Preoperative systemic immune-inflammation index as an incremental predictor of postoperative refractory hypoxemia in acute type a aortic dissection Authors: Zhang Liu, Yi Shen, Juan Zou, Weiqin Huang Published: 2026 ABSTRACT BackgroundPostoperative refractory hypoxemia (PRH) remains a clinically important pulmonary complication after surgery for acute type A aortic dissection (ATAAD). Because inflammation, coagulation activation, pulmonary vulnerability, and operative stress converge during ATAAD repair, a composite biomarker may provide more useful risk information than a single leukocyte-derived index. We evaluated whether the preoperative systemic immune-inflammation index (SII) was independently associated with PRH and whether it improved risk prediction beyond conventional clinical, pulmonary, and biomarker variables.MethodsIn this single-center retrospective cohort study, 490 patients undergoing surgery for ATAAD between January 2018 and December 2024 were analyzed. SII was calculated from absolute platelet, neutrophil, and lymphocyte counts and modeled primarily as a log2-transformed continuous variable, so that odds ratios represented the association per doubling of SII. The primary preoperative model included admission/preoperative variables, whereas an expanded perioperative model was used as a sensitivity analysis. Model discrimination, reclassification, decision-curve performance, and calibration were evaluated, and internal validation was performed using 1000 bootstrap resamples.ResultsPRH occurred in 158 of 490 patients (32.2%). Preoperative SII was higher in patients who developed PRH than in those who did not [1899.5 (1140.1-3043.6) vs. 667.5 (408.7-1173.7), P < 0.001]. In the primary preoperative prediction model, each doubling of SII was independently associated with PRH [adjusted odds ratio (OR), 2.80; 95% CI, 2.20–3.56; P < 0.001]. The base preoperative model had an AUC of 0.750 (95% CI, 0.705–0.794), whereas the SII-augmented model had an AUC of 0.846 (95% CI, 0.810–0.880). SII yielded the numerically largest incremental discrimination among SII, NLR, PLR, and MLR. Bootstrap validation of the SII-augmented model produced an optimism-corrected AUC of 0.834, calibration slope of 1.00, calibration intercept of 0.00, and Brier score of 0.153.ConclusionsPreoperative SII was independently associated with PRH after ATAAD surgery and provided incremental predictive information beyond conventional preoperative clinical, pulmonary, and biomarker variables. These findings support SII as a clinically accessible adjunctive risk-stratification marker rather than a stand-alone decision tool; external or temporal validation is required before routine implementation. Introduction Acute type A aortic dissection (ATAAD) is a cardiovascular emergency characterized by abrupt aortic wall injury, hemodynamic instability, coagulation activation, and systemic inflammation [1]. Although timely surgical repair remains the cornerstone of treatment, postoperative hypoxemia after ATAAD surgery is a common and clinically consequential complication [2]. PRH is particularly relevant to anesthesiologists, intensivists, and aortic surgeons because severe postoperative oxygenation failure can prolong ventilatory support, increase intensive care requirements, and signal broader postoperative lung injury rather than isolated impairment in gas exchange [3]. The pathogenesis of PRH after ATAAD repair is not reducible to a single perioperative insult. Cardiopulmonary bypass can trigger a systemic inflammatory response [4]. CPB-related lung injury may be amplified by ischemia-reperfusion, pulmonary endothelial stress, and alveolar-capillary barrier dysfunction [5]. In ATAAD surgery, aortic tissue injury, deep hypothermic circulatory arrest, massive transfusion, and postoperative capillary leak may further aggravate oxygenation impairment [2]. Obesity and pre-existing pulmonary vulnerability add patient-level susceptibility to postoperative lung injury [6]. This multifactorial physiology explains why risk stratification based only on operative duration or isolated inflammatory markers is often clinically incomplete. The concept of immunothrombosis provides a useful framework for ATAAD-associated lung injury. Inflammation and coagulation are biologically coupled processes in which intravascular thrombosis can function as an effector arm of innate immunity [7]. Activated platelets can present inflammatory signals to neutrophils and promote neutrophil extracellular trap formation [8]. Platelet-neutrophil interactions have also been implicated in transfusion-related acute lung injury, supporting the plausibility of a thromboinflammatory contribution to postoperative oxygenation failure [9]. At the same time, acute stress and systemic inflammation may be accompanied by lymphocyte suppression, reflecting impaired immune regulation and heightened physiologic stress [10]. Therefore, a composite biomarker incorporating neutrophil, platelet, and lymphocyte information may better approximate the host response than single-cell ratios alone. The systemic immune-inflammation index (SII), calculated as platelet count multiplied by neutrophil count and divided by lymphocyte count, was originally described as a composite prognostic marker integrating inflammation, thrombosis, and immune regulation [11]. Its prognostic relevance has also been reported in cardiac surgical cohorts, where preoperative SII was associated with short-term postoperative complications [12]. However, its specific value in ATAAD-associated PRH requires careful evaluation. First, SII must be compared with simpler inflammatory indices, such as the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and monocyte-to-lymphocyte ratio (MLR). Second, because ATAAD may involve false-lumen thrombosis and platelet consumption, the platelet component of SII requires cautious interpretation [13]. Third, model performance must be reported in a way that avoids overfitting and does not transform a single-center retrospective biomarker study into an overconfident clinical tool. Accordingly, this revised analysis evaluated whether preoperative SII was independently associated with PRH after ATAAD surgery and whether it provided incremental predictive information beyond conventional preoperative clinical, pulmonary, and biomarker variables. We further examined the robustness of the association in expanded perioperative sensitivity models, compared SII with NLR, PLR, and MLR, evaluated calibration and decision-curve performance, and used bootstrap internal validation to estimate optimism-corrected model performance. Materials and methods Study design and population This single-center retrospective cohort study included adult patients who underwent surgical repair for ATAAD at Wuhan Asia Heart Hospital between January 2018 and December 2024. Reporting was structured according to STROBE principles for observational cohort studies [14]. The study complied with the Declaration of Helsinki and was approved by the Institutional Review Board of Wuhan Asia Heart Hospital (Approval No. 2023-B022). The requirement for written informed consent was waived because anonymized retrospective clinical data were used. A total of 565 patients with ATAAD were screened. Patients were excluded if they required preoperative mechanical ventilation (n = 15), had fever or active infection before surgery (n = 17), received long-term steroids or immunosuppressive therapy (n = 3), had chronic inflammatory disease (n = 12), or had incomplete laboratory data required to calculate SII (n = 28). The final analytic cohort included 490 patients, of whom 158 developed PRH and 332 did not (Fig. 1). Data collection and variable definitions Clinical, laboratory, operative, and outcome data were extracted from the electronic medical record system. Baseline variables included age, sex, body mass index (BMI), smoking and drinking history, hypertension, diabetes, preoperative blood pressure, heart rate, left ventricular ejection fraction, pericardial effusion, aortic regurgitation, aortic dissection rupture, chronic lung disease history, preoperative oxygen saturation (SpO2), and Sequential Organ Failure Assessment (SOFA) score. Chronic lung disease was defined as documented chronic obstructive pulmonary disease or asthma. Preoperative SpO2 was defined as the earliest documented oxygen saturation after hospital presentation and before operative transfer. Laboratory indices were obtained from the first available preoperative blood tests. SII was calculated as platelet count x neutrophil count / lymphocyte count using absolute peripheral blood counts expressed in 10^9/L. Because SII is a composite index rather than a directly interpretable concentration, it was reported as an index without unit notation. NLR, PLR, and MLR were calculated using absolute cell counts and were also reported as indices. Additional laboratory markers included white blood cell count, neutrophil and lymphocyte counts, platelet count, fasting blood glucose, creatinine, blood urea nitrogen, D-dimer, high-sensitivity C-reactive protein, cardiac troponin I, and procalcitonin. Operative and perioperative variables included emergency surgery, cardiopulmonary bypass (CPB) time, aortic cross-clamp (ACC) time, total operation time, dexamethasone exposure, red blood cell transfusion, plasma transfusion, and platelet transfusion. Ulinastatin was not routinely used in the represented center during the study period and therefore was not included as an analyzable exposure. Because transfusion and treatment variables may occur after the preoperative risk state has already been established and may reflect bleeding burden, disease severity, treatment-selection bias, or timing bias, these variables were handled as expanded sensitivity/explanatory covariates rather than pure preoperative predictors. Outcome definition The primary outcome was PRH. PRH was defined as severe postoperative oxygenation impairment, operationalized as PaO2/FiO2 < 100 mmHg under high-level respiratory support, including positive end-expiratory pressure (PEEP) > = 5 cmH2O and high inspired oxygen concentration, after exclusion of dominant alternative causes such as cardiogenic pulmonary edema, established pneumonia, or pulmonary embolism. To improve reproducibility, PRH adjudication was based on postoperative arterial blood gas and respiratory support records together with clinical documentation and imaging findings. The final submitted version should ensure that the exact postoperative time window and measurement interval used in the local dataset are consistent with the source database. Model strategy The main analysis focused on the independent association and incremental predictive value of preoperative SII. SII was modeled primarily as log2-transformed SII; therefore, ORs represent the change in PRH odds associated with each doubling of SII. This approach was chosen to reduce the influence of scale ambiguity and to avoid unstable estimates generated by arbitrary raw-unit increments. Two model systems were prespecified. The primary preoperative prediction model included variables available before surgery or at admission: age, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, cTnI, and SII. An expanded perioperative sensitivity model added emergency surgery, CPB time, ACC time, operation time, dexamethasone, red blood cell transfusion, plasma transfusion, and any platelet transfusion. The expanded model was used to test whether the SII association persisted after accounting for operative burden and perioperative management variables, but it was not interpreted as a strictly preoperative deployment model. Statistical analysis Continuous variables were summarized as mean +/- standard deviation or median (interquartile range), depending on distribution. Categorical variables were summarized as n (%). Between-group comparisons were performed using the Student t test or Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables, as appropriate. Multivariable logistic regression was used to estimate the association between SII and PRH. Sequential models were constructed to show how the SII estimate changed after adjustment for demographic, pulmonary, operative, and perioperative covariates. Sensitivity analyses evaluated SII using a data-derived cutoff and quartiles, whereas the continuous log2-transformed SII analysis was retained as the primary specification. The prediction-model components were reported according to TRIPOD recommendations, including explicit separation of the base clinical model from marker-augmented models [15]. Model construction followed established clinical prediction-model principles, with attention to predictor timing, model specification, and internal validation [16]. Logistic regression modeling and bootstrap validation were guided by regression-modeling recommendations for avoiding overly optimistic apparent performance estimates [17]. Calibration was evaluated because discrimination alone cannot determine whether predicted probabilities are clinically reliable [18]. Calibration slope, calibration intercept, Brier score, and apparent decile calibration were reported to characterize agreement between predicted and observed PRH risk [19]. Incremental predictive value was examined by comparing the base preoperative model with models adding log-transformed SII, NLR, PLR, or MLR; continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used as complementary metrics of added predictive information [20]. AUC confidence intervals were estimated by bootstrap resampling, and DeLong tests were used to compare each marker-augmented model with the base model. Decision curve analysis was performed to estimate clinical net benefit across clinically relevant risk thresholds. All analyses were performed using Python 3.9 and R version 4.2. A two-sided P value < 0.05 was considered statistically significant. Results Study population and baseline characteristics Among 565 screened patients with ATAAD, 75 were excluded according to predefined criteria, leaving 490 patients in the final analytic cohort (Fig. 1). PRH developed in 158 patients (32.2%), whereas 332 patients did not develop PRH. Baseline and perioperative characteristics are shown in Table 1. Patients who developed PRH had a more adverse preoperative and perioperative profile than those without PRH. BMI was higher in the PRH group (26.5+/-3.1 vs. 24.9+/-3.1 kg/m2, P < 0.001), and preoperative SpO2 was lower (96.8+/-1.1% vs. 97.7+/-1.1%, P < 0.001). Heart rate, emergency surgery, pericardial effusion, SOFA score, creatinine, D-dimer, high-sensitivity C-reactive protein, red blood cell transfusion volume, and plasma transfusion volume were also higher or more frequent in patients with PRH. Operative burden was greater in the PRH group, reflected by longer CPB time (192.5+/-23.8 vs. 176.3+/-27.2 min, P < 0.001), longer ACC time (77.0+/-10.0 vs. 74.6+/-11.0 min, P = 0.016), and longer total operation time (334.8+/-42.5 vs. 319.4+/-42.2 min, P < 0.001). Inflammatory and immune-cell profiles differed substantially by PRH status. Median SII was nearly threefold higher in patients with PRH than in those without PRH [1899.5 (1140.1-3043.6) vs. 667.5 (408.7-1173.7), P < 0.001]. This difference was driven mainly by higher neutrophil counts and lower lymphocyte counts, whereas platelet count alone did not differ significantly between groups [177.8 (147.9-203.5) vs. 183.1 (152.6-214.9) x10^9/L, P = 0.121]. NLR, PLR, and MLR were also higher in the PRH group, supporting a broad pattern of preoperative inflammatory activation and lymphocyte suppression rather than an isolated platelet-count effect. Association between SII and PRH in staged models SII remained consistently associated with PRH across staged multivariable models (Table 2). In the demographic model adjusted for age, sex, and BMI, each doubling of SII was associated with a 3.05-fold increase in the odds of PRH (95% CI, 2.42–3.84; P < 0.001; AUC, 0.827). After additional adjustment for preoperative SpO2, chronic lung disease, D-dimer, and cTnI in the primary preoperative prediction model, the association remained strong [adjusted OR, 2.80 (95% CI, 2.20–3.56); P < 0.001; AUC, 0.846]. The estimate was not materially weakened after operative variables were added. In the model further adjusted for emergency surgery, CPB time, ACC time, and operation time, each doubling of SII was associated with an adjusted OR of 2.85 (95% CI, 2.21–3.67; P < 0.001; AUC, 0.862). In the expanded perioperative sensitivity model, which additionally accounted for dexamethasone and transfusion variables, the association remained directionally and statistically robust [adjusted OR, 2.96 (95% CI, 2.27–3.87); P < 0.001; AUC, 0.870]. Expanded perioperative sensitivity model The final expanded perioperative sensitivity model is shown in Table 3. Besides SII, BMI, preoperative SpO2, and CPB time retained clinically interpretable associations with PRH. Each 1-kg/m2 increase in BMI was associated with higher PRH odds [adjusted OR, 1.13 (95% CI, 1.04–1.23); P = 0.003], whereas each 1% higher preoperative SpO2 was associated with lower PRH odds [adjusted OR, 0.57 (95% CI, 0.45–0.74); P < 0.001]. CPB time also remained associated with PRH when scaled per 30 min [adjusted OR, 1.67 (95% CI, 1.17–2.38); P = 0.005]. The platelet-transfusion coefficient should not be interpreted causally. Any platelet transfusion showed an inverse association in the expanded model [adjusted OR, 0.49 (95% CI, 0.29–0.82); P = 0.007], but this variable was included only as a sensitivity/explanatory covariate. Its coefficient may reflect treatment-selection bias, bleeding pattern, timing of transfusion, unmeasured coagulation status, or perioperative management heterogeneity. Accordingly, the primary preoperative model remains the clinically cleaner model for risk stratification. Alternative SII specifications and incremental prediction Sensitivity analyses using alternative SII specifications supported the primary continuous analysis (Table 4). When high SII was defined using the data-derived cutoff of 1085.4, high SII was associated with markedly higher odds of PRH [adjusted OR, 7.58 (95% CI, 4.70-12.21); P < 0.001]. Quartile analysis showed a graded increase in risk: compared with Q1, adjusted ORs were 3.79 for Q2, 9.78 for Q3, and 28.88 for Q4 (all P < = 0.007) (Table 5). These categorical analyses provide clinical interpretability, but the continuous log2-transformed analysis was considered the primary specification because it uses the full distribution of SII and avoids overreliance on a single data-derived threshold. The adjusted dose-response curve demonstrated a broadly increasing relationship between preoperative SII and predicted PRH probability across the observed SII range, with observed decile event rates overlaid (Fig. 2). Unlike the earlier analysis, the revised model did not emphasize a specific inflection point or named curve shape, because more complete adjustment attenuated the previous non-linear signal. This more conservative interpretation better reflects the current data and avoids overfitting a visually attractive but potentially unstable threshold. When inflammatory indices were compared on the same base preoperative model, all marker-augmented models improved discrimination (Table 6; Fig. 3). The base model had an AUC of 0.750 (95% CI, 0.705–0.794). Adding log(SII) increased the AUC to 0.846 (95% CI, 0.810–0.880), with a delta AUC of 0.096, continuous NRI of 0.882, and IDI of 0.164. Adding log(NLR), log(PLR), and log(MLR) yielded AUCs of 0.839, 0.818, and 0.806, respectively. Thus, SII showed the numerically highest discrimination in this cohort, although its advantage over NLR was modest and should be interpreted as incremental rather than definitive superiority. Subgroup, decision-curve, calibration, and secondary outcome analyses Subgroup analysis showed directionally consistent associations between high SII and PRH across age, sex, BMI, emergency surgery, and plasma transfusion strata (Fig. 4). No subgroup suggested a qualitative reversal of the association, supporting the robustness of the SII-PRH relationship across clinically relevant patient subsets. The subgroup results should nevertheless be viewed as exploratory because the study was not powered to detect small interaction effects. Decision curve analysis demonstrated that the SII-augmented model provided greater net benefit than the base preoperative model over a clinically relevant range of threshold probabilities, particularly across the 10%-40% interval highlighted in Fig. 5. This suggests that SII may help refine preoperative respiratory risk stratification when clinicians are deciding whether a patient warrants closer postoperative respiratory monitoring, intensified lung-protective planning, or more cautious ICU resource allocation. It does not imply that SII alone should trigger a specific intervention. Internal calibration of the SII-augmented model is shown in Fig. 6. Bootstrap internal validation yielded an optimism-corrected AUC of 0.834, calibration slope of 1.00, calibration intercept of 0.00, and Brier score of 0.153. These findings indicate acceptable internal performance but do not substitute for external validation or temporal validation in later surgical cohorts. Secondary outcomes stratified by the SII cutoff are summarized in Table 7. Compared with patients in the low-SII group, patients in the high-SII group had longer mechanical ventilation [46.3 (25.6–69.8) vs. 24.5 (15.4–44.9) hours, P < 0.001], longer ICU stay [5.5 (3.8–7.6) vs. 4.0 (2.8–5.7) days, P < 0.001], longer hospital stay [16.7 (13.4–22.0) vs. 14.2 (11.4–18.7) days, P < 0.001], higher AKI incidence (23.2% vs. 11.7%, P < 0.001), and higher in-hospital mortality (10.1% vs. 4.6%, P = 0.017). These associations indicate that high SII identifies a higher-risk clinical phenotype, although causality cannot be inferred from the secondary outcome comparisons. Fig. 1Study flowchart.Flow diagram showing the screening of patients with acute type A aortic dissection and the final analytic cohort according to postoperative refractory hypoxemia status. ATAAD, acute type A aortic dissection; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Study flowchart.Flow diagram showing the screening of patients with acute type A aortic dissection and the final analytic cohort according to postoperative refractory hypoxemia status. ATAAD, acute type A aortic dissection; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Table 1Baseline and perioperative characteristics according to PRH statusVariableNo PRH (n = 332)PRH (n = 158)P valueAge, years61.7+/-11.562.8+/-11.20.303BMI, kg/m224.9+/-3.126.5+/-3.1< 0.001SBP, mmHg141.4+/-20.2140.2+/-20.60.552DBP, mmHg74.5+/-10.674.8+/-11.10.760Heart rate, bpm76.5+/-15.881.2+/-17.00.003LVEF, %54.9+/-7.254.3+/-8.30.502Preoperative SpO2, %97.7+/-1.196.8+/-1.1< 0.001CPB time, min176.3+/-27.2192.5+/-23.8< 0.001ACC time, min74.6+/-11.077.0+/-10.00.016Operation time, min319.4+/-42.2334.8+/-42.5< 0.001Male240 (72.3)113 (71.5)0.859Hypertension195 (58.7)103 (65.2)0.171Diabetes47 (14.2)22 (13.9)0.945Smoking186 (56.0)95 (60.1)0.391Drinking144 (43.4)73 (46.2)0.556Emergency surgery195 (58.7)111 (70.3)0.014Pericardial effusion28 (8.4)29 (18.4)0.001Aortic regurgitation60 (18.1)37 (23.4)0.165Aortic dissection rupture12 (3.6)4 (2.5)0.528Chronic lung disease (COPD/asthma history)17 (5.1)14 (8.9)0.112Dexamethasone143 (43.1)72 (45.6)0.603Any platelet transfusion188 (56.6)93 (58.9)0.640SOFA score3 (2–4)3 (3–4)0.019WBC, 10^9/L7.71 (6.50–9.62)10.50 (8.77–13.78)< 0.001Neutrophil count, 10^9/L5.48 (3.98–7.48)8.73 (6.90-11.96)< 0.001Lymphocyte count, 10^9/L1.38 (1.02–2.02)0.89 (0.62–1.15)< 0.001Platelets, 10^9/L183.1 (152.6-214.9)177.8 (147.9-203.5)0.121SII667.5 (408.7-1173.7)1899.5 (1140.1-3043.6)< 0.001NLR3.97 (2.25–6.40)9.88 (6.47–18.41)< 0.001PLR124.6 (90.6-174.8)206.1 (153.3–278.0)< 0.001MLR0.37 (0.23–0.56)0.68 (0.44–1.03)< 0.001FBG, mmol/L5.84 (5.07–6.55)6.06 (5.24–6.67)0.195Creatinine, umol/L80.8 (68.5–99.8)92.6 (75.1–110.0)< 0.001D-dimer, mg/L FEU31.8 (22.4–54.9)47.2 (29.5–71.8)< 0.001hs-CRP, mg/L27.4 (16.3–45.5)33.3 (20.0-57.4)0.007cTnI, ng/mL0.053 (0.028–0.097)0.057 (0.033–0.118)0.078PCT, ng/mL0.19 (0.12–0.30)0.21 (0.12–0.33)0.336RBC transfusion, units4.0 (3.0–6.0)5.5 (4.0–7.0)< 0.001Plasma transfusion, mL500 (300–650)650 (500–850)< 0.001Values are presented as mean +/- SD, median (interquartile range), or n (%), as appropriate. P values were calculated using the Student t test or Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. SII, NLR, PLR, and MLR are reported as indices without unit notation. ACC, aortic cross-clamp; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; DBP, diastolic blood pressure; FBG, fasting blood glucose; FEU, fibrinogen-equivalent units; hs-CRP, high-sensitivity C-reactive protein; LVEF, left ventricular ejection fraction; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; PCT, procalcitonin; PLR, platelet-to-lymphocyte ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SBP, systolic blood pressure; SII, systemic immune-inflammation index; SOFA, Sequential Organ Failure Assessment; SpO2, peripheral oxygen saturation; WBC, white blood cell count Table 1Baseline and perioperative characteristics according to PRH statusVariableNo PRH (n = 332)PRH (n = 158)P valueAge, years61.7+/-11.562.8+/-11.20.303BMI, kg/m224.9+/-3.126.5+/-3.1< 0.001SBP, mmHg141.4+/-20.2140.2+/-20.60.552DBP, mmHg74.5+/-10.674.8+/-11.10.760Heart rate, bpm76.5+/-15.881.2+/-17.00.003LVEF, %54.9+/-7.254.3+/-8.30.502Preoperative SpO2, %97.7+/-1.196.8+/-1.1< 0.001CPB time, min176.3+/-27.2192.5+/-23.8< 0.001ACC time, min74.6+/-11.077.0+/-10.00.016Operation time, min319.4+/-42.2334.8+/-42.5< 0.001Male240 (72.3)113 (71.5)0.859Hypertension195 (58.7)103 (65.2)0.171Diabetes47 (14.2)22 (13.9)0.945Smoking186 (56.0)95 (60.1)0.391Drinking144 (43.4)73 (46.2)0.556Emergency surgery195 (58.7)111 (70.3)0.014Pericardial effusion28 (8.4)29 (18.4)0.001Aortic regurgitation60 (18.1)37 (23.4)0.165Aortic dissection rupture12 (3.6)4 (2.5)0.528Chronic lung disease (COPD/asthma history)17 (5.1)14 (8.9)0.112Dexamethasone143 (43.1)72 (45.6)0.603Any platelet transfusion188 (56.6)93 (58.9)0.640SOFA score3 (2–4)3 (3–4)0.019WBC, 10^9/L7.71 (6.50–9.62)10.50 (8.77–13.78)< 0.001Neutrophil count, 10^9/L5.48 (3.98–7.48)8.73 (6.90-11.96)< 0.001Lymphocyte count, 10^9/L1.38 (1.02–2.02)0.89 (0.62–1.15)< 0.001Platelets, 10^9/L183.1 (152.6-214.9)177.8 (147.9-203.5)0.121SII667.5 (408.7-1173.7)1899.5 (1140.1-3043.6)< 0.001NLR3.97 (2.25–6.40)9.88 (6.47–18.41)< 0.001PLR124.6 (90.6-174.8)206.1 (153.3–278.0)< 0.001MLR0.37 (0.23–0.56)0.68 (0.44–1.03)< 0.001FBG, mmol/L5.84 (5.07–6.55)6.06 (5.24–6.67)0.195Creatinine, umol/L80.8 (68.5–99.8)92.6 (75.1–110.0)< 0.001D-dimer, mg/L FEU31.8 (22.4–54.9)47.2 (29.5–71.8)< 0.001hs-CRP, mg/L27.4 (16.3–45.5)33.3 (20.0-57.4)0.007cTnI, ng/mL0.053 (0.028–0.097)0.057 (0.033–0.118)0.078PCT, ng/mL0.19 (0.12–0.30)0.21 (0.12–0.33)0.336RBC transfusion, units4.0 (3.0–6.0)5.5 (4.0–7.0)< 0.001Plasma transfusion, mL500 (300–650)650 (500–850)< 0.001Values are presented as mean +/- SD, median (interquartile range), or n (%), as appropriate. P values were calculated using the Student t test or Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. SII, NLR, PLR, and MLR are reported as indices without unit notation. ACC, aortic cross-clamp; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; DBP, diastolic blood pressure; FBG, fasting blood glucose; FEU, fibrinogen-equivalent units; hs-CRP, high-sensitivity C-reactive protein; LVEF, left ventricular ejection fraction; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; PCT, procalcitonin; PLR, platelet-to-lymphocyte ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SBP, systolic blood pressure; SII, systemic immune-inflammation index; SOFA, Sequential Organ Failure Assessment; SpO2, peripheral oxygen saturation; WBC, white blood cell count Baseline and perioperative characteristics according to PRH status Values are presented as mean +/- SD, median (interquartile range), or n (%), as appropriate. P values were calculated using the Student t test or Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. SII, NLR, PLR, and MLR are reported as indices without unit notation. ACC, aortic cross-clamp; BMI, body mass index; COPD, chronic obstructive pulmonary disease; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; DBP, diastolic blood pressure; FBG, fasting blood glucose; FEU, fibrinogen-equivalent units; hs-CRP, high-sensitivity C-reactive protein; LVEF, left ventricular ejection fraction; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; PCT, procalcitonin; PLR, platelet-to-lymphocyte ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SBP, systolic blood pressure; SII, systemic immune-inflammation index; SOFA, Sequential Organ Failure Assessment; SpO2, peripheral oxygen saturation; WBC, white blood cell count Table 2Staged multivariable logistic regression models for SII and PRHModelAdjustment strategySII per doubling, OR (95% CI)P valueAUCModel 1: demographic modelAge, sex, BMI3.05 (2.42–3.84)< 0.0010.827Model 2: primary preoperative prediction modelAge, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, cTnI2.80 (2.20–3.56)< 0.0010.846Model 3: + operative variablesPrimary model + emergency surgery, CPB time, ACC time, operation time2.85 (2.21–3.67)< 0.0010.862Model 4: expanded perioperative sensitivity modelModel 3 + dexamethasone, RBC transfusion, plasma transfusion, any platelet transfusion2.96 (2.27–3.87)< 0.0010.870SII was modeled as log2-transformed SII; therefore, ORs correspond to the change in PRH odds associated with each doubling of SII. The AUC reflects discrimination of the corresponding model including SII. The primary preoperative model was considered the main risk-stratification model because all included covariates were available before or at the time of surgery. ACC, aortic cross-clamp; AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; OR, odds ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Table 2Staged multivariable logistic regression models for SII and PRHModelAdjustment strategySII per doubling, OR (95% CI)P valueAUCModel 1: demographic modelAge, sex, BMI3.05 (2.42–3.84)< 0.0010.827Model 2: primary preoperative prediction modelAge, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, cTnI2.80 (2.20–3.56)< 0.0010.846Model 3: + operative variablesPrimary model + emergency surgery, CPB time, ACC time, operation time2.85 (2.21–3.67)< 0.0010.862Model 4: expanded perioperative sensitivity modelModel 3 + dexamethasone, RBC transfusion, plasma transfusion, any platelet transfusion2.96 (2.27–3.87)< 0.0010.870SII was modeled as log2-transformed SII; therefore, ORs correspond to the change in PRH odds associated with each doubling of SII. The AUC reflects discrimination of the corresponding model including SII. The primary preoperative model was considered the main risk-stratification model because all included covariates were available before or at the time of surgery. ACC, aortic cross-clamp; AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; OR, odds ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Staged multivariable logistic regression models for SII and PRH SII was modeled as log2-transformed SII; therefore, ORs correspond to the change in PRH odds associated with each doubling of SII. The AUC reflects discrimination of the corresponding model including SII. The primary preoperative model was considered the main risk-stratification model because all included covariates were available before or at the time of surgery. ACC, aortic cross-clamp; AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; OR, odds ratio; PRH, postoperative refractory hypoxemia; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Table 3Final expanded perioperative sensitivity modelVariableAdjusted OR (95% CI)P valueInterpretation noteAge, per 10 years1.24 (0.99–1.56)0.063Male0.97 (0.56–1.67)0.907BMI, per kg/m21.13 (1.04–1.23)0.003Preoperative SpO2, per 1%0.57 (0.45–0.74)< 0.001Chronic lung disease0.68 (0.25–1.83)0.442D-dimer, per 10 mg/L FEU1.02 (0.96–1.09)0.532cTnI, per 0.01 ng/mL1.00 (0.97–1.04)0.813SII, per doubling2.96 (2.27–3.87)< 0.001Primary exposureEmergency surgery1.12 (0.66–1.89)0.683CPB time, per 30 min1.67 (1.17–2.38)0.005ACC time, per 30 min0.80 (0.39–1.64)0.544Operation time, per 30 min0.87 (0.71–1.06)0.173Dexamethasone1.14 (0.69–1.88)0.615Sensitivity covariateRBC transfusion, per unit1.07 (0.95–1.20)0.293Sensitivity covariatePlasma transfusion, per 100 mL1.06 (0.95–1.19)0.299Sensitivity covariateAny platelet transfusion0.49 (0.29–0.82)0.007Sensitivity covariate onlyThis model was designed as an expanded perioperative sensitivity analysis rather than a pure preoperative deployment model. Transfusion and therapy variables may be influenced by bleeding burden, clinical severity, timing of exposure, and treatment-selection bias; therefore, their coefficients should not be interpreted as causal effects. Any platelet transfusion should not be interpreted as protective. ACC, aortic cross-clamp; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; FEU, fibrinogen-equivalent units; OR, odds ratio; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Table 3Final expanded perioperative sensitivity modelVariableAdjusted OR (95% CI)P valueInterpretation noteAge, per 10 years1.24 (0.99–1.56)0.063Male0.97 (0.56–1.67)0.907BMI, per kg/m21.13 (1.04–1.23)0.003Preoperative SpO2, per 1%0.57 (0.45–0.74)< 0.001Chronic lung disease0.68 (0.25–1.83)0.442D-dimer, per 10 mg/L FEU1.02 (0.96–1.09)0.532cTnI, per 0.01 ng/mL1.00 (0.97–1.04)0.813SII, per doubling2.96 (2.27–3.87)< 0.001Primary exposureEmergency surgery1.12 (0.66–1.89)0.683CPB time, per 30 min1.67 (1.17–2.38)0.005ACC time, per 30 min0.80 (0.39–1.64)0.544Operation time, per 30 min0.87 (0.71–1.06)0.173Dexamethasone1.14 (0.69–1.88)0.615Sensitivity covariateRBC transfusion, per unit1.07 (0.95–1.20)0.293Sensitivity covariatePlasma transfusion, per 100 mL1.06 (0.95–1.19)0.299Sensitivity covariateAny platelet transfusion0.49 (0.29–0.82)0.007Sensitivity covariate onlyThis model was designed as an expanded perioperative sensitivity analysis rather than a pure preoperative deployment model. Transfusion and therapy variables may be influenced by bleeding burden, clinical severity, timing of exposure, and treatment-selection bias; therefore, their coefficients should not be interpreted as causal effects. Any platelet transfusion should not be interpreted as protective. ACC, aortic cross-clamp; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; FEU, fibrinogen-equivalent units; OR, odds ratio; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Final expanded perioperative sensitivity model This model was designed as an expanded perioperative sensitivity analysis rather than a pure preoperative deployment model. Transfusion and therapy variables may be influenced by bleeding burden, clinical severity, timing of exposure, and treatment-selection bias; therefore, their coefficients should not be interpreted as causal effects. Any platelet transfusion should not be interpreted as protective. ACC, aortic cross-clamp; BMI, body mass index; CI, confidence interval; CPB, cardiopulmonary bypass; cTnI, cardiac troponin I; FEU, fibrinogen-equivalent units; OR, odds ratio; RBC, red blood cell; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Table 4Sensitivity analyses using alternative SII specificationsAnalysisSII specificationAdjustment setEffect estimateP valuePrimary continuous analysislog2-transformed SIIPrimary preoperative prediction model2.80 (2.20–3.56)< 0.001Expanded perioperative sensitivity modellog2-transformed SIIPrimary model + perioperative variables2.96 (2.27–3.87)< 0.001Clinically interpretable cutoff modelHigh SII > 1085.4Primary preoperative prediction model7.58 (4.70-12.21)< 0.001Quartile modelQ4 vs. Q1; cut points 502.7, 921.7, 1796.7Primary preoperative prediction model28.88 (11.16–74.79)< 0.001Effect estimates are adjusted ORs with 95% CIs. Cutoff and quartile analyses were prespecified as secondary clinically interpretable analyses. The continuous log2-transformed SII model was retained as the primary analysis to minimize information loss and reduce dependence on data-derived thresholds. CI, confidence interval; OR, odds ratio; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Table 4Sensitivity analyses using alternative SII specificationsAnalysisSII specificationAdjustment setEffect estimateP valuePrimary continuous analysislog2-transformed SIIPrimary preoperative prediction model2.80 (2.20–3.56)< 0.001Expanded perioperative sensitivity modellog2-transformed SIIPrimary model + perioperative variables2.96 (2.27–3.87)< 0.001Clinically interpretable cutoff modelHigh SII > 1085.4Primary preoperative prediction model7.58 (4.70-12.21)< 0.001Quartile modelQ4 vs. Q1; cut points 502.7, 921.7, 1796.7Primary preoperative prediction model28.88 (11.16–74.79)< 0.001Effect estimates are adjusted ORs with 95% CIs. Cutoff and quartile analyses were prespecified as secondary clinically interpretable analyses. The continuous log2-transformed SII model was retained as the primary analysis to minimize information loss and reduce dependence on data-derived thresholds. CI, confidence interval; OR, odds ratio; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Sensitivity analyses using alternative SII specifications Effect estimates are adjusted ORs with 95% CIs. Cutoff and quartile analyses were prespecified as secondary clinically interpretable analyses. The continuous log2-transformed SII model was retained as the primary analysis to minimize information loss and reduce dependence on data-derived thresholds. CI, confidence interval; OR, odds ratio; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Table 5SII quartile analysisSII quartileAdjusted OR vs. Q1P valueQ2 (502.7 < SII < = 921.7)3.79 (1.44–10.01)0.007Q3 (921.7 < SII < = 1796.7)9.78 (3.83–24.98)< 0.001Q4 (SII > 1796.7)28.88 (11.16–74.79)< 0.001Q1 served as the reference group. Models were adjusted for the primary preoperative prediction covariates. The quartile analysis demonstrates a graded association across the observed SII distribution but should be interpreted as supportive of, not superior to, the primary continuous log2(SII) analysis. OR, odds ratio; SII, systemic immune-inflammation index Table 5SII quartile analysisSII quartileAdjusted OR vs. Q1P valueQ2 (502.7 < SII < = 921.7)3.79 (1.44–10.01)0.007Q3 (921.7 < SII < = 1796.7)9.78 (3.83–24.98)< 0.001Q4 (SII > 1796.7)28.88 (11.16–74.79)< 0.001Q1 served as the reference group. Models were adjusted for the primary preoperative prediction covariates. The quartile analysis demonstrates a graded association across the observed SII distribution but should be interpreted as supportive of, not superior to, the primary continuous log2(SII) analysis. OR, odds ratio; SII, systemic immune-inflammation index SII quartile analysis Q1 served as the reference group. Models were adjusted for the primary preoperative prediction covariates. The quartile analysis demonstrates a graded association across the observed SII distribution but should be interpreted as supportive of, not superior to, the primary continuous log2(SII) analysis. OR, odds ratio; SII, systemic immune-inflammation index Table 6Incremental predictive value of inflammatory indices added to the base modelModelAUC (95% CI)Delta AUC vs. baseDeLong P vs. baseContinuous NRIIDIBase primary preoperative model0.750 (0.705–0.794)RefRefRefRefBase + log(NLR)0.839 (0.803–0.874)0.089< 0.0010.7880.148Base + log(PLR)0.818 (0.774–0.855)0.068< 0.0010.7860.107Base + log(MLR)0.806 (0.765–0.844)0.057< 0.0010.6100.087Base + log(SII)0.846 (0.810–0.880)0.096< 0.0010.8820.164The base primary preoperative model included age, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, and cTnI. AUC confidence intervals were obtained by bootstrap resampling. DeLong P values compare each marker-augmented model with the base model, not with each other. Therefore, the numerically higher AUC for SII than for NLR should be interpreted cautiously unless a direct SII-vs-NLR comparison is performed. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; cTnI, cardiac troponin I; IDI, integrated discrimination improvement; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio [note: NLR in model row denotes neutrophil-to-lymphocyte ratio]; NRI, net reclassification improvement; PLR, platelet-to-lymphocyte ratio; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Table 6Incremental predictive value of inflammatory indices added to the base modelModelAUC (95% CI)Delta AUC vs. baseDeLong P vs. baseContinuous NRIIDIBase primary preoperative model0.750 (0.705–0.794)RefRefRefRefBase + log(NLR)0.839 (0.803–0.874)0.089< 0.0010.7880.148Base + log(PLR)0.818 (0.774–0.855)0.068< 0.0010.7860.107Base + log(MLR)0.806 (0.765–0.844)0.057< 0.0010.6100.087Base + log(SII)0.846 (0.810–0.880)0.096< 0.0010.8820.164The base primary preoperative model included age, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, and cTnI. AUC confidence intervals were obtained by bootstrap resampling. DeLong P values compare each marker-augmented model with the base model, not with each other. Therefore, the numerically higher AUC for SII than for NLR should be interpreted cautiously unless a direct SII-vs-NLR comparison is performed. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; cTnI, cardiac troponin I; IDI, integrated discrimination improvement; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio [note: NLR in model row denotes neutrophil-to-lymphocyte ratio]; NRI, net reclassification improvement; PLR, platelet-to-lymphocyte ratio; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Incremental predictive value of inflammatory indices added to the base model The base primary preoperative model included age, sex, BMI, preoperative SpO2, chronic lung disease, D-dimer, and cTnI. AUC confidence intervals were obtained by bootstrap resampling. DeLong P values compare each marker-augmented model with the base model, not with each other. Therefore, the numerically higher AUC for SII than for NLR should be interpreted cautiously unless a direct SII-vs-NLR comparison is performed. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; cTnI, cardiac troponin I; IDI, integrated discrimination improvement; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio [note: NLR in model row denotes neutrophil-to-lymphocyte ratio]; NRI, net reclassification improvement; PLR, platelet-to-lymphocyte ratio; SII, systemic immune-inflammation index; SpO2, peripheral oxygen saturation Fig. 2Adjusted dose-response association between preoperative SII and PRH.The curve was adjusted for variables in the primary preoperative model, and observed decile event rates were overlaid to show the empirical distribution of PRH across the SII range. The revised model supports a broadly increasing association but does not rely on a specific inflection point or named curve shape. PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Adjusted dose-response association between preoperative SII and PRH.The curve was adjusted for variables in the primary preoperative model, and observed decile event rates were overlaid to show the empirical distribution of PRH across the SII range. The revised model supports a broadly increasing association but does not rely on a specific inflection point or named curve shape. PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Fig. 3ROC curves comparing the primary base preoperative model with inflammatory-index-augmented models. All inflammatory-index-augmented models improved discrimination compared with the base preoperative model. SII produced the numerically highest AUC, although its advantage over NLR was modest. AUC, area under the curve; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; ROC, receiver operating characteristic; SII, systemic immune-inflammation index ROC curves comparing the primary base preoperative model with inflammatory-index-augmented models. All inflammatory-index-augmented models improved discrimination compared with the base preoperative model. SII produced the numerically highest AUC, although its advantage over NLR was modest. AUC, area under the curve; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; ROC, receiver operating characteristic; SII, systemic immune-inflammation index Fig. 4Subgroup forest plot for high SII and PRH risk.Odds ratios for high SII were directionally consistent across clinically relevant subgroups, including age, sex, BMI, emergency surgery, and plasma-transfusion strata. Subgroup analyses were exploratory and should not be interpreted as definitive interaction tests. BMI, body mass index; OR, odds ratio; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Subgroup forest plot for high SII and PRH risk.Odds ratios for high SII were directionally consistent across clinically relevant subgroups, including age, sex, BMI, emergency surgery, and plasma-transfusion strata. Subgroup analyses were exploratory and should not be interpreted as definitive interaction tests. BMI, body mass index; OR, odds ratio; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Fig. 5Decision curve analysis comparing the primary base model and the SII-augmented model.The SII-augmented model showed greater net benefit than the base preoperative model across the highlighted clinically relevant threshold probability range of approximately 10%-40%. Decision-curve analysis was used to evaluate potential clinical usefulness, not to define an intervention threshold. SII, systemic immune-inflammation index Decision curve analysis comparing the primary base model and the SII-augmented model.The SII-augmented model showed greater net benefit than the base preoperative model across the highlighted clinically relevant threshold probability range of approximately 10%-40%. Decision-curve analysis was used to evaluate potential clinical usefulness, not to define an intervention threshold. SII, systemic immune-inflammation index Fig. 6Internal calibration of the SII-augmented primary model.Apparent decile calibration is shown together with bootstrap-derived internal validation metrics, including optimism-corrected AUC, calibration slope, calibration intercept, and Brier score. Internal validation supports model consistency within the study dataset but does not replace external or temporal validation. AUC, area under the receiver operating characteristic curve; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Internal calibration of the SII-augmented primary model.Apparent decile calibration is shown together with bootstrap-derived internal validation metrics, including optimism-corrected AUC, calibration slope, calibration intercept, and Brier score. Internal validation supports model consistency within the study dataset but does not replace external or temporal validation. AUC, area under the receiver operating characteristic curve; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Table 7Secondary clinical outcomes stratified by SII groupOutcomeLow SII ( < = 1085.4) (n = 283)High SII (> 1085.4) (n = 207)P valueMechanical ventilation, hours24.5 (15.4–44.9)46.3 (25.6–69.8)< 0.001ICU length of stay, days4.0 (2.8–5.7)5.5 (3.8–7.6)< 0.001Hospital length of stay, days14.2 (11.4–18.7)16.7 (13.4–22.0)< 0.001PRH37 (13.1)121 (58.5)< 0.001AKI33 (11.7)48 (23.2)< 0.001In-hospital mortality13 (4.6)21 (10.1)0.017Pulmonary infection85 (30.0)75 (36.2)0.149Gastrointestinal bleeding30 (10.6)18 (8.7)0.483High SII was defined using the data-derived cutoff for clinically interpretable secondary analysis. Continuous outcomes are presented as median (interquartile range), and categorical outcomes as n (%). P values were calculated using the Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. Secondary outcome comparisons are descriptive and should not be interpreted causally. AKI, acute kidney injury; ICU, intensive care unit; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Table 7Secondary clinical outcomes stratified by SII groupOutcomeLow SII ( < = 1085.4) (n = 283)High SII (> 1085.4) (n = 207)P valueMechanical ventilation, hours24.5 (15.4–44.9)46.3 (25.6–69.8)< 0.001ICU length of stay, days4.0 (2.8–5.7)5.5 (3.8–7.6)< 0.001Hospital length of stay, days14.2 (11.4–18.7)16.7 (13.4–22.0)< 0.001PRH37 (13.1)121 (58.5)< 0.001AKI33 (11.7)48 (23.2)< 0.001In-hospital mortality13 (4.6)21 (10.1)0.017Pulmonary infection85 (30.0)75 (36.2)0.149Gastrointestinal bleeding30 (10.6)18 (8.7)0.483High SII was defined using the data-derived cutoff for clinically interpretable secondary analysis. Continuous outcomes are presented as median (interquartile range), and categorical outcomes as n (%). P values were calculated using the Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. Secondary outcome comparisons are descriptive and should not be interpreted causally. AKI, acute kidney injury; ICU, intensive care unit; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Secondary clinical outcomes stratified by SII group High SII was defined using the data-derived cutoff for clinically interpretable secondary analysis. Continuous outcomes are presented as median (interquartile range), and categorical outcomes as n (%). P values were calculated using the Mann-Whitney U test for continuous variables and the chi-square test or Fisher exact test for categorical variables. Secondary outcome comparisons are descriptive and should not be interpreted causally. AKI, acute kidney injury; ICU, intensive care unit; PRH, postoperative refractory hypoxemia; SII, systemic immune-inflammation index Discussion In this single-center retrospective cohort of patients undergoing surgery for ATAAD, preoperative SII was independently associated with PRH and provided incremental predictive information beyond conventional preoperative clinical, pulmonary, and biomarker variables. The association remained consistent after additional adjustment for operative burden, dexamethasone exposure, and transfusion variables. The revised analysis also addressed several methodological vulnerabilities of the earlier draft: it replaced unstable high-cutoff estimates with log2-transformed continuous modeling, compared SII with NLR, PLR, and MLR, added pulmonary variables to the primary model, reported bootstrap internal validation, and adopted a more restrained interpretation of dose-response findings. The clinical relevance of these findings lies in the pathobiology of ATAAD-associated lung injury. ATAAD is not merely a mechanical vascular event; inflammatory activation is already present during the acute dissection process [1]. Surgical repair then adds CPB-related systemic inflammation [4]. Pulmonary endothelial stress, ischemia-reperfusion injury, and postoperative alveolar-capillary barrier dysfunction provide a direct pathway to oxygenation failure [5]. Recent ATAAD hypoxemia literature also emphasizes the combined contribution of tissue injury, transfusion exposure, and complex perioperative management [2]. In this context, a preoperative marker that captures inflammatory and thrombotic stress before the major operative insult may help identify patients with limited pulmonary reserve or heightened susceptibility to postoperative oxygenation failure. SII may be biologically attractive because it integrates three complementary dimensions of the host response: neutrophil-mediated inflammation, platelet-related thromboinflammatory activity, and lymphocyte suppression. Neutrophils can contribute to endothelial injury, capillary leak, and neutrophil extracellular trap formation [8]. Platelets can interact with neutrophils and coagulation pathways in immunothrombotic responses [7, 9]. Lymphopenia may reflect stress-induced immune dysregulation rather than a benign hematologic fluctuation [10]. The current analysis supports this integrated interpretation because SII remained associated with PRH even after adjustment for pulmonary status, D-dimer, cTnI, and operative burden. At the same time, the platelet component of SII must be interpreted with care in ATAAD. Blood flow through the false lumen can activate the coagulation-fibrinolytic system and may contribute to platelet consumption [13]. In the present cohort, platelet count alone was not significantly higher in the PRH group. Therefore, the findings should not be read as evidence that platelet count independently drives hypoxemia. Rather, SII should be viewed as a composite immune-thrombotic-stress index in which platelet count modifies the information carried by neutrophilia and lymphopenia. This distinction prevents an oversimplified mechanistic claim and aligns the biomarker interpretation with ATAAD pathophysiology. The comparison with other inflammatory indices is also important. In the same base model, SII yielded the numerically highest AUC, NRI, and IDI among the evaluated indices. However, the AUC difference between SII and NLR was small (0.846 vs. 0.839). Therefore, the more defensible interpretation is that SII showed incremental performance and may slightly enrich the information captured by NLR, not that SII has been proven categorically superior. A direct SII-versus-NLR statistical comparison and validation in independent cohorts would be needed before making stronger claims. The revised model performance is more plausible than the near-perfect discrimination reported in the earlier analysis. AUC improvement from 0.750 to 0.846 is clinically meaningful without suggesting unrealistic separation of cases and non-cases. Bootstrap validation suggested acceptable internal consistency, and decision curve analysis indicated potential net benefit across clinically plausible thresholds. Nevertheless, internal validation cannot establish generalizability. Differences in surgical strategy, transfusion thresholds, ICU ventilation practices, oxygenation protocols, and baseline ATAAD severity across centers may alter model performance. External validation or at least temporal validation is therefore essential before translating this model into routine bedside use. Several clinical implications follow from these findings. First, SII may be useful as a preoperative adjunct to identify patients who warrant more intensive postoperative respiratory surveillance. Second, elevated SII should prompt consideration of a broader high-risk phenotype rather than a single-target intervention. Such patients may have higher BMI, lower preoperative oxygenation, greater inflammatory activation, and greater perioperative resource needs. Third, the association between high SII and longer ventilation, longer ICU stay, AKI, and mortality suggests that SII may reflect systemic vulnerability beyond pulmonary oxygenation alone. However, these secondary outcomes remain descriptive and should not be interpreted as proof that modifying SII would improve outcomes. This study has limitations. First, it was retrospective and single-center, leaving room for residual confounding and center-specific practice effects. Second, although bootstrap internal validation was performed, external and temporal validation were not available. Third, PRH adjudication depends on respiratory support settings, arterial blood gas timing, and exclusion of competing diagnoses, all of which should be standardized in future prospective studies. Fourth, therapy and transfusion variables are susceptible to timing bias and confounding by indication. Fifth, SII was measured only once preoperatively; dynamic postoperative changes in inflammatory indices may provide additional prognostic information. Finally, neutrophil extracellular traps were not measured despite their relevance to platelet-neutrophil lung injury [8, 9]. Direct endothelial dysfunction markers, including von Willebrand factor-related pathways, were also unavailable [21]. Fibrinogen-related inflammatory pathways were not measured [22]. Broader immunometabolic mediators that may connect obesity, inflammation, and postoperative lung vulnerability were also beyond the scope of this analysis [23]. Conclusions Preoperative SII was independently associated with PRH after ATAAD surgery and provided incremental predictive information beyond conventional preoperative clinical, pulmonary, and biomarker variables. SII should be interpreted as an adjunctive risk-stratification marker reflecting a composite immune-thrombotic-stress phenotype, not as a stand-alone decision tool or direct mechanistic proof. External or temporal validation is required before routine clinical implementation.