This study retrospectively and consecutively screened patients with LA-ESCC who received nICT at Sun Yat-sen University Cancer Center between 2020 and 2024. Patients were selected according to the predefined inclusion and exclusion criteria. The inclusion criteria were as follows: (1) histologically confirmed ESCC; (2) clinical Tumor-Node-Metastasis (cTNM) stage II-IVA according to the 8th edition of the American Joint Committee on Cancer (AJCC) staging system^[38]; and (3) administration of at least one cycle of PD-1 inhibitor combined with chemotherapy. Exclusion criteria included: (1) administration of other neoadjuvant treatment regimens such as radiotherapy or targeted therapy before surgery; (2) presence of other primary malignancies; (3) failure to undergo surgery after nICT due to disease progression during treatment, severe treatment-related toxicity, poor physical condition unsuitable for surgery, refusal of surgery, or other reasons; (4) non-R0 resection; (5) active infection, hematologic disorder, or autoimmune disease; and (6) incomplete baseline or follow-up data. All patients who met the inclusion criteria and did not meet any of the exclusion criteria were included in the final analysis. A complete-case analysis was performed. Missing or out-of-window data were minimal (2.99%) and mainly involved biochemical and hematological variables. Detailed information on missing or out-of-window data for each variable is provided in Supplementary Table 1. After screening, a total of 204 eligible patients with LA-ESCC were included in the final analysis. The study was performed in accordance with the Declaration of Helsinki and was approved by the ethics committee of Sun Yat-sen University Cancer Center (Approval No. B2025-776-01). The requirement for informed consent was waived by the institutional review board given the retrospective design and complete anonymization of all patient data. A flow diagram of the patient selection process is presented in Figure 1.

All patients received concurrent chemotherapy and immunotherapy before surgery, administered every three weeks. The chemotherapy regimens included either (1) a taxane-platinum combination, and (2) a taxane-fluoropyrimidine combination. The main chemotherapeutic agents used were paclitaxel, nab-paclitaxel, docetaxel, cisplatin, carboplatin, nedaplatin, lobaplatin, tegafur, and capecitabine. For immunotherapy, patients received one of the following PD-1 inhibitors: camrelizumab (AiRuiKa®, Hengrui Pharmaceuticals, Jiangsu, China), pembrolizumab (Keytruda®, Merck & Co., USA), tislelizumab (BaiZeAn®, BeiGene, Beijing, China), toripalimab (Tuoyi®, Junshi Biosciences, Shanghai, China), sintilimab (Tyvyt®, Innovent Biologics, Suzhou, China), or serplulimab (Hansizhuang®, Henlius Biotech, Shanghai, China).

After completion of neoadjuvant treatment, all patients underwent radical esophagectomy, most commonly minimally invasive esophagectomy (MIE) using the McKeown or Ivor Lewis approach combined with two-field or three-field lymphadenectomy. Although adjuvant immunotherapy following neoadjuvant chemoradiotherapy has been reported to improve survival outcomes^[39], there is currently no established consensus on adjuvant therapy after nICT. Therefore, postoperative adjuvant immunotherapy was allowed but not mandatory in this cohort.

Peripheral blood samples were collected within two weeks before the initiation of nICT and within one week before surgery. To minimize the influence of non-tumor-related inflammatory conditions, patients with clinically evident fever, active infection, hematologic disorder, autoimmune disease, or other conditions that might significantly affect peripheral blood inflammatory indicators at the time of blood sampling were excluded. In addition, available medical records were reviewed for potential confounding factors before blood sampling, including recent antibiotic use, corticosteroid use, granulocyte colony-stimulating factor support, blood transfusion, immune-related adverse events, and severe treatment-related toxicities. Owing to the retrospective nature of this study, these factors could not be systematically adjusted for in all patients. Laboratory tests including white blood cell count (WBC), neutrophil count (NEUT), lymphocyte count (LYM), monocyte count (MONO), platelet count (PLT), serum C-reactive protein (CRP) level, and serum albumin (ALB) level, were measured, and inflammation-based indices were subsequently calculated as follows: NLR = NEUT/LYM, PLR = PLT/LYM, LMR = LYM/MONO, and CAR = CRP/ALB. In addition, clinicopathological data were also collected, including age, sex, smoking history, body mass index (BMI), Eastern Cooperative Oncology Group performance status (ECOG-PS), tumor location, histologic differentiation, treatment regimen, cycles of neoadjuvant treatment, interval between the last treatment cycle and surgery, T stage, N stage, cTNM stage, pathologic findings, recurrence status, and survival outcomes.

Surgical specimens obtained after nICT were processed and evaluated according to the institutional standardized pathological protocol. The primary tumor bed and all dissected regional lymph nodes were systematically examined by experienced gastrointestinal pathologists. Cases with pCR or uncertain pathological response were reviewed and confirmed by a second pathologist. pCR was defined as the absence of viable tumor cells in both the resected primary tumor and all dissected regional lymph nodes, corresponding to ypT0N0. Tumor regression grade (TRG) was evaluated according to the modified Ryan scheme recommended by the College of American Pathologists (CAP) and the National Comprehensive Cancer Network (NCCN). TRG 0 was defined as complete response with no viable cancer cells; TRG 1 as near-complete response with single cells or rare small groups of cancer cells; TRG 2 as partial response with residual cancer showing evident tumor regression; and TRG 3 as poor or no response with extensive residual cancer and no evident tumor regression. Representative hematoxylin and eosin-stained sections of primary tumor tissue, adjacent non-tumor esophageal mucosa, and regional lymph node metastasis were reviewed and are presented in Supplementary Figure 1.

The primary endpoint of this study was pCR, as defined above. The secondary endpoints were DFS and OS. DFS was defined as the interval from the date of radical surgery to the first documented recurrence, disease progression, death from any cause, or last follow-up. OS was defined as the interval from the date of diagnosis to death from any cause or the last follow-up. DFS and OS were calculated from different starting points because DFS reflects postoperative disease control, whereas OS reflects the overall disease course from diagnosis. Survival analyses were restricted to patients who completed nICT and underwent curative surgery. After surgery, all patients were regularly followed up through outpatient visits and telephone interviews. Follow-up data were obtained from the hospital’s electronic medical record system and institutional follow-up database, with the last follow-up date on August 21, 2025.

Categorical variables were expressed as numbers and percentages and analyzed using the Chi-square test or Fisher’s exact test. The Shapiro-Wilk test was used to assess the normality of continuous variables. Variables with a normal distribution were expressed as mean ± standard deviation (SD) and compared using the independent-samples t-test, whereas variables with a non-normal distribution were presented as median and interquartile range (IQR) and compared using the Mann-Whitney U test.

ROC curve analysis was performed to evaluate the predictive performance of hematologic indicators for pCR. The optimal cut-off values were determined using the Youden index. The diagnostic performance of these cut-off values was further evaluated using sensitivity, specificity, positive predictive value, negative predictive value, and corresponding 95% confidence intervals.

Univariable and multivariable logistic regression analyses were conducted to identify independent predictors of pCR, with results presented as odds ratios (ORs) and corresponding 95% confidence intervals (CIs). Multicollinearity among candidate variables for multivariable logistic regression was assessed before model construction. The generalized variance inflation factor (GVIF) was calculated, and for categorical variables with more than one degree of freedom, GVIF^{1/(2 × Df)} was reported to allow comparison across variables. To evaluate the potential influence of treatment-related heterogeneity, additional treatment-adjusted models were constructed to assess whether the associations between key hematologic indicators and pCR remained robust after adjustment for treatment-related factors. The events per variable (EPV) was calculated to evaluate the risk of overfitting.

A nomogram model was developed based on the independent predictors identified in the multivariable logistic regression analysis. Internal validation was performed using 1,000 bootstrap resamples. The discriminative ability of the nomogram was evaluated using the area under the ROC curve (AUC) and the concordance index (C-index), while calibration curves were generated to assess the agreement between predicted and observed probabilities. Decision curve analysis (DCA) was further applied to evaluate the clinical utility and net benefit of the nomogram.

OS and DFS were estimated using the Kaplan-Meier method, and differences between groups were compared using the log-rank test. Univariable and multivariable Cox proportional hazards regression analyses were performed to identify independent prognostic factors, with results expressed as hazard ratios (HRs) and 95%CIs. The proportional hazards assumption for the final Cox models was assessed using Schoenfeld residuals.

All statistical analyses were two-sided, and a P-value < 0.05 was considered statistically significant. Analyses were conducted using R software (version 4.5.1, R Foundation for Statistical Computing, Vienna, Austria). The main R packages used included readxl and dplyr for data processing, pROC for ROC analysis, car for multicollinearity assessment, rms for nomogram construction and calibration, rmda for decision curve analysis, ResourceSelection for the Hosmer-Lemeshow goodness-of-fit test, ggplot2 for visualization, and survival and survminer for survival analysis.

The baseline clinicopathological characteristics of the patients are summarized in Table 1. A total of 204 patients with LA-ESCC were included in this study. The median age was 62.5 years (range, 42-75 years), and 169 (82.8%) were male, while 35 (17.2%) were female. A history of smoking was observed in 61.3% of patients. More than half of the tumors were located in the middle thoracic esophagus (60.3%). With respect to tumor differentiation, most cases were moderately differentiated (59.8%) or poorly differentiated (37.3%). According to the 8th edition of the AJCC TNM staging system, 14.2% of patients were classified as stage II, 56.4% as stage III, and 29.4% as stage IVA. Further analysis revealed that tumor differentiation, T stage, and cTNM stage were significantly associated with pCR (P < 0.05), whereas no significant associations were observed between pCR and other clinicopathological characteristics.

Detailed pathological outcomes after nICT are summarized in Supplementary Table 2. The median number of lymph nodes examined per patient was 40 (IQR, 30-50.25). Overall, 59 patients (28.9%) achieved pCR. Among the remaining patients, residual disease was confined to the primary tumor in 81 (39.7%), to regional lymph nodes in 9 (4.4%), and to both sites in 55 (27.0%).

Figure 2 illustrates the dynamic changes in hematologic parameters before and after nICT. The results showed that preoperative levels of NEUT, LYM, MONO, PLT, CRP, NLR, PLR, and CAR were significantly decreased compared with baseline values. No significant differences were observed in ALB and LMR between the pre- and post-treatment measurements. Furthermore, Table 2 summarizes the differences between the pCR and non-pCR groups. At baseline, the non-pCR group demonstrated higher NEUT (P = 0.002),, MONO (P = P = 0.035), NLR (P < 0.001), and PLR (P = 0.013) compared with the pCR group, whereas LYM (P = 0.027) and LMR were lower (P < 0.001). Prior to surgery, higher NEUT (P < 0.001) and NLR (P < 0.001) levels persisted in the non-pCR group. No other hematologic parameters differed significantly between the two groups.

ROC curves were generated and the AUC values were calculated to evaluate the predictive performance of baseline and preoperative hematologic indices for pCR [Figure 3]. For clinical interpretability, risk stratification, and visualization, the optimal cut-off values were determined using the Youden index. The diagnostic performance of the Youden index-derived cut-off values, including sensitivity, specificity, positive predictive value, negative predictive value, and corresponding 95%CIs, is summarized in Supplementary Table 3. The AUCs for baseline markers were 0.690 (95%CI: 0.607-0.772, P < 0.001) for NLR, 0.652 (95%CI: 0.570-0.734, P < 0.001) for LMR, 0.611 (95%CI: 0.525-0.698, P = 0.011) for PLR, and 0.528 (95%CI: 0.444-0.611, P = 0.515) for CAR. For preoperative markers, the AUCs were 0.668 (95%CI: 0.589-0.748, P < 0.001) for NLR, 0.586 (95%CI: 0.497-0.676, P = 0.059) for LMR, 0.471 (95%CI: 0.381-0.560, P = 0.518) for PLR, and 0.579 (95%CI: 0.494-0.664, P = 0.070) for CAR. The optimal cut-off values determined using the Youden index were as follows: baseline NLR = 2.080, LMR = 2.849, PLR = 104.982, and CAR = 0.128; preoperative NLR = 2.498, LMR = 4.698, PLR = 143.080, and CAR = 0.073. Patients were subsequently stratified into high and low groups according to these optimal cut-off values for subsequent analyses.

Clinicopathological and hematologic parameters associated with pCR were analyzed [Table 3]. In the univariate logistic regression analysis, sex (P = 0.049), tumor differentiation (P = 0.005), cTNM stage (P = 0.004), baseline NLR (P < 0.001), baseline LMR (P = 0.002), baseline PLR (P < 0.001), baseline CAR (P = 0.026), preoperative NLR (P < 0.001), preoperative LMR (P = 0.001), and preoperative CAR (P = 0.012) were significantly associated with pCR. Multicollinearity assessment showed no substantial collinearity among the candidate variables, with adjusted GVIF values ranging from 1.019 to 1.111 [Supplementary Table 4]. Multivariate logistic regression further demonstrated that tumor differentiation (P = 0.006), cTNM stage (P = 0.024), baseline NLR (P = 0.014), preoperative NLR (P = 0.006), and preoperative CAR (P = 0.037) were independently associated with pCR.

Subsequently, a nomogram was developed based on the independent predictors of pCR identified in the multivariate logistic regression analysis [Figure 4A]. Among these variables, baseline NLR, preoperative NLR, and cTNM stage contributed most to the prediction of pCR [Figure 4B].

To further evaluate the performance of the nomogram, ROC curve, calibration plot, and DCA were performed [Figure 5]. The model exhibited good discriminative ability with a C-index of 0.816 (95%CI: 0.755-0.877), and the bootstrap-corrected C-index after 1000 resamples was 0.792. The calibration plot based on 1000 bootstrap resamples showed excellent agreement between predicted and observed pCR probabilities (Brier score = 0.154; Hosmer-Lemeshow test, P = 0.907), confirming the good calibration and accuracy of the model. Furthermore, the DCA demonstrated that patients with LA-ESCC receiving nICT could gain a substantial net clinical benefit from the nomogram, underscoring its potential clinical utility. Compared with the clinical model including only tumor differentiation and cTNM stage, the full nomogram model showed improved discrimination (AUC: 0.816 vs. 0.683; bootstrap-corrected AUC: 0.792 vs. 0.667) and a lower Brier score (0.154 vs. 0.187). These results suggest that baseline NLR, preoperative NLR, and preoperative CAR provided incremental predictive value beyond conventional clinicopathological factors [Supplementary Table 5].

To address the potential redundancy between baseline and preoperative hematologic indicators, additional separated models were constructed. In the baseline model adjusted for tumor differentiation and cTNM stage, low baseline NLR was significantly associated with a higher probability of pCR (OR = 4.689, 95%CI: 2.246-10.033, P < 0.001). In the preoperative model, both low preoperative NLR (OR = 3.716, 95%CI: 1.815-8.039, P < 0.001) and low preoperative CAR (OR = 2.666, 95%CI: 1.197-6.409, P = 0.021) were significantly associated with pCR. These findings support the rationale that baseline and preoperative hematologic indicators may reflect distinct inflammatory states before and after nICT and provide complementary information for pCR prediction [Supplementary Table 6].

Sensitivity analyses were conducted to evaluate the potential impact of treatment heterogeneity on the pCR prediction model. After additional adjustment for PD-1 inhibitor type, chemotherapy regimen, number of nICT cycles, and interval from nICT to surgery, the associations of baseline NLR, preoperative NLR, and preoperative CAR with pCR remained directionally consistent and statistically significant across all models. These results indicated that the main findings were relatively robust after accounting for major treatment-related factors [Supplementary Table 7].

Kaplan-Meier survival analysis and log-rank tests were conducted to further evaluate the associations between hematologic markers and patient outcomes. The median follow-up time was 37.3 months (range, 6.8-71.3 months). During the follow-up, 54 patients (26.5%) experienced recurrence and 47 patients (23.0%) died. The median DFS and OS were not reached. Patients with low baseline NLR had significantly better DFS than those with high NLR (3-year DFS rate: 84.9% vs. 54.5%, P = 0.001) [Figure 6A]. Similarly, patients with low preoperative CAR achieved improved DFS (3-year DFS rate: 63.8% vs. 54.5%, P = 0.040) [Figure 6B]. Regarding OS, patients with low baseline NLR also demonstrated superior survival (3-year OS rate: 91.6% vs. 71.1%, P = 0.006) [Figure 6C]. Moreover, patients who achieved pCR exhibited significantly prolonged DFS and OS compared with those without pCR (3-year DFS rate: 88.3% vs. 50.4%, P < 0.001; 3-year OS rate: 88.3% vs. 70.0%, P = 0.005) [Figure 6D and E]. No statistically significant differences in DFS [Supplementary Figure 2] or OS [Supplementary Figure 3] were observed for other inflammatory markers.

In the univariate Cox proportional hazards analysis, female (P = 0.012), better tumor differentiation (P = 0.035), pCR (P < 0.001), low baseline NLR (P = 0.002), and low preoperative CAR (P = 0.042) were significantly associated with longer DFS. Multivariate Cox analysis further identified female (HR = 0.404, 95%CI: 0.174-0.934, P = 0.034), pCR (HR = 0.245, 95%CI: 0.104-0.580, P = 0.001), and low baseline NLR (HR = 0.448, 95%CI: 0.202-0.993, P = 0.047) as factors independently associated with longer DFS [Table 4]. Schoenfeld residual testing showed no significant violation of the proportional hazards assumption for the final DFS model, with a global P-value of 0.328.

When OS was analyzed as the endpoint, pCR (P = 0.009) and low baseline NLR (P = 0.011) were significantly associated with longer OS in univariate analysis. In multivariate analysis, pCR (HR = 0.413, 95%CI: 0.173-0.986, P = 0.046) and low baseline NLR (HR = 0.281, 95%CI: 0.086-0.920, P = 0.036) remained independently associated with longer OS [Table 5]. Schoenfeld residual testing also showed no significant violation of the proportional hazards assumption for the final OS model, with a global P-value of 0.392.