Baseline characteristics of study participants

This study analyzed data from 8,321 patients included in the MIMIC-IV database, of whom 3,848 met the inclusion criteria. Participants were stratified into four quartiles (Q1, Q2, Q3, and Q4) according to postoperative SHR percentiles. The baseline characteristics for each group are detailed in Table 1. The mean age of the cohort was 68 ± 12 years, with female participants comprising 30.6% (1,179). It is noteworthy that the Q4 group demonstrated the highest mean age and the largest proportion of female participants. The most prevalent comorbidities included MI, and HF, affecting 40.0%, and 37.6% of the population, respectively, with the Q4 group having the highest prevalence of these conditions. Furthermore, preoperative and postoperative WBC counts, creatinine levels, aps iii, and durations of mechanical ventilation, ICU stays, and total hospitalizations were consistently highest in the Q4 group.

Table 1 Baseline characteristics of patients grouped according to postoperative SHR index quartilesRelationship between postoperative SHR and clinical outcomes

Clinical outcomes varied significantly across postoperative SHR quartiles. Patients in the Q4 group exhibited the highest rates of in-hospital mortality (4.9%), 30-day mortality (4.7%), 90-day mortality (7.9%), and 360-day mortality (12.5%). An adjusted logistics regression analysis, accounting for sex, age, MI, HF, CVD, pulmonary disease, diabetes, renal disease, duration of mechanical ventilation, aps iii, and preoperative creatinine and WBC levels, revealed Q4 patients exhibiting higher risks of in-hospital mortality (OR = 3.323; 95% CI 1.558–7.089; p = 0.002), 30-day mortality (OR = 2.877; 95% CI 1.391–5.590; p = 0.004), 90-day mortality (OR = 1.918; 95% CI 1.187–3.099; p = 0.008), and 360-day mortality (OR = 1.485; 95% CI 1.031–2.138; p = 0.034). By contrast, no significant differences were observed between Q2 and Q1, or Q3 and Q1 (Table 2). These findings indicate that patients with an SHR index of ≥ 1.40 have a higher risk of in-hospital, 30-day, 90-day and 365-day all-cause mortality compared to those with an SHR index of < 1.40. Similar trends were observed for in-hospital, 30-day, 90-day and 365-day all-cause mortality, as detailed in Fig. 3.

Table 2 Logistic regression models for hospital, 30-day, 90-day and 360-day all-cause mortalitySurvival analysis

Kaplan–Meier survival analyses revealed significant differences in survival rates across postoperative SHR quartiles for in-hospital, 90-day and 360-day all-cause mortality. Patients in the Q4 group experienced lowest survival rates at all time points compared to those in lower postoperative SHR quartiles (log-rank p < 0.05). However, survival rates did not differ significantly among the Q1, Q2, and Q3 groups across any time point (Fig. 2).

Fig. 2

Kaplan–Meier survival analysis curves for all-cause mortality. Kaplan–Meier curves of hospital (A) 30-day (B), 90-day (C) and 360-day (B) all-cause mortality stratified by postoperative SHR index, SHR Stress hyperglycemia ratio

Predictive value and nonlinear relationship

The prognostic utility of preoperative SHR, postoperative SHR, and the rate of SHR change for in-hospital, 30-day, 90-day, and 360-day mortality was assessed using AUC analysis. Among these, postoperative SHR exhibited the strongest predictive value, with AUCs of 0.723, 0.710, 0.658, and 0.618, respectively (eFigure 1). Moreover, restricted cubic spline (RCS) analysis indicated a nonlinear association between postoperative SHR and all-cause mortality at all time points (in-hospital, 30-day, 90-day, and 360-day all-cause mortality). Increasing postoperative SHR values were consistently associated with higher mortality risks, demonstrating the nonlinear nature of this relationship (p for nonlinear < 0.05) (Fig. 3).

Fig. 3

RCS of SHR index with all-cause mortality. RCS of postoperative SHR index with hospital A 30-day B, 90-day C and 360-day B all-cause mortality. SHR Stress hyperglycemia ratio, RCS Restricted cubic splines

Stratified analyses

Subgroup analyses were conducted to explore potential effect modifications by sex, age, MI, HF, CVD, pulmonary disease, diabetes, and renal disease on the association between postoperative SHR and in-hospital, 30-day, 90-day, 360-day all-cause mortality (eFigure 2–5). Postoperative SHR was significantly associated with 30-day mortality among males, individuals aged ≤ 75 years, those with creatinine levels > 90, patients with MI, those without heart failure, individuals with CVD, and those without pulmonary or renal disease (p < 0.05). Conversely, postoperative SHR was no associated with 30-day mortality among females, individuals aged < 75 years, those with creatinine levels ≤ 90, and patients without MI, heart failure, CVD, pulmonary disease, or renal disease. However, postoperative SHR was significantly associated with 360-day mortality among those without HF.

Mediation analysis

Mediation analysis, conducted on 2,649 patients with complete data for monocytes, neutrophils, lymphocytes, and platelets, demonstrated that postoperative SHR indirectly influenced prolonged mechanical ventilation through its association with inflammatory markers (eTable 1). These markers include the neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), and systemic inflammation response index (SIRI). Prolonged mechanical ventilation, in turn, was associated with increased 30-day mortality risk (p < 0.05) (eTable 2, eFigure 3).

Machine learning

Using six machine learning algorithms were used to predict in-hospital all-cause mortality and were further extended to predict 360-day all-cause mortality. The model incorporated creatinine, SHR, HR, RR, HF. NB algorithm demonstrated the strongest predictive performance on in-hospital and 360-day all-cause mortality, achieving an AUC of 0.7936 and 0.7410 comparing with other models. (Fig. 4, eFigs. 7 and 8).

Fig. 4

The machine learning algorithm predicts in-hospital and 360-day all-cause mortality. AUC Area under the curve, XGBoost Extreme gradient boosting, SVM Support vector machine, AdaBoost Adaptive boosting, GBM Gradient boosting machine