A cohort of 1160 consecutively recruited patients with EC treated by radiotherapy at our institute was retrospectively analyzed from 2010 to 2019. The criteria for inclusion were as follows: 1) pathologically confirmed ESCC; 2) Karnofsky performance status (KPS) score ≥70; 3) no distant organ metastasis; 4) no history of a concomitant or previous malignancy; 5) underwent intensity-modulated radiation therapy (IMRT)-based definitive concurrent chemoradiotherapy; and 6) had unresectable EC or refused surgery. A total of 599 patients with ESCC met the above criteria and were selected for analysis. The flowchart of the screening process was shown in Figure 1 . The study was approved by the Ethics Committee of Tianjin Medical University Cancer Institute and Hospital. The patients were not required to sign an informed consent form for this retrospective study.

The end points included the overall survival (OS) rate, the progression-free survival (PFS) rate and treatment-related toxicity. We defined OS as the time from the first treatment to the date of death from any cause or the date of last follow-up. The PFS rate was calculated from the time of first treatment to disease progression, including local recurrence (LR), regional recurrence (RR), and distant metastasis (DM). Patients who did not experience an event of interest were censored at their last follow-up or the date of death.

The toxicities of treatment were evaluated according to the Common Terminology Criteria for Adverse Events (CTCAE, version 4.0). Patients attended regularly scheduled follow-up physical examinations, blood tests, chest and abdominal CT scans, barium esophagography, and ultrasonography at 3-month intervals for the first two years, every 6 months for the next three years, then annually.

Descriptive statistics were performed to examine the baseline characteristics of all patients. Age, KPS, radiation dose and primary tumor length were categorized with the median value as the cutoff. Grouping by the maximum lymph node diameter was performed as previously described (7). Continuous variables are shown as the means with standard deviation and were compared using the Mann-Whitney U test. Categorical variables were compared by using the chi-square test or Fisher’s exact test. Median follow-up was reported with reverse Kaplan-Meier estimate (10). Kaplan-Meier analysis with a log-rank test was used to compare the differences in the PFS and OS rates between the two groups. We used univariate and multivariate Cox proportional hazard models to examine the influence of different variables on the OS rate. Hazard ratios and their associated 95% confidence intervals (95% CIs) were obtained from the Cox regression analysis. To minimize potentially confounding effects between the two treatment groups, propensity score matching (PSM) was performed using Stata software. Using the OS rate as an endpoint, a clinical scoring model was performed by using the significant variables obtained from the multivariate analysis. The score was the weighted sum of the significant variables of which the weights were defined as the quotient of the corresponding estimated coefficient from a Cox regression analysis divided by the smallest chi-square coefficient (11, 12). All statistical tests were two-sided and P values < 0.05 were considered statistically significant. Analyses were conducted using SPSS v22.0 (IBM SPSS, New York), and figures were created using the Stata software program (version 14.1; StataCorp, College Station, TX, USA).

A total of 599 eligible patients with ESCC were included in the study, and they were divided into the following two treatment groups: IC followed by dCCRT (IC group, n=182) and dCCRT alone (non-IC group, n=417). The patient clinical and treatment characteristics are summarized in Table 1 . The median age of the whole cohort at diagnosis was 61 years (ranging from 30 to 86 years old). The sex ratio of males to females was 6:2. The median primary tumor length was 6.0 cm (range, 1.2-18.0 cm).

The median radiation dose was 54 Gy in the whole cohort (range, 41.4-70Gy). In our cohort, 65/599 patients (10.9%) developed esophageal fistula.

The differences in sex, smoking history, weight loss, pain in the chest and back, tumor location, clinical T stage, tumor length, and radiation dose were not statistically significant between the two groups (P > 0.05, Table 1 ). There were more younger patients in the IC group than the non-IC group (P=0.033). Compared with those in the non-IC group, patients in the IC group had a higher ratio of KPS ≥90 (78.0% vs. 60.7%, P<0.001), a higher rate of N1-3 (79.1% vs. 78.2%, P=0.022), a larger maximum lymph node diameter (58.2% vs. 44.4%, P=0.006), and a higher proportion of adjuvant chemotherapy (48.4% vs. 36.0%, P=0.004).

For patients treated with IC, the median number of cycles administered was 2 (range 1–6). A total of 59/182 (32.4%) patients received the three-drug combination IC regimen, including taxane (paclitaxel or docetaxel), platinum (nedaplatin, oxaliplatin or cisplatin), and fluoropyrimidine (5-fluorouacil, tegafu). Two drug treatment regimens were prescribed in 122/182 (67.0%) patients, with 75 patients receiving paclitaxel-platinum combined chemotherapy, 45 patients receiving platinum in combination with fluoropyrimidine, 1 patient receiving gemcitabine and cisplatin, and 1 patient receiving vinorelbine and nedaplatin. The remaining patient received single-agent paclitaxel therapy. The details of the use of IC were shown as Supplementary Table 1 . The regimens were selected based on patient age, general physical condition, response to chemotherapy, patient willingness, and physician judgment.

For subsequent adjuvant chemotherapy (ACT), a total of 238 patients (39.7%) receiving ACT following dCCRT in our study, the median number of chemotherapy cycles was 2 (range, 1-6). For patients treated with ACT, 51 patients (21.4%) had 1 cycle of ACT, 84 (35.3%) had two cycles, 45 (18.9%) had three cycles, 39 (16.4%) had four cycles, 6(2.5%) had five cycles, and 13(5.5%) had six cycles. A total of 11/238 (4.6%) patients received the three-drug combination of regimens based on platinum (oxaliplatin, cisplatin, nedaplatin) combined with taxane (paclitaxel or docetaxel) and fluoropyrimidine (5-fluorouacil, tegafu). Two ACT regimens were used in 217/238 (91.2%) patients, with 163/217 (75.1%) patients receiving paclitaxel-platinum combined chemotherapy, 50/217(23.0%) patients treated with platinum in combination with fluoropyrimidine, and 4/217(1.8%) patients receiving a combination of irinotecan and fluoropyrimidine. And 10/217(4.2%) patients were treated with single-agent chemotherapy regimen of tegafu.

Tumor response was evaluated in 96 patients (52.7%) after IC. Specifically, complete response (CR) was achieved in 1 patient (1.0%), partial response (PR) in 44 patients (45.8%), stable disease (SD) in 40 patients (41.7%), and progressive disease (PD) in 11 patients (11.5%). Overall, a total of 559 patients (93.3%) had post‐treatment tumor assessments. CR was obtained in 25 patients (4.2%), PR in 362 patients (60.4%), SD in 131 patients (21.9%), and PD in 41 patients (6.8%) for the whole cohort, respectively.

The observed treatment-related toxicities graded using the CTCAE are shown in Table 2 . The most common acute toxicities were radiation esophagitis and leukocytopenia, occurring in 43.9% and 52.0% of all patients with recorded toxicities. The incidence of hematological toxicities grade ≥2 in the IC group was higher than that in the non-IC group (leukocytopenia, 27.3% vs. 25.1%; anemia, 11.1% vs. 6.9%; thrombocytopenia, 9.3% vs. 6.1%), although the difference was not statistically significant (p=0.085, p= 0.068, p=0.077, respectively). There were no significant differences in the rates of radiation esophagitis, radiation pneumonitis and radiodermatitis between the two groups (P=0.225; P=0.205; P=0.663). No treatment-related deaths or grade 5 toxicities occurred. The comparison of radiation dose to OAR between the two groups is shown in Table 3 . No differences were observed in lung doses, heart doses or maximum spinal cord doses between the two groups (all p>0.05).

The median follow-up period was 33 months for the entire cohort. In total, 339 (56.6%) patients experienced disease recurrence, and 301 (50.3%) died over the course of follow-up. Distant metastases occurred in 151 patients (25.2%) and the metastatic sites included the lung, bone, liver, brain, pleura, kidney, adrenal gland, lymph nodes and soft tissue. Of these patients, 122 developed distant metastases in a single organ: 66 lung metastases, 20 liver metastases, 19 bone metastases, 8 soft tissue metastases,4 brain metastases, 3 pleura metastases, 1 kidney metastases, and 1 adrenal gland metastases and 29 had multi-site metastases. Ninety-two patients in the IC group experienced recurrence, of which 54 (29.7%) experienced local recurrence (LR), 44 (24.2%) experienced regional recurrence (RR), and 43 (23.6%) experienced distant metastases (DM). In the non-IC group, a total of 247 (59.2%) patients experienced relapse, of which 153 (36.7%) experienced LR, 105 (25.2%) experienced RR and 108 (25.9%) experienced DM. There were no significant differences in the LR, RR or DM rates between the two groups (P=0.097, P=0.794, P=0.556, respectively).

The median PFS and OS rates for the entire group of patients were 15 months (95% CI, 12.4-17.6) and 25 months (95% CI, 19.7-30.3), respectively. The 1-, 3-, and 5-year PFS rates were 58.1%, 39.3%, and 33.4% in the IC group and 52.2%, 29.1%, and 24.7% in the non-IC group, respectively (P=0.078) ( Figure 2A ). No significant difference in OS was observed between the IC group (1-year OS, 71.4%; 3-year OS, 40.3%; 5-year OS, 30.3%) and the non-IC group (1-year OS, 69.9%; 3-year OS, 43.5%; 5-year OS, 32.9%) (P=0.532) ( Figure 2B ).

We used univariate and multivariate Cox proportional hazard regressions to analyze prognostic factors influencing the OS rate in patients with ESCC ( Table 4 ). The univariate analysis revealed that the OS rate was significantly associated with the following factors: weight loss (p=0.038), tumor location (p=0.002), clinical T stage (p=0.010), clinical N stage (p<0.001), tumor length (p<0.001), maximum lymph node diameter (p<0.001) and adjuvant chemotherapy (p=0.004). Further multivariate analysis showed that clinical T stage (P=0.029), tumor length (p=0.004), maximum lymph node diameter (p<0.001) and adjuvant chemotherapy (p=0.007) were independent factors affecting patient survival ( Table 4 ).

The parameters with statistical significance in the multivariate analysis are shown in Table 4 . The score of each parameter is weighted according to its relative contribution determined by the chi-square score. Specifically, the score for each variable was calculated by dividing its chi-square value obtained from the Cox regression analysis by the smallest chi-square coefficient. The score ranged from 0 to 10.4 after adding up these variables. Using the median value as a cutoff point, we then separated patients into high (>5.5) and low (≤5.5) risk groups. We evaluated the PFS rates of the high-risk group (n=291) and low-risk group (n=308), and the PFS rate of the high-risk group was significantly shorter than that of the control group (p<0.001, Figure 3A ). Moreover, patients in the high-risk group had worse OS rates than those in the low-risk group (p<0.001, Figure 3B ).

The baseline clinical and treatment characteristics of the high-risk group of 291 patients are shown in Table 5 . A total of 106 patients received IC followed by dCCRT, and 185 underwent dCCRT alone. The baseline characteristics were similar between the two groups, with no significant differences. The PFS rate of patients receiving IC was comparatively higher than that of patients who did not receive IC (P=0.034, Figure 4A ). The prognosis of the IC group, with a 5-year OS of 21.6%, was comparable to that of the non-IC group, with a 5-year OS of 18.8% (P=0.794, Figure 4B ).

In total, 308 patients in the low-risk group received IC (IC group), and 232 patients did not receive IC (non-IC group), as shown in Supplementary Table 2 . Kaplan-Meier analysis of the PFS and OS rates revealed no significant difference between the groups (PFS: P=0.207, Figure 5A ; OS: P=0.997, Figure 5B ). The baseline characteristics were balanced between the two groups using propensity score matching (PSM) ( Supplementary Table 2 ). Fifty patients who received IC were matched with thirty-five patients without IC treatment by PSM analysis. Likewise, no significant differences were found between the two treatment groups in the PFS (P=0.772) or OS rates (P=0.777) ( Figures 6A, B ). Furthermore, In low-risk group, we found patients showed a longer OS treated with ACT (P=0.003), while no significant difference in PFS (P=0.224). The survival analysis revealed that ACT affected the OS of patients (P=0.049) but had no significant impact on PFS (P=0.261) in high-risk group.

Because the wide variation in the IC regimen, we then investigated whether the number of chemotherapy combination and the regimens would have an effect on PFS in patients with high-risk group. In patients who received two drug treatment regimens (66/291, 22.7%) or three drug combination regimens (39/291, 13.4%), no significant difference in PFS (P=0.259) between two groups was observed. In addition, for 40 high-risk patients treated with taxane in combination with platinum and 25 patients receiving platinum-fluoropyrimidine combined chemotherapy regimens, similarly, the PFS did not differ for the two different regimens (P=0.591).