It is well known that ESCC is associated with poor prognosis and the five-year overall survival rate is about 20%-40%. More and more research is being directed at developing better treatment and improving prognosis for patients diagnosed with ESCC. Previous studies suggested that certain factors had an effect on the prognosis of ESCC patients however, the systematic analysis of these factors as a prognostic took has not been done[19–24]. In the present study, we have analyzed the ESCC patients’ data from two academic institutes, SYSUCC and HNCH. Then, we divided randomly data of SYSUCC into two groups, training group (SYUCC-A, N = 555) and internal validation group (SYUCC-B, N = 241). Next, we analyzed the information of training group by applying univariate and multivariable Cox regression. We have identified the five following factors: age, Alb levels, T stage, LN classification, and tumor differentiation as independent prognostic factors. To construct the SPM, we processed the data based on assessing these five significant variables mentioned above in the training cohort (Table 3). We regarded SC as prognostic score of patients, which was derived from SPM. Mean SC was presented as a cutoff value for classifying the prognosis. In the present study, we found that our SPM could provide a stratified effect on the patients’ prognosis, and we could detect the group with the poor after esophagectomy (Fig. 2A, HR 1.009, 95% CI, 1.006–1.011, P < 0.001). To validate the SPM, we used the same method and cutoff values in the internal validation and external validation cohorts. As we have expected, there was a strong stratified effect in the validation cohorts (Fig. 2BC). The SPM yielded a C index of 0.654 ± 0.022 (SYSUCC-A), 0.630 ± 0.034 (SYSUCC-B) and 0.688 ± 0.04 (HNCH) for OS, respectively. In the present study, we have established and validated a survival prognostic model (SPM) that could predict the OS for patients with stage T1-3N0M0 ESCC patients, and might help clinicians to screen subgroups in order to identify patients with poor prognosis. To further explore the stratified effect of SPM in the different T stages, we used the K-M analysis to draw a survival curve. The results, as expected, had shown that SPM had a better distinction power in stage T2-3 (all P < 0.05) than in stage T1 (P = 0.061, Fig. 3).
The selected factors from clinical characteristics could be easily obtained from patient’ medical record. We could get the information on the age at the time of diagnosis from patients’ medical record, and obtain additional data on tumor differentiation, T stage, and the number of lymph nodes removed from patients’ pathological reports. Preoperative level of Alb could be accessed using the results of labor test. Five factors listed above were very common in patients’ records and could be easily obtained, which was very beneficial for the testing of our model. In addition, the areas with high incidence of esophageal cancer in China were mainly located in Henan Province, Hebei Province, Chaoshan region (Guangdong Province), and other regions[25–27]. SYSUCC is located far from the HNCH, and both cancer centers accept patients with ESCC across China. Our model was constructed using the cases from SYSUCC-A and further validated using the cases from both: SYSUCC-B and HNCH. Model validation using two patients’ cohorts from SYSUCC-B and HNCH also fully illustrated the applicability of the five-factor model presented in this study. We excluded the ESCC patients with stage T4 and those with metastases in lymph nodes and other organs because previous study showed that neoadjuvant therapy could improve the prognosis in patients who were diagnosed with metastases in lymph nodes before surgery. Accordingly, in the present study, we classified the LNs according to the approach used in our previous study.
However, there are some limitations in the study presented here. First of all, the sample size of ESCC patients was not large enough; the T stage was restricted to only the T1-3, and the data distribution of the T stage was not balanced. For further work, to improve this aspect, the sample size would need to be expended. Second, the patients’ data used in this study originated from two academic institutes only, which might also impact the accuracy of our model, so a larger scale using patients’ information from multiple centers was necessary to confirm our findings. While the results presented here proved that our five-factor prognosis model could be useful when predicting OS for postoperative T1-3N0M0 ESCC patients, data from more centers would still be required to verify the applicability of this model. Third, this model could only provide a certain reference information to the clinicians but not the treatment recommendations. The doctors would need to make decisions on the patients’ treatment according to the relevant guidelines and clinical experience. Fourth, the follow-up time period for the HNCH cohort was not long enough. The main observation end-point was 4-year OS for the patients who underwent surgery during the period between January 2014 and December 2015. More follow-up research should be done for evaluating prognosis for these patients. Fifth, the patients’ data was lacking the results of molecular diagnosis, such as the information on the expression of Programmed death-ligand 1 (PD-L1) which is an effective prognostic factor that could contribute to the more accurate prediction[30, 31]. Therefore, a substantial research at the molecular level is still needed for further improvement of the proposed model.
In conclusion, we have established and validated a SPM that can predict the OS for patients with stage T1-3N0M0 ESCC and could help clinicians screen for subgroups with poor prognosis. In addition, we suggested that this model from two academic institutes had a value of utilization for other researchers apply on their own data.