Recently, several disease centers have published their own treatment experience, but each single-center study of the disease was generally limited by the sample size, which had a certain impact on the accuracy of the results. Modified Kadish staging system was the most widely used ENB staging system [10,22]. Recent studies have shown that the survival of patients with ENB was significantly associated with pathologic grade and age [20, 23]. To the best of our knowledge, this is the first study to use the nomogram model to combine age, pathologic grade, and modified Kadish staging systems to predict the prognosis of patients with ENB.
Although surgery and chemotherapy were associated with patient outcomes in univariate analysis, they were not prognostic independent predictors in multivariate analysis. Patients receiving chemotherapy often have large local tumors or distant metastases [6]. For patients with locally advanced tumors, chemotherapy decreased the risks of systemic failure by acting on systemic micrometastasis [24]. For patients with distant metastases who did not undergo radical surgery, chemotherapy may be a suitable treatment and control the lesions. Compared with the SEER cohort, the Chinese cohort had relatively fewer patients undergoing surgery and a higher proportion of patients receiving chemotherapy. One reason for this phenomenon was that the proportion of distant metastasis in the Chinese cohort was significantly higher than that in the Western cohort (26.4% vs. 11.6). This may be due to bias caused by too small a sample size. Another potential reason was that Sun Yat-sen University Cancer Center and The First Affiliated Hospital of Sun Yat-sen University were two famous hospitals in China, a bias toward more advanced disease among those referred to these two hospitals. Last but not least, the lack of ascertainment in SEER and the inevitable selection bias might be weighted towards a surgical group.
The role of age in ENB is still controversial and unclear. In this study, the best cutoff values of 54 and 70 were calculated by X-tile [25], and the prognosis was the best in the group of patients younger than 54 years old. Although these patients all received the same treatment strategy, this study still showed different survival trends in three groups. Yin et al. showed that patients older than 60 years of age had a worse prognosis [23]. Previous studies have shown that young patients with ENB have more aggressive disease, but these patients are sensitive to chemotherapy and can achieve good results through a combination of chemotherapy and radiotherapy [26]. We recommend that young, locally advanced patients enter Multi-Disciplinary Therapy Meeting (MDT) to discuss and determine treatment options.
The standard for pathological grading of ENB is the Hyams standard. At present, some studies have reported that pathological graded survival was significantly correlated and was an independent predictor of survival in patients with ENB [19, 27-29]. For Chinese cohort, the tumor differentiation criterion based on the Hyams Grading System while the SEER cohort used the tumor differentiation grading scheme. The indicators for evaluating cell differentiation included mitotic index and nuclear polymorphism, which were also part of the Hyams scoring system. Limited by the SEER database, it could not provide Hyams grading information, but the impact on nomogram might be slight. Tajudeen et al. considered that tumor differentiation grading scheme roughly corresponded to the Hyams grading scale [19]. Significant differences in survival can be seen in the pathological graded polarization of ENB, and high-grade pathological differentiation grades tend to have a worse prognosis [28]. In this study, we defined grade I and grade II tumors as low-grade tumors and defined grade III and grade IV tumors as high-grade tumors. In these two groups of patients, we observed significant differences in both training cohort and validation cohort, while in multivariate analysis, high-grade tumors were risk factors for prognosis. For SEER grading scheme may not be interpreted as a true Hyams grade, but it roughly corresponds to the Hymas grading scale. The bias caused by this method requires a large sample size cohort containing Hyams grading information as training cohort to reconstruct a nomogram. However, due to the rarity of ENB, the SEER database was the largest cohort that could be obtained, and variability was minimized by grouping patients into low-grade and high-grade tumor groups.
A number of studies evaluated the predictive power of the modified Kadish staging system [6, 22, 30]. Although it was partly confirmed that the modified Kadish staging system can effectively predict the prognosis of patients, some of them did not confirm its predictive efficacy. The reason for this phenomenon was the lack of sample size, selection bias, or a defect in the modified Kadish staging system itself. In the present study, we did not find statistically significant differences in survival between modified Kadish A and B, either in the SEER cohort or in the Chinese cohort. This was consistent with the conclusions of some previous studies [30]. Therefore, to improve statistical performance, stage A and stage B ENB patients were combined together. The prognosis of these patients was significantly better than that of patients with stage C and stage D disease.
The prognostic significance of clinical staging and pathologic grading were perhaps confounded often by each other [6, 27]. These factors explained the limitations of using pathological grading and clinical grading alone to some extent. The use of the modified Kadish staging system and pathological grading system was not sufficiently accurate. We established a nomogram to predict the prognosis of patients. Based on the Cox regression risk model, the model calculated the likelihood of 3-year and 5-year survival based on the patient’s age, clinical stage and tumor pathologic grade. Clinical application was simple and convenient. Here, we demonstrated a nomogram application example based on the calculation of the nomogram. The patient was a 55-year-old male who was diagnosed with clinical stage C in 2012. The tumor pathologic differentiation grade was a low-risk group (grade I well differentiated), and the patient underwent both surgery and radiotherapy. According to the nomogram, the measured probability of 3-year survival was about 83%, and the 5-year survival was about 80%. This patient was in the low-risk group. When this new method of evaluating prognosis is extended to patients in the non-SEER cohort, we recommended using this nomogram after the validation step. It can reduce the bias caused by selection and regional differences in medical levels. It may be necessary to build a nomogram with the data of the non-SEER population, and then determine the cut-off value of the risk stratification according to the actual situation of the population by using our methods.
The c-indexes for internal and external validation were 0.737 (95% CI, 0.709 to 0.765) and 0.791 (95% CI, 0.767 to 0.815), respectively, which showed that the present nomogram was a repeatable and accurate prognostic tool for predicting 3- and 5-year OS in patients. However, some of ENB patients could have a long naturel history, OS may not be the most relevant endpoint. Quality of life for ENB patients was one of dominant components of the treatment evaluation. Thus, DFS may be more relevant than OS, but SEER database only provides OS and disease-specific survival as primary endpoints. Further improvement of our nomogram by using patient series with data for quality of life and DFS is needed. Nonetheless, this nomogram could act as a tool to select high-risk patients and make individualized treatment and follow-up schedules.
There were several limitations in our study. First, this was a retrospective analysis study that inevitably had a selective bias. One of the enrollment criteria used positive histology only, which resulted in some patients who did not receive surgery and lacked pathological data were excluded from the cohort. A total of 599 patients were excluded due to unknow demographic and clinicopathological information and 279 out of 599 patients (46.6%) had no surgery performed. The proportion of excluded patients who received surgery was lower than the proportion of enrolled patients (53.4% vs. 90.2%). This might increase the proportion of patients in the SEER cohort who received surgery. Second, the SEER database did not provide detailed chemotherapy information. In this study, we were unable to confirm information about the course of treatment for patients with chemotherapy, which may lead to bias in the treatment results. In addition, the detailed radiation therapy data are not provided, and it was hard to evaluate the treatment impact on the SEER cohort’s patients. Third, the SEER database did not provide the patient’s surgical methods, so it was impossible to make comparisons on the influence of the surgical approach. Finally, the SEER database did not provide patients’ information about modified Kadish stage and Hyams grade. Modified Kadish staging transformation depended on the accurateness of SEER data and its coding system. Nonetheless, the results were still novel, we successful provided insight into the utility of the nomogram and to verify the repeatability and practicability of the nomogram in validation cohort.