Association Between the Endothelial Activation and Stress Index and All-Cause Mortality in Critically Ill Patients with Stage 4 Cardiovascular-Kidney-Metabolic Syndrome: A Retrospective Cohort Study Based on MIMIC-IV and External Validation in eICU-C

Abstract Objective To investigate the association between the endothelial activation and stress index (EASIX) measured within 24 hours of admission to the intensive care unit (ICU) and 30-, 90- and 365-day all-cause mortality in patients with stage 4 cardiovascular-kidney-metabolic (CKM) syndrome, and to characterize the shape of this association as it changes with time since admission. Methods Retrospective cohort study; data were obtained from MIMIC-IV. Adults who met stage 4 CKM criteria at ICU admission, survived to 24 hours after admission, and had EASIX measurable within 24 hours were included. The first EASIX value within 0–24 hours was taken, and log₂(EASIX) was used as a continuous exposure. The primary outcome was 30-day all-cause mortality, and the secondary outcomes were 90-day and 365-day all-cause mortality. Cox proportional hazards models were used, with adjustment from M1 (unadjusted) to M4 (fully adjusted); the proportional hazards assumption was tested with scaled Schoenfeld residuals, categorical covariates violating the assumption were changed to stratification factors, continuous covariates were entered with a time-varying coefficient (γ × ln t), and hazard ratios (HRs) were reported at different time points. Restricted cubic splines, quartile analysis and directly standardized survival curves were also fitted, and subgroup analysis and E-value analysis were performed. Patients with stage 4 CKM were identified in eICU-CRD by an approximate definition for external validation; the endpoint in that database was all-cause mortality during hospitalization. Results A total of 9,766 patients with stage 4 CKM were finally included. Median age was 71 (61–81) years, 5,694 patients (58.3%) were male, and median EASIX was 2.1 (1.0–5.0). All-cause deaths at 30, 90 and 365 days were 2,278 (23.3%), 2,938 (30.1%) and 3,802 (38.9%), respectively. In the fully adjusted model, for each doubling of EASIX, the HR for 30-day mortality was 1.243 (95%CI 1.182–1.306) at 1 day after admission and 1.131 (1.079–1.185) at 30 days, and decreased with longer time since admission (γ = −0.0277, P = 0.026); for the 365-day outcome, the HR at the 365-day time point was 1.038 (0.993–1.086). For each 1-standard-deviation increase in log₂(EASIX) (1.861), the HR for 30-day mortality was 1.257 (1.152–1.372). Compared with Q1, the HR for 30-day mortality in Q4 was 1.660 (1.273–2.164), and the differences of Q2 and Q3 from Q1 were not statistically significant. The P values of the non-linearity test were 0.002 to < 0.001. Among the 9 subgroups, at the 30-day time point the interaction was statistically significant for heart failure, SOFA-4 and malignancy (P = 0.003, < 0.001, 0.039), and CKM stage 4a/4b did not reach significance (0.079). In the type 2 diabetes subgroup (n = 3,459), comparing the highest with the lowest quartile, the HR for 30-day mortality was 1.704 (1.061–2.735). The E value estimated at the 30-day time point was 1.40 (1.29). External validation included 1,103 patients, 173 of whom died during hospitalization (15.7%); after full adjustment, the HR of each doubling of EASIX for in-hospital death was 1.121 (1.045–1.203). Conclusions In ICU patients with stage 4 CKM, a higher EASIX calculated within 24 hours of admission was associated with a higher risk of all-cause mortality; the association was strongest early after admission and decreased over time, and was more evident in the higher EASIX range. In eICU-CRD the direction of this association was consistent, but the endpoint in that database was in-hospital mortality, which does not constitute a replication on the 30-day endpoint. The above findings should be interpreted as a descriptive association rather than a validated risk stratification tool; prospective validation on a consistent endpoint is still required before clinical implementation.