This is a retrospective cohort study that adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) (24) statement. Most of the data were extracted from Hospital Information System (HIS) and Anesthesia Information Management System (AIMS). In-hospital deaths were documented with death certificates and medical record reviews, and out-of-hospital deaths and dates of death were obtained through telephone follow-up. Because of the sensitive nature of the data collected for this study, data collection was performed by unwitting hospital information center staff. After extracting the raw data of the preoperative evaluation sheets, the baseline characteristics were organized into a standardized form by independent investigators who were blinded to the outcomes. Data analysis was performed by qualified researchers trained in human subject confidentiality agreements. For purposes of confidentiality, all data were de-identified and analyzed anonymously. Ethical approval was obtained from the Ethics Committee of Sichuan University (Project No. 1082 in 2021). The study was registered at chictr.org (ChiCTR2200058376).

We identified consecutive patients who underwent surgery between February 2018 and November 2020 at West China Hospital, Sichuan University. Inclusion criteria: (1) Age > 16 years; (2) Patients who underwent non-cardiac surgery; (3) Patients who underwent hs-cTnT examination within 7 days before surgery; (4) ASA Grade I-V. Exclusion criteria: (1) Patients undergoing diagnostic, cardiac, obstetric, and ophthalmic surgery; (2) Hospital stay less than 24 h (day surgery).

The primary outcome measure was defined as all-cause mortality 1 year after surgery. The secondary outcome measures were defined as MINS, length of hospital stay (LOS), and postoperative ICU admission. MINS (8) was defined as the concentration of hs-cTnT ≥20 ng/L and the absolute value change ≥5 ng/L, or the absolute value of hs-cTnT ≥65 ng/L, and no such detection was considered as no postoperative myocardial injury. Peak postoperative values regardless of the day of sampling were used in these calculations.

We also generated a list of risk-adjustment variables (Supplementary Table 1), including patient age, sex, body mass index (BMI), 15 preoperative comorbidities, 12 preoperative laboratory tests, 5 prognostic models (ASA, CCI, RCRI, Ex-care, SORT), types of surgery, and detail of anesthesia and intraoperative management. Preoperative anemia was defined as hemoglobin <130 g /L for men and < 120 g/L for women, preoperative increased creatinine was defined as plasma levels of creatinine 100 mmol/L for men and 90 mmol/L for women, intraoperative transfusion was defined as intraoperative transfusion of any blood product, and intraoperative hypotension was defined as MAP<55 mmHg at any time intraoperatively (regardless of duration).

Serum TnT examination is not a routine preoperative test for surgical patients in our institute. The clinicians screened high-risk groups to detect myocardial injury according to clinical guidelines and experience. We defined the preoperative hs-cTnT level as the serum concentration of the last hs-cTnT measurement within 7 days before surgery. We obtained fresh serum aliquots from supernatants of peripheral or central blood samples collected in lithium heparinized tubes that were previously centrifuged for 5 min at 3500× g in our Core Laboratory. The levels of serum hs-cTnT were measured by Elecsys Troponin T hs of non-competitive enzyme-linked immunosorbent assays (ELISA) using a Cobas e602Modular AutoAnalyzer (Roche Diagnostics, Basel, Switzerland). The minimum detectable value was 3 ng/L.

We first compared characteristics of patients with or without elevated preoperative hs-cTnT using Fisher exact tests or χ² tests for categorical measures and 2-sample t tests or Mann–Whitney U tests for continuous measures. Summary statistics are presented as mean (standard deviation), median [inter-quartile range (IQR)], or number (%).

To assess unadjusted associations between different hs-cTnT levels and time to the primary composite outcome, we first used stratified Kaplan–Meier plots with log-rank tests. Hazard ratios (HRs) were then reported using univariate Cox regression models that included only hs-cTnT levels as explanatory variables. We then fitted multivariable Cox regression models, adjusting for patient demographic characteristics, comorbidities, preoperative laboratory tests, hepatic and renal function, and detail of anesthesia and intraoperative managements (25). Univariate analysis of each factor was performed (Supplementary Table 1). Variables were considered candidates for inclusion in the risk-adjustment model if there were bivariable associations with the primary outcome event at p < 0.10. We selected final variables for the multivariable Cox model using the both-sided stepwise regression method. Collinearity was evaluated by the variance inflation factor (VIF) (26), and only variables with VIF ≤ 10 were input into the final model. We assessed multivariable Cox model appropriateness by receiver operating curve (ROC) and calibration curve (27, 28). Logistic regression was also used to assess the association of preoperative hs-cTnT and MINS, LOS, and ICU admission reported as odds ratios (ORs) with 95% confidence intervals (CIs). Statistical analyses were performed using R 4.0.2 and R-packages (tableone, survminer, survival, rms, MASS, AUC, calibrate and ggsurvplot).

The sample size required for developing a clinical prediction model is calculated according to the guidance published in 2020 (29). Use pmsampsize to calculate the minimum sample size required for developing a multivariable prediction model with a survival outcome using 50 candidate predictors. We expected all-cause mortality of about 7% after noncardiac surgery, a mean follow-up of more than 200 days, and an adjusted R-squared of 0.19. We select a time point of interest for prediction using the newly developed model of 1 year. The result indicates that at least 3,205 samples are required, corresponding to 1756.2 person-time of follow-up and 2.46 events per covariate. Sample size calculation was performed using R-packages (pmsampsize).

We conducted subgroup analyses to determine whether associations between hs-cTnT levels and one-year mortality varied according to different ages, BMI, ASA-PS scores, emergency cases, surgical categories, and comorbidities. Due to the minimum sample size limitation, only partial subgroup analyses were performed for the type of surgery. For every risk-factor subgroup, the respective variables defining the risk factor were removed from the analyses. Every subgroup was treated as new data to calculate the HR for the association between the preoperative hs-cTnT level and outcomes. All p values were 2-sided, with p < 0.05 defined as the threshold of statistical significance. Forestplot of subgroup analyses was performed using R-packages (forestplot).

To investigate the mediation effect (30, 31) of MINS between preoperative hs-cTnT level and one-year mortality, structural equation modeling (SEM) analysis (32) was performed using the lavaan package (33) on the R software and graphically reported using the lavaanPlot package. Two models were estimated: a multivariate logistic regression model for MINS (mediator) conditional on hs-cTnT and all study confounders and a multivariate linear regression model for the mortality (outcome) conditional on hs-cTnT, MINS, and all study confounders. The natural direct effects (NDE) represented the effect of hs-cTnT on mortality that was independent of MINS. The natural indirect effects (NIE) represented the proportion of hs-cTnT that could be explained by its association with MINS. To quantify the magnitude of mediation, the study estimated the proportion of the association mediated by MINS (NIE/[NDE + NIE]). All analyses were estimated using bootstrapping (1,000 replications) to recover the correct standard errors for direct and indirect effects (32). All p values were 2-tailed, and statistical significance was set at a p value less than.05.

A total of 7,156 who underwent hs-cTnT measurements before non-cardiac surgery were enrolled in our trial. Participant flow is outlined in Supplementary Figure 1. In 2151 patients (30.1%), the hs-cTnT concentration was above the 99th percentile of the upper range limit (URL ≥ 14 ng/L). The median preoperative hs-cTnT was 9.20 ng/L (IQR 6.10–23.75). The mean time for hs-cTnT testing before surgery was 2.3 days (SD 1.28).

The characteristics of included patients are summarized in Table 1. A relatively small proportion (ie, 589 [8.23%]) were censored on February 1, 2022, with more than 1 year of follow-up. We identified 617 (9.2%) deaths, 644 (9.0%) MINS, and 2099 (29.3%) ICU admission after surgery during follow-up. The median length of hospital stay was 12.0 days (IQR 8.0–20.0). Patients with higher ASA and RCRI grades, SORT, and Ex-care scores also had higher hs-cTnT levels before surgery. The preoperative hs-cTnT level is correlated with intraoperative hypotension, rapid heart rate, increased blood loss, and transfusion. However, age, gender, BMI, and operation duration had no significant correlation with preoperative hs-cTnT concentration. Although differences in all preoperative laboratory tests were statistically significant in this large data set, most of the differences were generally of small magnitude. Compared with patients with normal preoperative hs-cTnT, those with elevated hs-cTnT were more likely to have more ICU admission (1,286 [25.7%] vs. 813 [62.2%]; p < 0.001), more MINS (231 [4.6%] vs. 413 [19.2%]; p < 0.001), and a longer median (IQR) stay in hospital (11 [8–18] days vs. 15 [8–24] days; p < 0.001). Similarly, preoperative hs-cTnT elevation was significantly associated with 30-day (162 [3.5%] vs. 248 [12.9%]; p < 0.001), 90-day (191 [4.1%] vs. 302 [15.8%]; p < 0.001) and one-year mortality (192 [3.9%] vs. 308 [14.8%]; p < 0.001).

Preoperative elevated hs-cTnT was associated with higher mortality in unadjusted survival analysis (p < 0.001 by log-rank test) (Figure 1). The hazard ratio of 30-day mortality was 3.90 (95% CI, 3.20–4.75; p < 0.001). It steadily increased and maintained statistical significance over the one-year period after surgery (HR at 1 year, 4.13; 95% CI, 3.45–4.95; p < 0.001). It’s worth noting that most events occurred within 3 months after surgery, afterwards the Kaplan–Meier plots did not seperate further. Figure 2 summarizes the multivariable Cox regression risk-adjustment model. The area under the ROC curve of the model was 0.88. The ROC and calibration curves were shown in Supplementary Figure 2. Results were similar in a multivariable Cox regression model that adjusted for all covariates in Table 2 (adjusted HR at 1 year, 1.93; 95% CI, 1.58–2.36; p < 0.001; Table 2).

Preoperative elevated hs-cTnT was also associated with MINS, ICU admission, and LOS (p < 0.001 by Wald test for each outcome). In a univariable logistic regression model, ORs for associations of elevated hs-cTnT with MINS, ICU admission, and LOS were significant (MINS: odds ratio[OR] 4.91; 95% CI, 4.14–5.81, p < 0.001; LOS:OR 1.84; 95% CI, 1.68–2.01, p < 0.001; ICU admission: OR 1.76; 95%CI, 1.58–1.96, p < 0.001). Results were similar in a multivariable model adjusting for patient characteristics during the initial admission (ie, adjusted for all covariates in Table 2), with a significant association of with higher incidence of MINS, ICU admission, and LOS (MINS: adjusted odds ratio[aOR] 3.01; 95% CI, 2.46–3.69; p < 0.001; LOS: aOR 1.48, 95%CI 1.34–1.641; p < 0.001; ICU admission: aOR 1.52, 95%CI 1.31–1.76; p < 0.001).

In subgroup analyses stratified by sex, age, BMI, ASA grades, surgical categories, degrees of surgical urgency, and preoperative comorbidities, the associations between preoperative hs-cTnT and one-year mortality remained in all categories (Figure 3). Subgroup analysis of diabetic patients showed insignificant results due to the small sample size, but the trend was consistent with the overall. In a subgroup analysis, preoperative hs-cTnT was most strongly associated with mortality among patients after general surgery (aHR, 3.34; 95% CI, 2.13–5.23; p < 0.001). In a subgroup analysis by ASA scores, preoperative hs-cTnT was most strongly associated with adverse outcomes among patients with ASA GradeI-II, followed by those with Grade III and IV-V (GradeI-II: aHR, 3.81; 95% CI, 1.51–9.62; p < 0.001; GradeIII: aHR, 1.82; 95%CI, 1.37–2.42; p < 0.001; GradeIV-V: aHR, 1.98; 95% CI, 1.46–2.67; p < 0.001). After adjusting for multiple factors, the results of a subgroup analysis suggest that preoperative hs-cTnT remains significantly correlated with mortality in patients who have undergone either elective surgery (aHR 2.28; 95% CI 1.51–3.44; p < 0.001) or emergency surgery (aHR 1.82; 95% CI 1.45–2.29; p < 0.001).

aHRs of other subgroups remained stable between 1.71 and 2.13.

The SEM analysis showed that a significant main effect was found for elevated hs-cTnT with one-year mortality after surgery (β = 0.043, 95%CI, 0.037–0.049; p < 0.001). The natural direct effects (NDE) of TnT indicated that we would, on average, observe a 0.029 point (95% CI, 0.028–0.030; p < 0.001) increase in mortality if all study participants were free of MINS. The natural indirect effects (NIE) via MINS implied that we would, observe a 0.014 point (95% CI, 0.013–0.015; p < 0.001) increase in mortality among participants with elevated TnT. The proportion of the association between TnT and mortality mediated by MINS was 33.6% (95%CI 28.1–39.1%). Figure 4 illustrates a simplified SEM representation of the mediation effect of MINS.