This was a multicenter retrospective study that included 106 patients treated at Wuhan Union Hospital (WHUH) between October 2020 and December 2023, 204 patients treated at Fujian Cancer Hospital (FJCH) between September 2019 and December 2024, and 35 patients treated at 900th Hospital of the Joint Service Support Force of the People’s Liberation Army of China (900th Hospital) between February 2020 and July 2023, with a total of 78 (22.6%) patients achieving pCR after nICT treatment. All enrolled patients had stage II to IV ESCC and received nICT followed by radical surgery. Inclusion criteria: (1) Pathologically confirmed diagnosis of stage II to IV esophageal cancer; (2) Radical surgery with available pathology after undergoing nICT; (3) Age ≥ 18 years. Exclusion criteria (1) Non-radical resection (R1/R2 resection) after undergoing nICT; (2) Other pathological types of EC (non-ESCC); (3) Concomitant or pre-existing malignant tumors at other sites; (4) incomplete prognostic data.

We retrospectively collected baseline data on clinical case patients, including patient demographics (age, gender, body mass index [BMI], Eastern Cooperative Oncology Group Performance Status [ECOG PS] score, hypertension, smoking, drinking), biochemical data (carcinoembryonic antigen [CEA], total bilirubin [TBIL], alanine aminotransferase [ALT], aspartate aminotransferase [AST], alkaline phosphatase [ALP], albumin [ALB], globulin [GLOB], creatinine [CRE], urea nitrogen [BUN], fasting blood glucose [FBG], total cholesterol [TC], triglycerides [TG], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C], lactate dehydrogenase [LDH], calcium [Ca], white blood cell count [WBC], hemoglobin [HGB], platelet count [PLT], neutrophil count [NEU], lymphocyte count [LYM], and monocyte count [MONO]), and tumor-related information (tumor site, tumor regression grade [TRG], tumor stage, perineural invasion, and vessel invasion). Except for post neoadjuvant therapy pathological Tumor, Node, Metastasis [ypTNM] stage and TRG, all baseline information was based on the outcome of the first hospitalization.

PNI and SII were defined as follows: SII = PLT (10^⁹/L) × NEU (10^⁹/L)/LYM (10^⁹/L); PNI = ALB (g/L) + 5 × LYM (10^⁹/L). Using the combined cohort (Combined from FJCH cohort, WHUH cohort, 900th Hospital cohort) data, we calculated the Youden index (sensitivity + specificity-1) maxima to determine the optimal cutoff values, The specific analysis process was as follows: First, the Youden index was calculated for all possible cutoff points using the R software (version 4.3.0). The value that maximized this index was selected as the optimal threshold. The results showed that the optimal cutoff value for SII was 593.590 (AUC = 0.679, 95% CI: 0.616-0.742), and the optimal cutoff value for PNI was 49.525 (AUC = 0.667, 95% CI: 0.598-0.737). We categorized SII and PNI into low SII, high SII, low PNI, and high PNI based on the optimal cutoff values. Based on this, patients were categorized into three groups: those with an SII-PNI score of 2 (high SII and high PNI), an SII-PNI score of 1 (either high SII and low PNI or low SII and high PNI), and an SII-PNI score of 0 (low SII and low PNI).

Follow-up information was obtained through clinical consultation records, imaging data, and telephone follow-up visits, which continued until October 31, 2024 or the patient’s death, whichever occurred first. The primary outcome of this study was pCR, defined as surgical resection specimens that showed no tumor cell remnants on pathological examination (i.e., ypT₀N₀cM₀). the secondary outcomes were DFS, defined as the time from surgery to disease recurrence or death from any cause (whichever occurred first), and OS, defined as the time from the start of nICT to death from any cause.

Most patients in this study received two to three cycles of neoadjuvant immunotherapy combined with chemotherapy before surgery. This regimen consisted of an intravenous infusion of an immune checkpoint inhibitor (camrelizumab 200 mg, tislelizumab 200 mg, sintilimab 200 mg, pembrolizumab 2 mg/kg, or nivolumab 3 mg/kg) on ​​day 1 of each cycle, combined with a paclitaxel- and platinum-based chemotherapy regimen (e.g., nab-paclitaxel 125 mg/m² administered intravenously on days 1 and 8 of each cycle, or paclitaxel 175 mg/m² administered intravenously on day 1 of each cycle combined with cisplatin 75 mg/m² administered intravenously on day 1 of each cycle). Curative resection, typically using an Ivor Lewis or McKeown esophagectomy, was performed 4 to 6 weeks after completion of neoadjuvant therapy. Postoperative adjuvant therapy is not mandatory and needs to be formulated according to the postoperative pathological results and the specific conditions of the patient. However, for patients with esophageal squamous cell carcinoma with pathological stage ypT_0-4aN₊M₀, adjuvant therapy is recommended. For patients with ypT_0-4aN₀M₀, adjuvant therapy is recommended based on close monitoring. Postoperative adjuvant therapy options include chemotherapy combined with immunotherapy, immunotherapy, chemotherapy.

Continuous variables were expressed and statistically tested according to their data distribution characteristics. One-way ANOVA was used and expressed as mean ± standard deviation (Mean ± SD) if they were normally distributed and met the requirements to satisfy variance chi-square. If they did not meet the criteria for normal distribution, they were analyzed by the Kruskal-Wallis rank sum test and expressed as median and interquartile range (Median, IQR). Categorical variables were expressed as frequencies and proportions (n, %), and Pearson’s chi-square test or the continuity-corrected chi-square test was used. We used maximization of the Youden index to calculate the optimal thresholds for SII and PNI, ROC curves to assess the predictive performance of the SII-PNI scores. When performing univariate and multivariate logistic regression analyses of independent predictors of pCR, we performed multicollinearity diagnosis and calculated the variance inflation factor (VIF). The VIF of all variables was less than 4, indicating that there was no serious multicollinearity problem. Survival analyses were performed using the Kaplan-Meier method; overall and pairwise comparisons were performed using the log-rank test, and p-value correction was performed using the Bonferroni method when comparing differences in DFS and OS among the three patient groups. Univariate and multivariate Cox were used to estimate the hazard ratios for DFS and OS in the three patient groups. Multicollinearity was also diagnosed in the Cox regression analysis, and the VIFs for all variables were less than 4, confirming model stability. Predictors meeting the p-value threshold (P < 0.1) in univariate analysis were included in the multivariate Cox regression. In subgroup analyses, hazard ratios (HRs) for DFS and OS were calculated using non-stratified univariate Cox, and the results were presented as forest plots. Two-tailed tests were applied for all statistical analyses, with a significance level of P < 0.05. All statistical analyses were performed using SPSS software version 26.0 (IBM, Chicago, IL, USA) and R software version 4.3.0 (R Foundation) for data analysis.

A total of 345 patients with ESCC treated with nICT were included in this retrospective multicenter cohort study, including 70 patients with an SII-PNI score of 0, 149 patients with a score of 1, and 126 patients with a score of 2. The baseline demographic characteristics of patients in the different SII-PNI score groups in the combined cohort are shown in Table 1. The different SII-PNI score groups differed in terms of gender (P = 0.005), BMI (P = 0.034), smoking (P = 0.002), drinking (P = 0.040), cTNM stage (P = 0.002), ypTNM stage (P < 0.001), and TRG (P < 0.001). Specifically, there were more patients with cTNM stage IV (38.9%), ypTNM stage III (43.7%) and IVA (7.9%), and TRG grade 3 (40.5%) tumors in the 2-score group. In contrast, there were fewer patients with cTNM stage IV tumors (22.9%), ypTNM stage III (14.3%) and IVA (2.9%), and TRG grade 3 (17.1%) in the 0-score group, with the 1-score group in between. In addition, we compared hematologic characteristics among the different SII-PNI score groups (Supplementary Table 1).

The predictive efficacy of the SII-PNI score and its components (SII and PNI) for pCR after nICT was assessed using ROC curves in four cohorts (Figure 1). The DeLong’s test confirmed that the SII-PNI score had superior predictive power compared to its individual components. The statistical significance and specific AUC values are shown in Supplementary Figure 1. In the combined cohort (Figure 1A), the AUC of the SII-PNI score was 0.803 (95% CI: 0.753-0.854), and, compared to the SII-PNI score, SII (AUC = 0.679, 95% CI: 0.616-0.742, P < 0.001) and PNI (AUC = 0.667, 95% CI: 0.598-0.737, P < 0.001) had lower predictive efficacy. This trend remained stable in the other cohorts: the FJCH cohort (Figure 1B) (AUC: SII-PNI score = 0.807 [reference], SII = 0.681 [95% CI: 0.600-0.763, P < 0.001], PNI = 0.671 [95% CI: 0.586-0.757, P < 0.001]), the WHUH cohort (Figure 1C) (AUC: SII-PNI score = 0.803 [reference], SII = 0.698 [95% CI: 0.593-0.802, P = 0.012], PNI = 0.693 [95% CI: 0.563-0.823, P = 0.014]), and 900th Hospital cohort (Figure 1D) (AUC: SII-PNI score = 0.787 [reference], SII = 0.655 [95% CI: 0.419-0.891, P = 0.131], PNI = 0.629 [95% CI: 0.341-0.918, P = 0.141]). The trend that the predictive efficacy of SII-PNI score was higher than that of SII and PNI in all cohorts was consistent across centers, which suggests that the SII-PNI score has stable predictive performance and can be used as a reliable clinical tool for predicting pCR after nICT.

Univariate logistic regression analysis of pCR in the combined cohort, SII-PNI score, BMI, cTNM staging, vessel invasion, and perineural invasion were identified as potential predictors of pCR and were subsequently included in multivariate analysis. Multivariate logistic regression analysis revealed that the SII-PNI score, BMI, cTNM stage, and vessel invasion were independent predictors of pCR. Specifically, Compared to the 0-score group, patients in the 1-score group had a lower pCR rate (OR = 0.159, 95% CI:0.080-0.319, P < 0.001), while those in the 2-score group showed an even lower pCR rate (OR = 0.025, 95% CI:0.009-0.073, P < 0.001). Additionally, in cTNM staging, stage III patients had a lower pCR rate compared to stage II patients (OR = 0.397, 95% CI:0.171-0.919, P = 0.031), and stage IVA patients exhibited a further reduction in pCR rate (OR = 0.250, 95% CI:0.094-0.664, P = 0.005). An increase in BMI was significantly associated with a higher pCR rate (OR = 1.145, 95% CI:1.019-1.287, P = 0.023). Patients with positive vessel invasion had a reduced pCR rate (OR = 0.414, 95% CI:0.177-0.968, P = 0.042). Multicollinearity diagnosis confirmed that the VIF of all variables in the multivariate logistic regression was less than 4, indicating that there was no serious multicollinearity, ensuring the stability of the model estimates (Table 2).

The nomogram constructed based on independent predictors identified through multivariate logistic regression demonstrated reliable predictive performance (Figure 2A). Both the predicted probability curve and calibration curve closely aligned with the ideal reference line, indicating that the model-predicted probabilities closely matched the actual observed values, confirming the model’s good accuracy (Figure 2B). Decision curve analysis (DCA) further validated the clinical utility of the integrated model and SII-PNI score: the multivariate logistic model showed higher net benefit than other clinical indicators and the “All” and “None” reference lines within the 10%-80% risk threshold range, while the SII-PNI score outperformed other clinical indicators and the reference lines within the 10%-60% risk threshold range, indicating that the net benefits of the integrated model and SII-PNI score were significantly higher than those of other clinical indicators (Figure 2C).

The median follow-up time was 25.66 months (interquartile range [IQR]: 14.03-39.26 months; IQR = 25.23 months), 9 of 70 patients (12.9%) in the 0-score group, 31 of 149 patients (20.8%) in the 1-score group, and 48 of 126 patients (38.1%) in the 2-score group died. Significant differences in OS (overall log-rank P < 0.001) and DFS (overall log-rank P < 0.001) were found in the combined cohort for each SII-PNI score group. Specifically, compared with the 0-score group, the 1-score and 2-score groups showed higher risks for both DFS and OS. For DFS (Figure 3A), compared with the 0-score group, the 1-score group had an elevated risk (HR = 1.816, 95% CI:1.106-2.981, adjusted P = 0.101), and the 2-score group showed a further increased risk (HR = 2.904, 95% CI:1.850-4.558, adjusted P < 0.001). Compared with the 1-score group, the 2-score group also exhibited a higher risk (HR = 1.622, 95% CI:1.116-2.357, adjusted P = 0.029). For OS (Figure 3B), compared with the 0-score group, the 1-score group demonstrated an elevated risk (HR = 2.040, 95% CI:1.075-3.874, adjusted P = 0.160), and the 2-score group displayed a further increased risk (HR = 4.321, 95% CI:2.564-7.284, adjusted P < 0.001). Compared with the 1-score group, the 2-score group was associated with a similarly elevated risk (HR = 2.157, 95% CI:1.379-3.373, adjusted P < 0.001). The median DFS and OS in the 2-score group were 20.86 months and 38.17 months, respectively. Due to insufficient follow-up, the median DFS and OS could not be determined for the 0-score and 1-score groups. We also constructed Kaplan-Meier curves for DFS and OS in the FJCH cohort, WHUH cohort, and 900th Hospital cohort, respectively, which yielded results consistent with those in the integrated cohort (Supplementary Figure 2).

In univariate Cox proportional hazards regression analysis of OS in the combined cohort, the SII-PNI score, ECOG PS score, cTNM stage, vessel invasion, and perineural invasion were identified as potential predictors and included in the multivariate analysis. In the multivariate Cox proportional hazards regression analysis, independent risk factors for OS included a poorer SII-PNI score (1-score group: HR = 2.169, 95% CI:1.006-4.676, P = 0.048; 2-score group: HR = 4.473, 95% CI:2.138-9.357, P < 0.001) and ECOG PS score (HR = 1.699, 95% CI:1.109-2.602, P = 0.015), both of which were significantly associated with reduced OS (Table 3). Similarly, in univariate Cox proportional hazards regression analysis of DFS in the combined cohort, the SII-PNI score, cTNM stage, vascular invasion, and perineural invasion were identified as potential predictors and included in the multivariate analysis. 2-score group (HR = 2.487, 95% CI:1.414-4.374, P = 0.002), cTNM stage IV (HR = 2.010, 95% CI:1.005-4.021, P = 0.048), and perineural invasion (HR = 1.607, 95% CI:1.052-2.455, P = 0.028) emerged as independent risk factors for reduced DFS (Supplementary Table 2). For both multivariate Cox proportional hazards regression analysis (OS and DFS), the VIF for all included variables were below 4, confirming the absence of severe multicollinearity and the stability of the estimates.

To clarify the association between SII-PNI score and OS/DFS risk across baseline characteristics (including gender, ECOG PS score, smoking history, alcohol history, clinical TNM stage, hypertension, vascular invasion, perineural invasion, and tumor location), We performed subgroup analyses in the combined cohort. Results demonstrated that, in most subgroups, patients with poorer SII-PNI scores (1-score group and 2-score group) showed an increasing trend in mortality risk for OS (Figure 4) and disease progression risk for DFS (Supplementary Figure 3) compared to 0-score group, with risk exhibiting a graded dependence on score severity (higher risk in 2-score group than in 1-score group).