This retrospective cohort research comprised ESCC cases receiving NICT during 2019-2021. Peripheral blood biomarkers and demographic-clinical datasets were systematically collected. Exclusion parameters encompassed: (i) non-squamous pathological subtypes; (ii) concurrent antineoplastic interventions; (iii) incomplete tumor resection post-NICT; (iv) previous or concurrent with other malignancies; (v) coexisting with other hematologic, autoimmune, or inflammatory diseases; and (vi) inadequate clinical records or follow-up durations. Supplementary Figure S1 delineates the participant selection workflow. Tumor staging in the current research was determined based on the AJCC/UICC TNM staging criteria (8th edition) (21). The institutional review board of Zhejiang Cancer Hospital authorized the research protocol (IRB2020183) in compliance with Helsinki Declaration ethical guidelines.

NICT, in the current study, was administered in two cycles at 21-day intervals as part of the preoperative management. The ICIs, such as camrelizumab (200 mg), sintilimab (200 mg), or tislelizumab (200 mg), were included in the regimens on day 1. Chemotherapeutic protocols included albumin-bound paclitaxel (120 mg/m²) on days 1 and 8, combined with carboplatin (AUC=5 mg/mL/min) administered on day 1. Surgical intervention - employing either the Ivor Lewis or McKeown approach - was scheduled for execution between 4–6 weeks post-NICT completion. Current clinical guidelines lack consensus on postoperative adjuvant regimens, though the CheckMate 577 trial suggests adjuvant immunotherapy post-NCRT demonstrates therapeutic advantages (22). Accordingly, selective adjuvant treatment, but not mandatory, was administered post-resection, prioritizing cases with histopathological confirmation of ypN1–3 or ypT3/T4a. Longitudinal monitoring continues through December 2024.

The Kolmogorov-Smirnov test for continuous data in this investigation revealed that all of the data were non-normally distributed. Consequently, the median and interquartile ranges (IQR) were used to present all continuous data, whereas frequency and percentage were also used to report categorical variables. Utilizing the restricted cubic spline (RCS), the ideal threshold for CCR was established based on the examination of the non-linear relationship between disease-free survival (DFS)/overall survival (OS) and CCR. Using measures of variance inflation factor (VIF) and tolerance, the multicollinearity between different independent hematological variables was examined (23). Severe collinearity of the variables was indicated by a tolerance <0.1 or a VIF > 5 (24). The clinical applicability and prognostic evaluation of CCR and other hematological indices were compared and evaluated by using decision curve analyses (DCAs), receiver operator characteristic curves (ROCs), and time-dependent areas under the curves (AUCs). Log-rank tests and Kaplan-Meier analyses were also utilized to compare the DFS/OS between the low and high groups. DFS and OS were assessed using hazards models based on Cox proportional analyses, which were represented as hazard ratios (HRs) with 95% CIs. Firstly, factors with statistical differences were screened out based on univariate regression analysis. Then, a multivariate analysis with stepwise regression was conducted on the above-mentioned statistically significant factors. An established nomogram model was based on the findings of multifactor Cox proportional risk regression analysis. Internal verification was conducted using the bootstrap method, which involves sampling 1000 times. The model’s prediction performance was represented by the concordance index (C-index), and its prediction conformance was visually represented using the calibration approach. DCA assessed the model’s applicability value. The application value of the model was evaluated by ROCs and DCAs. Using R 4.1.2 and SPSS 20.0, all analyses were conducted with a significance criterion of P < 0.05.

According to inclusion and exclusion criteria, 202 cases were finally eligible to be included in this cohort. There were 20 (9.9%) females and 182 (90.1%) males, with a mean age of 64 years (range: 45–75 years). After surgery, there were 118 (58.4%) patients in the ypT0–2 stage and 84 (41.6%) patients in the ypT3-4a stage. Additionally, 83 patients (41.1%) had positive lymph nodes. Finally, pCR was seen in sixty-one (30.2%) individuals after NICT. After a median follow-up period of 40 months (range: 7–54 months), 77 (38.1%) cases experienced relapses and 54 (26.7%) instances resulted in death.

In the current study, CCR’s prognostic usefulness was assessed by contrasting it with various traditional hematological parameters including NLR, PLR and PNI. To address the collinearity, VIF and tolerance were included. There was no obvious collinearity between the variables, according to the analysis (VIFs between 1.014 and 1.264 and tolerances between 0.791 and 0.986) ( Supplementary Table S1 ). The correlation diagram for all hematological indices is depicted in Figures 1A, B . When compared to other hematological indices, the ROCs showed that CCR had the biggest AUC (OS=0.665, DFS=0.688), suggesting a superior capacity for prediction ( Figures 1C, D ). The DCAs also approved of CCR’s greater clinical applicability in both OS and DFS ( Figures 1E, F ). As for the predictive value in the time-dependent AUCs, additionally, CCR again fared better than other traditional indices ( Figures 1G, H ). Therefore, CCR exhibited the strongest clinical application and predictive power among all these indicators.

The link between survival (DFS/OS) and CCR is depicted in Figures 1I, J , indicating a non-linear relationship. The optimal CCR cutoff value was determined by the RCS model to be 97.5 ( Figures 1K, L ). A threshold of 97.5 (as the ideal CCR cutoff point) was then used to divide the patients into two groups. There were no statistically significant variations in the other clinical variables between these two cohorts, with the exception of the statistically significant differences in sex (P=0.014), tumor length (P=0.014), and ypT stage (P=0.021) ( Table 1 ). Therefore, patients with a low CCR were associated with increased prevalence of ypT disease and tumor length. Regarding postoperative complications, patients with a low CCR were associated with increased prevalence of respiratory complications (P=0.038).

Patients exhibiting low CCR demonstrated significantly worse 3-year DFS (48.5% vs. 74.3%, P<0.001; Figure 2A ) and OS (62.9% vs. 82.9%, P=0.001; Figure 2B ) compared to those with high CCR. To better understand the prognostic significance of CCR, we conducted a subgroup analysis according to the ypT stage, ypN stage, and pCR. The results of the analyses showed that CCR was able to distinguish the OS and DFS for each subgroup analysis ( Figures 2C–N ). In the subgroup analyses of ypT stage, there were statistically significant differences for DFS (ypT0-2: P=0.016; ypT3-4a: P=0.049), while for OS, it was more statistically significant in the ypT3-4a stage (ypT0-2: P=0.422; ypT3-4a: P=0.023) ( Figures 2C–F ). In the ypN stage subgroup analysis, we found that CCR can effectively stratify the prognosis of ESCC patients in both DFS and OS ( Figures 2G–J ). Similarly, in the pCR subgroup analyses, there were statistically significant differences in DFS, while for OS, there were no statistically significant differences among pCR patients ( Figures 2K–N ). Hence, the results indicated that CCR had prognostic significance for the ESCC cases stratified according to subgroup analyses.

In the current study, RCS showed that with the increase of CCR, the DFS and OS of ESCC patients receiving NICT gradually increased. After correcting for confounding factors, there was still a negative linear relationship between DFS/OS and CCR. In univariate analysis, DFS and OS were significantly affected by the following clinical characteristics: vessel invasion, perineural invasion, tumor length, ypT stage, ypN stage, pCR, and CCR ( Figure 3A ). Subsequent multivariate analysis revealed that CCR was found to be an independent predictor of both OS (HR=0.482, 95% CI=0.271-0.857, P=0.013) and DFS (HR=0.440, 95% CI=0.273-0.709, P=0.001) ( Figure 3B ). Compared to the low CCR group, the high CCR group reduced the risk of recurrence by 56.0% and the risk of death by 51.8%, respectively. The Sankey diagram, which was used to analyze the relationship between clinical outcomes and CCR, revealed that the lowest CCR group was more likely to die and experience recurrence ( Figure 3C ). Finally, we conducted a risk-score analysis based on the three indicators with statistical differences in multivariate regression. Through this analysis, prognosis stratification can be better guided in both DFS ( Figure 3D ) and OS ( Figure 3E ).

Using a nomogram based on statistically significant variables in Cox regression analysis, the model’s prediction outcomes can be visually shown ( Supplementary Figures S2A, B ). The model predicted DFS and OS with a C-index of 0.785 (95% CI: 0.761–0.810) and 0.789 (95% CI: 0.762–0.816), respectively, according to the findings of the model’s prediction ability using the C-index. Internal validation of the model was performed using the Bootstrap self-sampling technique (B=1000). The calibration curve demonstrated that the actual observation probability and the anticipated DFS and OS were reasonably consistent ( Supplementary Figures S2C, D ). To assess the model’s accuracy, ROC curves for 1- and 3-year survival DFS/OS were plotted based on independent parameters ( Supplementary Figures S2E, F ). The model’s application value was assessed using DCA curves, and the findings also demonstrated good clinical applicability ( Supplementary Figures S2G–J ).