Patients who underwent non-cardiac surgery between January 2008 and August 2019 at Chinese PLA General Hospital were initially screened from a perioperative retrospective database. The inclusion criteria were as follows: (1) aged 65 yr or older; (2) underwent non-cardiac surgery; (3) received general anesthesia; and (4) were with duration of surgery >60 min. Patients who presented with an American Society of Anesthesiologists (ASA) classification of ≥IV, were performed under regional anesthesia, or had missing clinical data were excluded. Among patients who underwent multiple surgeries during the study period, only the first eligible surgery was considered. A flow diagram of the patient selection process is displayed in Figure 1.

The primary outcome of interest was perioperative ischemic stroke, defined as an episode of neurological dysfunction, such as motor, sensory, or cognitive dysfunction, caused by focal cerebral, spinal, or retinal infarction within 30 postoperative days (Sacco et al., 2013). Diagnoses of stroke are confirmed by a combination of neuroimaging and clinical evidence of cerebrovascular ischemia during hospital stay. In our study, perioperative ischemic stroke patients were identified if discharge records included at least 1 ICD-9-CM/ICD-10-CM diagnosis code for stroke (Supplementary Table 1).

Serum concentrations of preoperative peripheral platelet, neutrophil, and lymphocyte were the most recent blood counts measured within 3 days prior to surgery. The SII, NLR, and PLR were calculated as follows: SII = platelet × neutrophil/lymphocyte, NLR = neutrophil/lymphocyte, PLR = platelet/lymphocyte. In order to visualize perioperative ischemic stroke in relation to SII, NLR, and PLR, parameters were grouped into high and low by the optimal cut-off points using the receiver operating characteristic curve (ROC).

The exposure of interest was preoperative SII. Result from the ROC analysis indicated that the optimal cut-off value for the SII was set at 583 × 10⁹ for predicting perioperative ischemic stroke. Consequently, the SII was stratified into low (< 583 × 10⁹) and high (≥ 583 × 10⁹) for subsequent analyses.

Preoperative covariates of interest, such as age, sex, body mass index (BMI), ASA classification, hypertension, diabetes mellitus, prior ischemic stroke, coronary heart disease, arterial fibrillation, peripheral vascular disease, renal dysfunction, and preoperative use of β-blockers or aspirin, were noted. The indices derived from the preoperative laboratory data, including hemoglobin, albumin, total bilirubin, prothrombin time, were defined as the most recent blood counts measured within 3 days prior to surgery. In addition, we collected surgical and anesthetic information, such as preoperative mean arterial pressure (MAP), surgical procedures, duration of procedures, estimated blood loss, intraoperative hypotension (MAP ≤ 65 mmHg), crystalloid or colloid infusion, blood product transfusion, non-steroid anti-inflammatory drugs (NSAIDs) use, glucocorticoid medication, cumulative opioid consumption, and volatile anesthetic use.

Propensity score matching (PSM) was applied to address the potential residual confounding effect of the covariates such as demographic variables and clinical parameters, which would result in a biased outcome. The propensity score, a composite score, was computed based on the probability of patients having different levels of SII and derived from synthesized baseline parameters using the multivariate logistic regression model (Williamson and Forbes, 2014). After the propensity score generation, patients with SII < 583 and SII ≥ 583 were randomly matched at a 1:1 ratio using the greedy nearest-neighbor matching approach (maximum caliper width, 0.1). Kernel density plots of the propensity scores were applied to examine the equivalence between the matched patients. The standardized mean difference (SMD) was reported and performed to evaluate the baseline differences between the two groups, where a value < 0.1 indicated minor differences (Cheung et al., 2019).

Data were statistically analyzed using IBM SPSS Statistics for Windows, version 26.0 (IBM, Corp.) and R statistical software (R version 4.0.5, R Foundation for Statistical Computing). Continuous variables were summarized as mean [standard deviation (SD)], or median [interquartile range (IQR)]. Categorical data were displayed as numbers and percentages. The NLR or PLR score was converted into a two-level categorical variable based on the ROC analysis. The SII was also evaluated as continuous variable. We applied an extended multivariate logistic regression model approach to evaluate the clinical effect of the SII on perioperative ischemic stroke. Based on the type of surgical procedure, we conducted sensitivity analysis by only including non-neurosurgical patients to test the robustness of our findings. Subsequently, we further analyzed subgroups to appraise the primary correlation between the SII and perioperative ischemic stroke according to age, sex, intraoperative hypotension, diabetes mellitus, and prior ischemic stroke (Selim, 2007). The comparison group in this study referred to patients with SII < 583, unless stated otherwise. For all tests, statistical significance was inferred at a two-sided P-value < 0.05.

A total of 58,776 older patients who underwent non-cardiac surgery from January 2008 to August 2019 were enrolled in the study. Figure 1 illustrates the flow diagram of patient selection. After applying the inclusion and exclusion criteria, 40,670 eligible patients remained in the analysis, with a median age of 70 years (IQR: 67, 74), of whom 18,334 (45.1%) were women. Among the patients, 14,811 (36.4%) underwent intra-abdominal surgery, 1,903 (4.7%) underwent neurosurgery; and 6,224 (15.3%) received blood product transfusion. The prevalence of intraoperative hypotension (MAP ≤ 65 mmHg) in the cohort was 45.1% (n = 18,349). Of the entire patient cohort, 237 (0.58%) experienced an ischemic stroke within 30 days after surgery. The stroke incidence was consistent with previously reported rates of 0.1–0.7% in patients who underwent non-cardiac surgery (Vasivej et al., 2016; Woo et al., 2021).

The median NLR, PLR, and SII values were 2.00 (IQR:1.49, 2.83), 120.5 (IQR:93.1, 159.0), and 413.0 (IQR:287.0, 626.5), respectively. Using the ROC analysis, NLR, PLR, and SII accurately predicted perioperative ischemic stroke, with area under curve of 0.599 (95% CI: 0.534–0.632), 0.560 (95% CI: 0.505–0.616), and 0.721 (95% CI: 0.673–0.758), respectively. The Youden index of NLR, PLR, and SII were 0.165, 0.129, and 0.184, respectively. The optimal cut-off points derived for NLR, PLR, and SII to predict perioperative ischemic stroke were 3, 119, and 583 × 10⁹, respectively. The ROC curve of NLR, PLR, and SII for stroke risk classifier are shown in Supplementary Figures 1–3.

The patients were then grouped into low (< 583, n = 29,060, 71.5%) and high (≥ 583, n = 11,610, 28.5%) SII groups. The differences in baseline characteristics between the groups are presented in Table 1. Some baseline clinical characteristics in the study were relatively similar between the SII < 583 and SII ≥ 583 groups. However, other characteristics, such as sex, ASA classification, preoperative hemoglobin level, surgical procedures, duration of procedures, estimated blood loss, intraoperative hypotension, colloid infusion, blood transfusion, and opioid analgesic medication, differed between the two groups. Patients with SII ≥ 583 had more cardiovascular and cerebrovascular comorbidities (hypertension, diabetes mellitus, prior ischemic stroke, coronary heart disease, and peripheral vascular disease) and higher long-term administration of β-blockers and aspirin, than did those with SII < 583. Compared to patients with SII < 583, those with SII ≥ 583 had higher preoperative NLR and PLR scores.

We initially evaluated the association between the SII as continuous variable and perioperative ischemic stroke using univariate and multivariate logistic analyses. The univariate analysis showed that the SII was associated with perioperative ischemic stroke [odds ratio (OR): 1.285; 95% confidence interval (CI): 1.174–1.542; P < 0.001]. In the multivariate logistic regression model, the adjusted OR of SII was 1.213 (95% CI: 1.169–1.468; P < 0.001). The SII as continuous variable was an independent predictor of perioperative ischemic stroke (Supplementary Table 2).

When investigating the prognostic values of SII as categorical variable, the results of our univariate analysis revealed that age, hypertension, diabetes mellitus, prior ischemic stroke, coronary heart disease, arterial fibrillation, peripheral vascular disease, β-blockers use, aspirin medication, preoperative MAP, surgical procedures, duration of procedures, estimated blood loss, intraoperative hypotension (MAP ≤ 65 mmHg), crystalloid infusion, NSAIDs use, NLR, PLR, and SII were prognostic factors for perioperative ischemic stroke, whereas sex, BMI, ASA classification, renal dysfunction, preoperative laboratory data (hemoglobin, albumin, total bilirubin, and prothrombin time), colloid infusion, blood product transfusion, glucocorticoid use, opioid analgesic medication, and volatile anesthetic use had no prognostic significance (Supplementary Table 3). An elevated SII was strongly associated with an increased risk of perioperative ischemic stroke in the unadjusted analysis [OR: 2.217; 95% (CI): 1.714–2.863; P < 0.001] (Table 2).

Based on the extended multivariate logistic regression models (Table 2), patients with SII ≥ 583 showed a significantly higher OR of perioperative ischemic stroke in all three models consistently (OR range: 1.843–2.301, P < 0.001 for all) (Table 2). The NLR, but not PLR, was also an independent predictor of perioperative ischemic stroke in the adjusted analysis (OR: 1.036; 95% CI: 1.007–1.062; P = 0.007) (Supplementary Table 3). The other independent risk factors for predicting perioperative ischemic stroke are shown in Supplementary Table 3.

Then, we performed PSM analysis to assess the prognostic value of the SII further. Prior to matching, the median propensity score in older patients with SII ≥ 583 was 0.266 (IQR: 0.203–0.377) vs. 0.231 (IQR: 0.194–0.346) in those without SII < 583. PSM resulted in 10,826 patients in the SII ≥ 583 group matched to 10,826 patients in the SII < 583 group. The distribution of propensity scores in the high and low SII groups is graphically illustrated before and after matching (Figure 2). After matching, the mean (SD) propensity score was similar between those with high [0.314 (0.127)] and low [0.318 (0.121)] SII. The baseline demographic and clinical characteristics of the variables were generally well balanced between the two groups, with SMD less than 0.10 for all covariates (Table 1). Following multivariate logistic regression adjustment after PSM (n = 10,826), the correlation between SII and perioperative ischemic stroke remained significantly robust (OR: 2.195; 95% CI: 1.574–3.106; P < 0.001) (Table 2 and Supplementary Table 4).

The incidence of perioperative ischemic stroke is closely associated with the type and complexity of surgical procedures (Selim, 2007). The surgical procedures with the incidence of perioperative ischemic stroke at Chinese PLA General Hospital are shown in Figure 3. The surgical categories with highest overall incidence of perioperative ischemic stroke were neurosurgery (3.153%), followed by spine surgery (0.837%) and joint arthroplasty (0.631%). To further test the robustness of our results, we performed a sensitivity analysis excluding neurosurgeries. The adjusted OR of perioperative ischemic stroke in neurosurgical patients with the SII was 2.297 (95% CI: 1.192–4.429; P = 0.013). The association between preoperative SII and perioperative ischemic stroke remained stable in those non-neurosurgical patients (OR: 1.691; 95% CI: 1.202–2.376; P < 0.002) (Table 3).

Among 11,610 older patients with SII ≥ 583, 925 (8.0%) were aged ≥ 80 years, 4,683 (40.3%) were female, 5,749 (49.5%) had a duration of intraoperative hypotension (MAP ≤ 65 mmHg) ≥ 10 min, 8,854 (23.7%) presented with diabetes mellitus, and 847 (7.3%) had prior ischemic stroke. The subgroup analyses for the prediction of a poor neurological outcome were summarized according to age, sex, intraoperative hypotension, diabetes mellitus, and prior ischemic stroke (Figure 4). The OR of the SII was significant for age subgroups [≥ 80 years: OR (95% CI): 4.368 (1.127–6.508), P = 0.036; < 80 years: OR (95% CI): 1.766 (1.298–2.401), P < 0.001]. Additionally, an increased risk of perioperative ischemic stroke was observed in both male (OR: 1.900; 95% CI: 1.262–2.883; P = 0.002) and female (OR: 1.724; 95% CI: 1.109–2.655; P = 0.014). In patients with suboptimal intraoperative hypotension (duration of MAP ≤ 65 mmHg ≥ 10 min), the SII was significantly correlated with perioperative ischemic stroke (OR: 2.104; 95% CI: 1.382–2.184; P < 0.001). This increased risk was also significant in those with intraoperative hypotension (MAP duration ≤ 65 mmHg < 10 min). The correlation between preoperative SII and perioperative ischemic stroke was significant in individuals with (OR: 3.074; 95% CI: 1.765–5.404; P < 0.001) and without (OR: 1.520; 95% CI: 1.053–2.184; P = 0.024) diabetes. Additionally, an elevated SII was only significantly associated with an increased risk of perioperative ischemic stroke in the prior ischemic stroke group (OR: 2.678; 95% CI: 1.534–4.733; P < 0.001), whereas it was not significant in the no prior ischemic stroke group (OR: 1.440; 95% CI: 0.997–2.069; P = 0.055).