This multicenter retrospective study included consecutive patients aged ≥ 65 years with documented CAD who underwent non-cardiac surgery at two tertiary academic medical centers in Zhejiang, China. Participants were recruited from the First Affiliated Hospital of Zhejiang University School of Medicine (AHZU) between January 1, 2013 and May 31, 2021, and from the Fourth AHZU between October 1, 2020, and October 31, 2024.

This study was adhered to the Declaration of Helsinki and received ethical approval from the Institutional Review Boards (IRB) of both participating institutions. The First AHZU granted approval (Approval No. IIT20230114A; February 2023) for data collection spanning from 2013 to 2021. Subsequently, the Fourth AHZU provided approval (Approval No. K2024222; December 2024) for data collection covering the period from 2020 to 2024. The overlapping data collection period (1 October 2020 to 31 May 2021) falls within the valid approval periods of both institutions. Due to the retrospective nature of the study, the requirement for written informed consent was waived. All data were anonymized and de-identified prior to analysis.

The CAD was defined based on any of the following criteria: angiographic evidence of coronary stenosis > 50%; documented myocardial infarction > 3 months prior to enrollment; coronary revascularization > 3 months prior to enrolment; positive results on myocardial perfusion scintigraphy or exercise stress test; or typical anginal symptoms accompanied by electrocardiographic evidence of myocardial ischemia (10). Surgical procedures included elective non-cardiac surgery, classified according to the American College of Cardiology (ACC)/American Heart Association (AHA) guidelines for perioperative cardiovascular assessment (11). Exclusion criteria comprised: day surgery; emergency surgery; patients who underwent multiple (≥2) operations during a single hospital admission; and incomplete or insufficient clinical data for comprehensive analysis. All patients underwent routine preoperative evaluations following established perioperative management guidelines.

Data were obtained from the integrated electronic medical record systems of the First and Fourth AHZU. Initial patient identification was conducted using the International Classification of Diseases, Tenth Revision (ICD-10) coding system to identify all surgical department discharges with CAD diagnoses during the study period. Each case was manually reviewed and rigorously assessed against the predefined inclusion and exclusion criteria. Clinical data extracted from electronic medical records included demographic characteristics, preoperative evaluations, American Society of Anesthesiologists (ASA) physical status classifications, surgical types, anesthesia techniques, perioperative cardiovascular complications, and other relevant perioperative information. All preoperative assessments were conducted within 30 days preceding surgery. Extraneous data were excluded from analysis. When multiple measurements were available during this period, the temporally closest value to the surgical date was selected to maximize clinical relevance.

Geriatric nutritional risk index was calculated using the following formula: GNRI = 1.489 × albumin (g/L) + 41.7 × (current weight/ideal body weight). Ideal body weight was derived from the Lorentz equations (12).

The RCRI consists of six components, each assigned a binary score of 0 (absent) or 1 (present): history of ischemic heart disease; history of congestive heart failure; history of cerebrovascular disease; insulin-dependent diabetes mellitus; creatinine > 2 mg/dL; and high-risk surgery (suprainguinal vascular, intraperitoneal, or intrathoracic procedures) (13). The total RCRI score was calculated as the sum of these components.

The primary outcome was a composite of PCEs, including all-cause death, resuscitated cardiac arrest, myocardial infarction, heart failure, and stroke, occurring intraoperatively or during postoperative hospitalization. Cardiac arrest was defined as the loss of circulation requiring chest compressions, defibrillation, or both (14). Myocardial infarction was defined as acute myocardial injury with clinical evidence of acute myocardial ischemia, diagnosed based on a rise or fall in cardiac troponin values (at least one value above the 99th percentile upper reference limit) accompanied by one or more of the following: symptoms of myocardial ischemia; new ischemic ECG changes; development of pathological Q waves; imaging evidence of new loss of viable myocardium or new regional wall motion abnormality consistent with ischemia; and identification of a coronary thrombus by angiography or autopsy (15). Cardiac biomarkers were assessed only when myocardial infarction was clinically suspected or ischemic ECG changes were observed. Heart failure was diagnosed based on clinical symptoms or physical examination findings, including orthopnea, dyspnea, jugular venous distention, peripheral edema, third heart sound, rales, or chest X-ray evidence of pulmonary edema or vascular redistribution (16). Stroke was diagnosed by a neurologist based on new neurological deficits confirmed by imaging (17).

Data were systematically entered into Microsoft Excel (Microsoft, Redmond, Washington) and analyzed using Statistical Package for Social Sciences (SPSS, version 23, IBM, Armonk, New York). Date distribution was assessed using histograms and Q–Q plots. Continuous variables were summarized as median (interquartile range, IQR) or mean ± standard deviation (SD), depending on their distribution. Categorical variables were presented as frequencies and percentages. Group comparisons were performed using the Kruskal–Wallis rank-sum test or variance test for continuous variables, depending on the distribution, and the chi-squared test or Fisher’s exact test for categorical variables, as appropriate. Univariate and multivariate logistic regression analyses were conducted to identify predictors associated with outcomes. Restricted cubic spline (RCS) curves were generated using R software (version 4.2.2) with the “ggplot2” and “rcs” packages, based on logistic regression models. The optimal GNRI cutoff value was determined by the Youden index. Model performance was evaluated using receiver operating characteristic (ROC) curves, with the area under the curve (AUC) as a measure of discrimination. The DeLong test was used to compare AUC values between models. Calibration was assessed using Hosmer-Lemeshow test and calibration plots. Decision curve analysis (DCA) was performed to evaluate the clinical utility of the model. Statistical parameters, including odds ratios (OR) and 95% confidence intervals (CI), were reported. A two-tailed p-value < 0.05 was considered statistically significant for all analyses.

A total of 7,272 patients aged ≥ 65 years with CAD undergoing non-cardiac surgery were included in this study, with a median age of 73 years (IQR, 69–78). Figure 1 illustrates the patient enrollment and analysis flowchart. Baseline clinical characteristics and their association with perioperative outcomes are comprehensively presented in Table 1.

Patients underwent diverse surgical procedures at two tertiary referral centers, predominantly comprising general, urologic, orthopedic, thoracic, and vascular surgeries.

The PCEs occurred in 408 patients, representing a prevalence rate of 5.6%. Compared to patients without PCEs, those with PCEs were significantly older (median age: 76 vs. 73 years, p < 0.001), had lower body mass index (median body mass index: 22.31 vs. 23.60 kg/m², p < 0.001), and included more males (74.8% vs. 65.0%, p = 0.001). The PCEs group demonstrated significantly higher prevalence of comorbidities: diabetes mellitus (34.6% vs. 27.4%, p = 0.002), stroke (14.7% vs. 9.5%, p = 0.001), dialysis (5.6% vs. 0.9%, p < 0.001), ischemic heart disease (61.0% vs. 40.1%, p < 0.001), heart failure (18.4% vs. 4.9%, p < 0.001), atrial fibrillation (15.2% vs. 5.6%, p < 0.001), and valvular heart disease (4.2% vs. 1.4%, p < 0.001). ASA classification distributions also differed significantly, with the PCEs group having higher proportions of ASA III (75.2% vs. 58.8%; p < 0.001) and ASA IV (7.6% vs. 0.4%; p < 0.001) patients.

Preoperative laboratory analysis revealed significant differences between groups. The PCEs group had elevated leukocyte counts and creatinine levels but lower hemoglobin levels, platelet counts, and albumin concentrations compared to the non-PCEs group.

Regarding surgical characteristics, the PCEs group had higher rates of general anesthesia use (80.6% vs. 71.4%, p < 0.001) and were more likely to undergo general abdominal (38.7% vs. 21.9%, p < 0.001), neurological (6.1% vs. 3.5%, p = 0.006), and vascular surgeries (15.9% vs. 9.3%, p = 0.001) compared to the non-PCEs group.

A total of 408 patients experienced PCEs. Heart failure was the most prevalent complication (58.6%, n = 239), followed by myocardial infarction (54.2%, n = 221), while all-cause mortality occurred in 16.4% (n = 67) of cases. The detailed composition of PCEs is provided in Table 2.

The GNRI was significantly lower in the PCEs group compared to the non-PCEs group (median GNRI: 96 vs. 107, p < 0.001), as detailed in Table 1. RCS curves with five knots at the 5th, 28th, 50th, 72th, and 95th percentiles were used to model the association between GNRI and PCEs (Figure 2). The RCS analysis revealed a significant inverse linear correlation between GNRI levels and PCEs (OR = 0.92; 95% CI: 0.91–0.93; p < 0.001).

Using the Youden index-derived optimal cutoff value (GNRI = 98), the cohort was stratified into two groups: at-risk group (GNRI < 98) and no-risk group (GNRI ≥ 98). Threshold effect analysis was performed for each group (Table 3). Univariate and multivariate regression analyses were conducted to evaluate the association between at-risk GNRI and perioperative outcomes (Table 4). Univariate analysis identified potential predictors, while multivariate analysis, after adjusting for confounders, confirmed the independent predictive value of at-risk GNRI (OR = 1.919; 95% CI: 1.496–2.461; p < 0.001).

Table 5 presents univariate and multivariate analyses of GNRI and RCRI associations with perioperative outcomes. Using the multivariate regression coefficients, a weighted scoring system was developed, assigning two points to GNRI and one point to each RCRI component. This integration created a novel composite prognostic index, the GNRI plus RCRI model, with a total possible score of 8 points (2 points from GNRI and 6 points from RCRI).

The ROC analysis was used to evaluate the discriminatory ability of GNRI, RCRI, and the composite model (Figure 3). While GNRI demonstrated comparable discriminatory ability to RCRI (AUC: 0.676 vs. 0.694, p = 0.309), the GNRI plus RCRI model significantly outperformed both individual indices (vs. RCRI: AUC 0.768 vs. 0.694, p < 0.001; vs. GNRI: AUC 0.768 vs. 0.676, p < 0.001). The composite model exhibited good calibration, as indicated by a non-significant Hosmer-Lemeshow test (p = 0.391) and agreement between predicted and observed probabilities in the calibration curve (Figure 4). Decision curve analysis confirmed the superior clinical utility of the GNRI plus RCRI model across a wide range of threshold probabilities (Figure 5).

Subgroup analysis revealed significant associations between at-risk GNRI and perioperative outcomes across various subgroups (Figure 6). Stratified by age, the interaction was not significant (P for interaction = 0.069). Sex-specific analysis showed no significant interaction (P for interaction = 0.386), with ORs of 4.50 (95% CI: 3.55–5.71, p < 0.001) for males and 5.53 (95% CI: 3.70–8.27, p < 0.001) for females. Hypertension did not significantly modify the association (P for interaction = 0.155), with ORs of 6.31 (95% CI: 4.28–9.32, p < 0.001) for non-hypertensive and 4.52 (95% CI: 3.54–5.78, p < 0.001) for hypertensive individuals. Diabetes mellitus showed a significant interaction (P for interaction = 0.038), with higher ORs in non-diabetic (OR = 5.80, 95% CI: 4.49–7.49, p < 0.001) compared to diabetic patients (OR = 3.66, 95% CI: 2.58–5.20, p < 0.001). Ischemic heart disease approached significance (P for interaction = 0.052), with ORs of 6.03 (95% CI: 4.36–8.36, p < 0.001) for non-ischemic and 3.98 (95% CI: 3.05–5.19, p < 0.001) for ischemic cases. ASA class did not significantly interact (P for interaction = 0.132). Among surgical types, thoracic surgery showed the highest OR (6.48, 95% CI: 2.93–14.32, p < 0.001), with no significant interaction (P for interaction = 0.111). General anesthesia showed a borderline significant interaction (P for interaction = 0.053), with ORs of 3.28 (95% CI: 2.09–5.17, p < 0.001) for non-general anesthesia and 5.43 (95% CI: 4.32–6.83, p < 0.001) for general anesthesia.