Our institutional ethics committee approved this study, and an informed consent was signed by each patient before partaking in this study.

We retrospectively collected and analyzed patients with advanced ESCC who received neoadjuvant chemotherapy plus immunotherapy in our hospital from January 2020 to September 2022. The inclusion criteria were as follows: 1) diagnosis of ESCC based on preoperative endoscopic pathological examination, 2) patients underwent CT before neoadjuvant therapy, and 3) patients’ clinical TNM stage was cT_3-4aN_0-2M₀ as depicted on CT. According to the inclusion criteria, we collected 175 consecutive cases. The exclusion criteria were any of the following: 1) patients received any tumor-related treatments (e.g., chemotherapy or radiotherapy) before undergoing CT (n = 10), 2) patients had concomitant malignant tumors of another type (n = 2), or 3) the quality of the images was not good enough (n = 5). Therefore, 17 cases were excluded from our study, and 158 patients were ultimately included in the study. All participants were randomly assigned to a training cohort (TC, n = 121) and a validation cohort (VC, n = 37) with SPSS (version 25, Chicago, IL, USA) (20). All the patients underwent CT examinations before and after two cycles of neoadjuvant chemotherapy plus immunotherapy. The age, gender, and anatomic distribution of the tumor in the TC and VC are depicted in Table 1 .

All image data were acquired using a 64-section multidetector computed tomography (MDCT) (LightSpeed VCT, GE Medical Systems, USA). Patients were orally given 100 ml to 200 ml of water as a negative contrast agent before CT data collection. The CT examinations were performed in a supine position. First, the routine unenhanced scans were performed. Subsequently, a 70–100-ml contrast agent (Omnipaque, Iohexol, GE Healthcare, USA), calculated based on a ratio of 1.5 ml/kg body weight, was injected into an antecubital vein with a 20-G needle at the rate of 3.0 ml/s, and then, 20 ml of saline was rinsed with a pump syringe (Vistron CT Injection System, Medrad, USA). Twenty to 30 s after the contrast injection, contrast-enhanced CT data were obtained. The CT scanning parameters were as follows: peak voltage 120 kV, tube current 200 mA (using automatic exposure control), rotation time 0.5 s, collimation 64 × 0.6 mm, pitch 0.9, section thickness 5 mm, and matrix 512 × 512 mm. The CT scan covered the area from the neck to the middle of the left kidney. During one breath-hold with a full-held inspiration for 10–15 s, each examination was performed. Finally, CT data were transmitted directly to the General Electric Advantage Workstation 4.4 at the mediastinal window settings, and the window width was 400 HU with a window level of 40 HU.

All enrolled patients received two cycles of neoadjuvant chemotherapy plus immunotherapy. The concrete treatment drugs were as follows (21, 22): taxel (docetaxel or paclitaxel) and platinum doublet (cisplatin or carboplatin) plus immunotherapy (sintilimab, pembrolizumab, or camrelizumab). The details of the above chemotherapy were as follows: during each cycle of the 21-day duration, cisplatin or carboplatin (75 mg/m²) on day 1, docetaxel or paclitaxel (135 mg/m²) on days 1 and 8, and sintilimab, pembrolizumab, or camrelizumab (200 mg) on day 1 were administered intravenously. All patients were re-evaluated with a CT scan of the chest and upper abdomen approximately 4 to 6 weeks after neoadjuvant chemotherapy plus immunotherapy. The neoadjuvant regimens of the patients in the TC and VC are depicted in Table 1 .

Due to the absence of the phenomenon of pseudoprogression in this study, tumor response was evaluated based on the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 as follows (23): complete response (CR) was the disappearance of all target lesions; partial response (PR) was at minimum a 30% reduction in the sum of the longest target lesion diameters using the sum of the longest target lesion diameters at baseline as reference; progressive disease (PD) was at minimum a 20% increase in the sum of the longest target lesion diameters using the minimum sum of the longest target lesion diameters recorded since the start of treatment (rock bottom) as reference, or the appearance of at least one new lesion; and stable disease (SD) was neither PR nor PD. The patients with CR or PR according to RECIST (v.1.1) within two treatment cycles were considered “responders,” while those who suffered from PD or SD were considered “non-responders.” As reported (24), the application of iRECIST had no impact on response-related endpoints, compared to RECIST 1.1, and we used RECIST 1.1 to assess the tumoral treatment response.

The pretherapeutic CT features of advanced ESCC were assessed on the 3D-slicer 4.11. The esophageal wall thickness that exceeded 5 mm on axial contrast-enhanced CT was considered abnormal thickness caused by the tumor (25). The pretherapeutic CT features included cT stage, cN stage, anatomic distribution, gross tumor volume (GTV), gross volume of all enlarged lymph nodes (GVALN), and tumor length. The shape of the esophageal tumor was manually outlined around the abnormal tissue on enhanced CT ( Figure 1 ) independently by two radiologists (observer 1 with 3 years of expertise in radiology and observer 2 with 4 years of expertise in radiology), and then the above software automatically calculated the GTV. The volume of each enlarged lymph node was also obtained by the previous two radiologists in a way similar to the previous GTV, and GVALN was obtained by the sum of the volume of each enlarged lymph node. The tumor lengths were independently assessed by observers 1 and 2 on axial and sagittal contrast-enhanced CT images as follows: the upper and lower edges of the tumor on axial contrast-enhanced CT images were determined and marked, and then we measured the tumor length on sagittal contrast-enhanced CT images through the markers with the referring standards of double contrast barium examinations. The measurements of GTV, GVALN, and tumor length of ESCC independently by the above two radiologists were used to test the interobserver reproducibility. Observer 1 remeasured the GTV, GVALN, and tumor length of all target lesions 1 month later to test the intraobserver reproducibility. Before their measurements, a radiology professor with 25 years of radiology experience taught them how to assess quantitative CT tumor features randomly in 20 patients.

For assessing the pretherapeutic categorical CT features including cT stages, cN stages, and anatomic distribution, observer 1 discussed with observer 2 to reach a consensus. When observer 1 and observer 2 disagreed with each other, they could consult the above professor. The cT and cN stages were assessed using the pretherapeutic CT data based on the 8th edition of the American Joint Committee on Cancer (AJCC) staging of cancers of the esophagus and esophagogastric junction (26). For the cT stage, cT₃ of the advanced tumor infiltrated the adventitia, and the cT₄ tumor infiltrated adjacent structures. For the cN stage, the number of involved nodes determines the N stage: the N₀ stage involves no node, the N₁ stage involves one to two nodes, the N₂ stage involves three to six regional nodes, and the N₃ stage involves seven or more regional nodes. Based on the AJCC staging of cancers of the esophagus and esophagogastric junction, the short axis of nodes can be easily measured on CT, and the intrathoracic and abdominal lymph nodes >1 cm can be considered enlarged. All evaluations were performed without knowledge of the histological results.

Statistical analyses were performed by SPSS (version 25, Chicago, IL, USA). According to the published study on pancreatitis using similar statistical analysis (27), univariate analysis was performed to identify the pretherapeutic CT features associated with the response prediction. Chi-squared test or Fisher’s exact test was used to compare categorical variables, and the independent-sample t-test or Mann–Whitney U test was used to compare continuous variables. Binary logistic analysis was performed for statistically significant CT features from the univariate analysis and showed no multicollinearity (27). A P <0.05 was considered statistically significant for all statistical tests.

The R software (version 4.2.2) was used to develop and validate the nomogram. For the development of the nomogram, only independent predictive factors determined by binary logistic analysis were selected. In our nomogram, the regression coefficient of each independent predictive factor in binary logistic regression was proportionally transformed into a specific number on a scale of 0 to 100 points. Moreover, the accuracy of the predictive nomogram was assessed using the concordance index (C-index) and the calibration curves. The receiver-operating characteristic (ROC) curve and the largest area under the receiver-operating characteristic curve (AUC) were obtained with the optimal cutoff point in the nomogram. Additionally, we used decision curve analysis (DCA) to validate the clinical application value of our model because DCA is a novel algorithm for evaluating the net benefit value of a model under different thresholds (28).

In the TC, 9 (7.4%), 60 (49.6%), 41 (33.9%), and 11 (9.1%) patients demonstrated CR, PR, PD, and SD, respectively. In the VC, 2 (5.4%), 20 (54.1%), 11 (29.7%), and 4 (10.8%) patients demonstrated CR, PR, PD, and SD, respectively. Hence, there were 69 responders and 52 non-responders in the TC and 22 responders and 15 non-responders in the VC.

Intraobserver and interobserver agreements of GTV, GVALN, and tumor length are shown in Table 2 . Both intra- and interobserver ICC values of the above measurements were >0.90 (P < 0.001 for all). Thus, the GTV, GVALN, and tumor length obtained by the first measurements from the first observer were used for the subsequent analyses.

The pretherapeutic CT and clinical features are listed in Table 1 . According to our univariate analysis, the cT stage, cN stage, GTV, GVALN, and tumor length were associated with response in patients with ESCC after neoadjuvant chemotherapy plus immunotherapy (P < 0.05 for all). In detail, patients with lower cT stage, lower cN stage, lower GTV, lower GVALN, and lower tumor length were more likely to be responders (P = 0.001, < 0.001, = 0.001, = 0.002, and < 0.001, respectively). However, there were no statistically significant differences in gender, age, neoadjuvant regimen, and anatomic distribution between responders and non-responders (P = 0.376, 0.717, 0.617, and 0.094, respectively).

As for potential independent predictive factors for therapeutic response after neoadjuvant chemotherapy plus immunotherapy including cT stage, cN stage, GTV, GVALN, and tumor length, the binary logistic regression analysis was used for identifying the independent predictive factors. The results of the binary logistic regression analysis indicated that cT stage, cN stage, and tumor length were independent predictive factors in patients with advanced ESCC (P = 0.039, 0.001, and 0.002, respectively).

Based on the above binary analysis in the TC, a nomogram was constructed to predict the response of patients with ESCC to neoadjuvant chemotherapy plus immunotherapy that incorporated the three independent predictive factors: cT stage, cN stage, and tumor length ( Figure 2 ). A total score was calculated with the use of cT stage, cN stage, and tumor length which were reachable on CT. In our nomogram, the value of each variable was scored on the axis of the score scale. By adding each score, the total score could be easily calculated, and we were able to predict patients’ response after the neoadjuvant treatment by projecting the total score onto the total point scale.

In the TC and VC, an individual calibration curve was used to confirm the accuracy of the nomogram based on the predictive nomogram, showing that the predicted values of our model coincided with the actual values ( Figure 3 ). The AUC values of 0.813 [95% confidence interval (CI), 0.735–0.890] in the TC and of 0.797 (95%CI, 0.643–0.950) in the VC suggested excellent predictive power, respectively ( Figure 4 ). DCA demonstrated good clinical application of this nomogram in the TC and VC ( Figure 5 ).