an optimal agreement and highly accurate absolute risk estimation, without significant overestimation or underestimation.

• 

  FIGURE 6. Reclassification improvement analyses (NRI and IDI) for the integration of lnEASIX with established intensive care scoring systems. (A–D) Scatter plots illustrating the changes in predicted mortality risk when lnEASIX is added to baseline clinical models (SOFA, LODS, and GCS) for predicting (A) 28‐day, (B) in‐hospital, (C) 90‐day, and (D) 180‐day mortality. The x‐axis represents the predicted risk calculated by the baseline clinical score alone, while the y‐axis represents the updated predicted risk after incorporating lnEASIX. Each dot represents an individual patient: red dots indicate patients who died (events), and blue dots indicate patients who survived (non‐events). The dashed black diagonal reference line represents no change in predicted risk. The net reclassification improvement (NRI) and integrated discrimination improvement (IDI) metrics, along with their respective p‐values, are displayed above each subplot. The consistently positive NRI and IDI values (all p < 0.05) demonstrate that the addition of lnEASIX significantly reclassifies patients into more accurate risk categories, thereby substantially enhancing the overall prognostic performance of standard ICU scoring…

• 

  FIGURE 7. Subgroup analyses of the association between continuous lnEASIX and mortality outcomes. Forest plots illustrating the subgroup analyses for (A) 28‐day all‐cause mortality, (B) in‐hospital mortality, (C) 90‐day all‐cause mortality, and (D) 180‐day all‐cause mortality in critically ill patients with ischemic stroke. Hazard ratios (HRs) and 95% confidence intervals (CIs) for a per‐unit increase in lnEASIX were estimated using Cox proportional hazards regression models across various clinical strata. Interaction tests (p for interaction) were performed to assess effect modification by baseline characteristics. The consistently positive HRs (> 1) across the vast majority of subgroups demonstrate the robust and independent prognostic value of lnEASIX.