This is a single-center retrospective cohort study. Patients with TESCC who received nICT or nCT alone then followed by surgery in Fujian Medical University Cancer Hospital from January 2017 to December 2023 were screened and analyzed, as shown in the flowchart (Figure 1). The inclusion criteria were: (1) tumor locating in the thoracic esophagus; (2) histologically confirmed as ESCC; (3) underwent preoperative nICT or nCT; (4) underwent esophagectomy and lymph node dissection; and (5) absence of non-regional lymph node metastases and/or distant organ metastases, except for supraclavicular lymph node metastases. The exclusion criteria were: (1) cervical or esophagogastric junction ESCC; (2) esophageal adenocarcinoma, small cell carcinoma, or other histology; (3) received preoperative nCRT or other antitumor therapies; (4) simultaneously combined with other malignant tumors; and (5) with incomplete medical records. This study was approved by the Human Ethics Review Committee (Approval No.: K2025-340-01) of Fujian Medical University Cancer Hospital. The primary endpoint was LNM rate. Secondary endpoints included postoperative recurrence patterns, OS and EFS. OS was measured from the initiation of neoadjuvant therapy until death from any cause, whereas EFS was measured from the same start point to the first event among local recurrence, metastasis, or death from any cause.

Data were collected from the hospital’s medical record, including: gender, age, body mass index (BMI), personal medical history, esophageal tumor location, neoadjuvant treatment regimen, surgical data, pathological data, TNM stage, auxiliary examination data (complete blood count, biochemistry, endoscopy, CT, PET-CT, etc.), postoperative treatment, and follow-up data. Pretreatment clinical staging (cTNM) was determined using chest CT, PET-CT, and endoscopic ultrasound. TNM staging was based on the 8^th AJCC edition (27). Follow-up was conducted through outpatient reviews, inpatient records, and telephone inquiry.

All patients received at least one cycle of neoadjuvant therapy administered every 3 weeks (q3w) before surgery. The nCT consisted of taxane and platinum doublet agents. Taxanes included paclitaxel (150 mg/m² on day 1), nab-paclitaxel (260 mg/m² on day 1), or docetaxel (75 mg/m² on day 1). Platinum agents included cisplatin (75 mg/m² on day 1), carboplatin (AUC 5–6 on day 1), lobaplatin (30 mg/m² on day 1), or nedaplatin (80 mg/m² on day 1). For nICT, a PD-1 inhibitor was added to the taxane and platinum doublet agents. PD-1 inhibitors included pembrolizumab, camrelizumab, sintilimab and tislelizumab (each 200 mg intravenously on day 1, q3w), as well as toripalimab (240 mg intravenously on day 1, q3w) and serplulimab (300 mg intravenously on day 1, q3w). The specific agents were selected at the discretion of the treating physicians based on individual patient characteristics and clinical considerations.

All patients proceeded to esophagectomy with lymph node dissection 3–8 weeks after completion of the last neoadjuvant cycle. Surgical approach was categorized as open esophagectomy or minimally invasive esophagectomy (MIE). The operative procedure was further classified as McKeown (cervical anastomosis) or Ivor-Lewis (intrathoracic anastomosis) esophagectomy. The extent of lymphadenectomy was defined as two-field (thoracic and abdominal) or three-field (cervical, thoracic and abdominal) dissection. Cervical lymphadenectomy, when performed, included at minimum the cervical paraesophageal and supraclavicular nodes as defined by the Japan Esophageal Society (JES) nodal map (28). Lymph node stations were recorded according to the Japanese Classification of Esophageal Cancer (JES, 12^th edition) for cervical and thoracic stations (e.g., No.101-112), and abdominal nodal stations were documented following the Japanese gastric nodal station system adopted by the JES classification. The definitions of lymphadenectomy fields (two-field vs three-field) and the detailed lymph node station map are provided in Figure 2A, Supplementary Table 1. Resected tumor and lymph node specimens were independently evaluated by two pathologists (W. Xiao and Y. Shi).

Two senior pathologists evaluated the surgical specimens and issued the pathologic reports. Assessment included ypTNM stage, histological type, differentiation, circumferential resection margin (CRM), tumor regression, pathological response, immunohistochemistry and number of lymph nodes. Based on tumor regression, pathological complete response (pCR) was defined as the absence of viable tumor cells in both the primary site and resected lymph nodes (including carcinoma in situ at the primary site), and major pathological response (MPR) was defined as < 10% viable tumor cells in the primary tumor bed (29). R0 was defined as negative longitudinal resection margin and CRM, and R0 (close margin) was defined as negative longitudinal resection margin and tumor 0-1mm from CRM (30).

Lymph node analysis and pattern summarization were conducted using the JES (28). To more comprehensively assess the true degree of lymph node involvement, we used the following metrics: ➊Lymph Node Ratio (LNR): Also known as the positive lymph node ratio, calculated as the total number of metastatic lymph nodes divided by the total number of resected lymph nodes (31); ➋Lymph Node Metastasis Rate (LNM_Rate): The proportion of patients with lymph node metastasis among the total patient population; ➌Actual Lymph Node Metastasis Rate (actual LNM_Rate): The proportion of patients with lymph node metastasis in a specific station among those who actually underwent dissection of that station; ➍For detailed analysis of specific lymph node stations, the following terms were defined: I. LNMR1_Level X = LNR for a specific level = Number of metastatic LNs in level X/Total number of LNs dissected from each patient; II. LNMR2_Level X = Lymph Node metastasis rate for level X = Number of patients with metastasis in level X/Total number of patients; III. LNMR3_Level X = Lymph Node actual metastasis rate for level X = Number of patients with metastasis in level X/Number of patients who actually had level X dissected. Based on previous studies (32), LNM rates were categorized into three levels: < 5%, 5-10%, and > 10%. Maps of metastasis rates for different stations were generated using yellow, orange, and red to represent these levels, respectively.

Postoperative tumor recurrence was categorized as locoregional recurrence (LR) or distant recurrence (DR). LR was defined as recurrence at the esophagogastric anastomotic site or in regional lymph nodes (according to JES stations). DR was defined as tumor occurrence in other organs or non-regional lymph nodes. Recurrence or metastasis was determined based on pathological reports or comprehensive review of imaging findings. Specific positive imaging criteria included (1): higher avidity/uptake (SUVmax > 3.0) in PET-CT (2); irregular nodularity or irregular thickening > 5 mm with abnormal enhancement at the site of anastomosis in CT scanning (3); bulky, clustered or fused lymph nodes, or a short-axis diameter greater than > 10 mm in contrasted CT scanning (4); brain, bone, liver or pelvic metastases verified in combination with magnetic resonance imaging (MRI) (33, 34). To quantify the recurrence/metastasis burden, we established a scoring system: 1 point for a single LR site; 2 points for a single DR site; a total score of 8 points if the number of recurrence/metastasis sites was ≥ 4.

To reduce the influence of potential confounders in the two groups, we used propensity-score matching (PSM). We matched the two groups according to the 1:1 optimal matching principle with the use of the MatchIt package of R software. The characteristics of the two groups of patients were included for propensity score matching method, including age, gender, smoking, drinking, tumor location, clinical TNM stage, number of neoadjuvant therapy cycles, surgical approach, type of surgery, extent of resection, number of lymph node dissection and postoperative adjuvant therapy. Multiple statistical measures were used, with median (interquartile range) for continuous variables and frequency (percentage) for categorical variables. Categorical variables were analyzed with the use of the chi-square test or Fisher’s exact test, and continuous variables were compared with the use of the Wilcoxon rank-sum test or the Kruskal-Wallis test. Subgroup analyses were performed to assess the association between treatment group and pCR across different strata. Prespecified subgroup analyses were performed by number of neoadjuvant cycles, taxane and platinum doublet agent. Within each subgroup, multivariable logistic regression estimated adjusted ORs (95% CIs) for pCR comparing nICT with nCT, adjusting for sex, age, BMI, tumor location, cT, and cN. Treatment-by-subgroup interactions were tested and summarized in a forest plot. A receiver operating characteristic (ROC) curve was used to determine the optimal LNR cut-off using the Youden Index. Univariate logistic or Cox Regression analyses were first performed, followed by multivariate models including variables with p < 0.1. Kaplan-Meier curves were generated, and survival difference analysis was assessed using the Log-rank test. Statistical significance was defined as p < 0.05 (two-sided). Analyses were conducted using R software (Version 4.2.2).

As shown in Table 1, a total of 441 eligible TESCC patients were included in our study. The overall male-to-female ratio was 3.5:1, with a median age of 61 years old (range: 44–78 y.o.). Amongst them, 179 (40.6%) received nCT, and 262 (59.4%) received nICT. In comparison to nCT, nICT was associated with higher odds of pCR (12/179 [6.7%] vs 58/262 [22.1%], with an adjusted OR of 4.08, 95% CI 2.09-7.97; p < 0.001). The association was most predominant in patients receiving two cycles of nCT (adjusted OR 4.61, 95% CI 2.17-9.77; p < 0.001), with no significant interaction with number of cycles, taxane or platinum agent (p for interaction = 0.171, 0.108, and 0.093, respectively) (Supplementary Figure 1). In addition, Table 1 still provided a comprehensive comparison of baseline characteristics between nCT group and nICT group before and after PSM. Baseline characteristics were well-balanced within groups (p > 0.05). Detailed distributions of chemotherapy agents and PD-1 inhibitors are provided in Supplementary Table 2.

Postoperative pathologic evaluation confirmed a LNM rate of 46.3% (204/441) for the entire cohort, with no statistical difference between nCT and nICT groups (49.7% vs. 43.9%, p = 0.228). However, regarding LNR, the nICT group showed a significant advantage, with a significantly lower LNR than that of the nCT group before (3.5% vs. 6.2%, p < 0.001) and after (4.2% vs. 6.2%, p < 0.001) matching. ROC curve analysis indicated that LNR had good predictive value for OS, with an AUC of 0.675 (95% CI: 0.629-0.720). The optimal LNR cut-off value was 0.041, as was determined by the maximum Youden index. Based on this cut-off value, LNR was dichotomized into High (≥ 0.041) and Low (< 0.041) subgroups. In the nICT group, the proportion of patients with High LNR was significantly lower than in the nCT group (24.8% vs. 41.9%, p < 0.001). Detailed lymph node dissections and metastasis data are provided in Supplementary Table 3.

Heatmaps illustrating lymph node level-specific metastasis rates were shown in Figures 2B, C, 3–6. There were clear station-specific differences in lymph node metastasis patterns between groups (Figures 3-6). Further analysis of lymph node metastasis patterns revealed that nICT group had significantly lower LNMR1 than that in nCT group regarding several key lymph node stations, including level 104L, 106recR, 107, 110, 1 and 2 (all p < 0.05), as illustrated in Figure 3. Importantly, level 7 (lymph nodes along the left gastric artery) was a common metastatic hotspot in the nCT group. Both the lymph node metastasis rate (LNMR2: 7.6% vs. 14.5%, p = 0.020) and the actual metastasis rate (LNMR3: 8.4% vs. 16.8%, p = 0.012) for level 7 were significantly lower in the nICT group, suggesting that nICT would specifically influence the lymphatic metastasis pathway of ESCC. Furthermore, the number of stations with LNMR2 and LNMR3 greater than 10% was higher in the nCT group than that in the nICT group. In nCT and nICT groups, the highest LNMR2 values were for level 107 (12.8%) and 106recR (12.6%), respectively, and the highest LNMR3 values were for level 104L (24.3%) and 104R (21.8%), respectively. Metastasis rates for different lymph node levels in each group were showed in Figure 4. After matching, nICT group had a lower median LNMR1 at level 2 (0.036 vs. 0.053, p = 0.027), and there was no significant difference in LNMR2 and LNMR3 between the two groups. Complete lymph node metastasis data were provided in Supplementary Tables 4-6.

Until September 1, 2025, 7 patients (1.6%) were lost to follow-up (3 in nCT group, 4 in nICT group). The median follow-up time was 5.97 years in nCT group and 3.16 years in nICT group. Survival analyses indicated that nICT would significantly improve EFS (HR = 0.65, 95% CI: 0.50-0.84, p = 0.001). Although nICT demonstrated a trend towards better OS, the difference was not statistically significant (HR = 0.90, 95% CI: 0.69-1.17, p = 0.436). Regarding the recurrence patterns, the overall recurrence/metastasis rate was significantly lower in the nICT group than that in the nCT group (36.2% vs. 56.8%, p < 0.001), and the median recurrence/metastasis score was lower (p < 0.001) too. Mediastinal LNM and anastomotic site recurrence were the main recurrence patterns. nICT significantly reduced the risk of anastomotic site recurrence (9.6% vs. 18.2%, p = 0.009). Supraclavicular lymph nodes were the most common site of non-regional LNM, and lung, bone, and liver were the most common organs of distant metastases, as shown in Table 2.

As shown in Figure 7, multivariate logistic regression analysis showed that patients receiving nICT had a significantly lower risk of recurrence/metastasis compared to those receiving nCT alone (adjusted OR = 0.55, 95% CI: 0.35-0.88, p = 0.013). MIE significantly reduced the risk of recurrence or metastasis compared to open or converted open surgery (adjusted OR = 0.47, 95% CI: 0.28-0.79, p = 0.005). R0 (close margin) & R1 resection showed a trend toward higher risk of recurrence/metastasis (adjusted OR = 1.9, 95% CI: 0.99-3.64, p = 0.054), and achieving MPR showed a trend towards reduction of risk of recurrence/metastasis (adjusted OR = 0.47, 95% CI: 0.21-1.02, p = 0.057).

As indicated in Figure 8, multivariate Cox regression analyses revealed several factors independently associated with OS. Firstly, non-R0 resection (including close margins and R1 resection), higher ypT stage (ypT3-4), and presence of postoperative recurrence/metastasis were factors significantly associated with higher risk of cancer-related death, especially postoperative recurrence/metastasis (HR = 8.07, 95%CI: 5.75-11.33, p < 0.001). However, receiving postoperative adjuvant therapy was an independent beneficiary factor (HR = 0.59, 95% CI: 0.43-0.80, p < 0.001) favoring longer survival. Kaplan-Meier survival curves for OS were shown in Figure 9.

As shown in Figure 10, multivariate Cox regression analysis showed that EFS was significantly associated with several factors. MIE significantly reduced the risk of events (adjusted HR = 0.66; 95% CI: 0.49-0.88; p = 0.005). R0 resection (close margin) & R1 resection (adjusted HR = 1.57; 95% CI: 1.10-2.26; p = 0.014) and high LNR (adjusted HR = 1.74; 95% CI: 1.10-2.77; p = 0.019) significantly increased the risk of events. For each scale increment in the number of metastatic LNs, the risk of events increased correspondingly (adjusted HR = 1.06; 95% CI: 1.01-1.12; p = 0.013). Corresponding Kaplan-Meier survival curves for EFS were shown in Figure 11.