Prognosis and nomogram for predicting postoperative survival of duodenal adenocarcinoma: A retrospective study in China and the SEER database

As primary duodenal adenocarcinoma is rare, the prognostic factors of this disease remain insufficiently explored, especially in China. We identified postoperative duodenal adenocarcinoma patients at a Chinese double-center (from 2006 to 2016) or who were registered with the Surveillance, Epidemiology, and End Results (SEER) database (from 2004 to 2014). Clinicopathological features and significant prognostic factors for cancer-specific survival (CSS) were reviewed and analyzed by using univariate and multivariate Cox proportional hazards regression. Then, a nomogram predicting CSS was constructed based on the SEER database and validated externally by using the separate Chinese cohort. Totally, 137 patients from the Chinese double-center and 698 patients from the SEER database were included for analysis. The multivariate analyses showed that age, tumor grade and TNM stage were independent prognostic factors. The nomogram constructed using these factors showed a clear prognostic superiority to the AJCC-TNM classification, 7th ed. (C-index: SEER cohort, 0.693 vs 0.625, P < 0.001; Chinese cohort, 0.677 vs 0.659, P < 0.001, respectively). In summary, the valuable prognostic factors in patients with duodenal adenocarcinoma were age, tumor grade and TNM stage. This study developed a nomogram that can precisely predict the CSS for postoperative duodenal adenocarcinoma patients.

period, 49 out of 137 (35.8%) patients died of cancer-associated death in the Chinese cohort compared with 308 out of 698 (44.1%) patients in the SEER cohort. The 3-and 5-year cancer-specific survival (CSS) values were 67% and 39% in the Chinese cohort and 52% and 38% in the SEER cohort, respectively. Histopathologically, most patients had moderately differentiated tumors in the Chinese and SEER cohort (n = 68, 49.6% vs n = 365, 52.3%). However, patients with duodenal adenocarcinoma tended to present with more poorly differentiated tumors in the SEER cohort than those in the Chinese cohort (n = 264, 37.8% vs n = 28, 20.4%). According to the AJCC-TNM classification, 7 th ed., patients in the SEER cohort were approximately twice as likely to be diagnosed as having stage III cancer than those in the Chinese cohort (55.4% vs 26.3%). Conversely, the percentage of patients with stage I cancer in the Chinese cohort was significantly higher than that of patients in the SEER cohort (26.3% vs 8.9%). The majority of the patients had pathologic T4 cancer in both the Chinese and SEER cohorts. Patients in the SEER cohort were more likely to present with LN metastasis than those in the Chinese cohort (61.4% vs 30.7%). Distant metastasis was found in 16 patients (11.7%, including 9 patients with liver metastases, 4 with omentum metastases, and 3 with abdominal wall metastases) in the Chinese cohort vs 48 patients (6.9%) in the SEER cohort. In addition, tumor vascular and perineural invasion occurred in 18.2% and 17.5% of the Chinese cohort, respectively.
Univariate and multivariate analyses of CSS prognostic factors. In the univariate analysis, age, tumor grade, tumor size and TNM category were closely related to CSS in the SEER cohort, but only LN and distant metastasis was associated with CSS in the Chinese cohort ( Table 3). The multivariate analyses identified five variables, including older age, advanced grade and TNM category, to be significantly associated with CSS in the SEER cohort. However, in the Chinese cohort, only LN and distant metastasis was statistically significantly different in the CSS (

Risk of lymph node metastasis.
Patients with stage I to III duodenal adenocarcinoma were grouped according to LN metastasis (Table 5). T-stage classification was independently associated with the increased risk of LN metastasis both in the Chinese and SEER cohorts ( Table 6). Tumor grade is an independent factor of LN metastasis in the SEER cohort but not in the Chinese cohort. The rate of LN metastasis increased with a higher Construction and validation of the CSS nomogram. The TNM-based nomogram, incorporating all the significant independent factors for predicting 3-and 5-year CSS based on the SEER training cohort, was established. Figure    In the test cohort, the C-index of the TNM-based CSS nomogram was (0.681; 95% CI, 0.642 to 0.719) higher than that of the 7th AJCC system (0.634; 95% CI, 0.593 to 0.675; P < 0.001). Consistently, in the Chinese external validation cohort, the TNM-based nomogram (0.677; 95% CI, 0.634 to 0.719) still showed superior discrimination compared to the 7th AJCC system (0.659; 95% CI, 0.618 to 0.701; P < 0.001).
The calibration plots of the train cohort and the external validation cohort are presented in Fig. 3, which shows the predicted 3-and 5-year CSS probabilities for both the SEER training cohort and the Chinese validation cohort compared with the actual observations.

Discussion
Duodenal adenocarcinoma is a rare cancer affecting approximately less than 0.5 per 100,000 individuals 5 . Due to the low incidence and prevalence of duodenal adenocarcinoma, few studies have been published, and the relevant survival factors are still controversial 11 . In this retrospective study, we evaluated the clinicopathological characteristics and independent prognostic factors in duodenal adenocarcinoma patients who underwent surgical excision from the Chinese double-center and the SEER database. Independent prognostic factors for CSS were related to age, tumor grade and TNM stage. The TNM-based nomogram included these factors and predicted CSS better than the AJCC TNM staging system, 7 th ed.
Several similarities and differences between the Chinese and SEER cohorts were observed in this study. In Chinese cohort, 61.3% were male and 38.7% were female. It is very different from that of the United States (Table 1). Considering this is a retrospective analysis from the double-center, which may result in a selection bias. However, we found our difference of gender incidence in small bowel cancers was in line with two studies of small bowel cancer in the Chinese population [12][13][14] . So there may be difference between the male and female incidence of small bowel cancers in Asian and Caucasians population, and future research is warranted. In the SEER cohort, patients tended to present with more poorly differentiated and advanced stage tumors than those in  Table 4. Multivariate analysis of patients with postoperative duodenal adenocarcinoma in Chinese and SEER cohort.
the Chinese cohort. This is partially because of the differences in race, geographic patterns and diet. International data show that the incidence of small bowel adenocarcinoma in North America, Western Europe and Oceania is higher than that in Asia 12,15 . African Americans have substantially higher incidence rates and worse small bowel adenocarcinoma survival compared to Caucasians 16 . Studies also suggested that dietary factors are related to the risk of small bowel adenocarcinoma 17,18 . High-fat diet led to an increase risk of small bowel tumors in mice. Mechanistically, high-fat diet mediated carcinogenesis may due to DNA damage caused by bile acid. One possible mechanism is that bile acids induce oxidative stress and frequent apoptosis that then causes DNA damage 19 .
Other factors, such as alcohol consumption and cigarette smoking, have been suggested to be associated with the risk of small bowel adenocarcinoma 20 . The disparities between the Chinese and SEER cohorts likely reflect a complex interaction between race, geography, environment, socioeconomic and genetic inequalities.   The prognostic factors of resected primary duodenal adenocarcinoma remain controversial. Ryder et al. demonstrated that larger tumor size, advanced histological grade, and transmural invasion are associated with decreased survival 10 . Qing-Long Jiang et al. revealed that LN metastasis and vascular invasion were independent prognostic factors that were negatively associated with survival in patients undergoing curative resection 21 . In this study, through univariable analysis and subsequent multivariable Cox regression analysis, we identified patients with elder age, worse tumor grade and advanced TNM stage had shorter CSS. Most studies suggested that regional LN metastasis is associated with prognosis [22][23][24] . The incidence of LN metastasis in patients with duodenal adenocarcinoma has been reported to range from 22% to 76% 25 . Our study, in accordance with previous reports, has shown that patients with nodal metastasis had diminished survival in both the Chinese and SEER cohorts. However, patients in the SEER cohort were approximately twice as likely to be diagnosed as having LN metastasis cancer than those in the Chinese cohort (61.4% vs 30.7%). This may be due to the differences in the T-stage classification between the Chinese cohort and the SEER cohort. T-stage classification was the strongest predictor of LN metastasis in our study, as reported by a matched cohort study based on the National Cancer Database 26 .
The application of nomograms in individualized risk prediction and stratification by incorporating TNM stage and other key prognostic factors is well recognized in a wide variety of cancers, such as prostate, breast, gastric and colorectal cancer [27][28][29][30] . In this study, we first constructed a nomogram based on TNM stage along with other clinicopathologic parameters. We found that the TNM-based nomogram predicts CSS more accurately than the AJCC-TNM staging system (C-index value: 0.693 vs 0.625, P < 0.001) in the train cohort and 0.677 vs 0.659 (P < 0.001) in the Chinese external validation cohort. The calibration plots showed excellent agreement in the training cohort between the prediction probabilities and the actual observations, which ensured the reliability and repeatability of the constructed nomogram. Although there are some differences between the Chinese and SEER cohorts, our nomogram still showed acceptable agreement in the external validation cohort. This nomogram would allow clinicians to identify high risk for poor survival, to make better clinical decisions and provide follow-up surveillance for patients with duodenal adenocarcinoma.
This study has several limitations. Firstly, our retrospective study only included duodenal adenocarcinoma patients received surgical resection, which may result in selection bias. Secondly, variables such as adjuvant chemotherapy and radiation therapy are not available in our study; therefore, some treatment bias is present. Thirdly, the molecular pathologic characteristics are not included in this study, which may result in a limitation on the survival and LN metastasis analysis.
In conclusion, we identified the prognostic factors of duodenal adenocarcinoma patients who underwent curative resection based on two institutions from China and the SEER database. According to the factors, we developed and validated a novel nomogram for predicting postoperative survival of duodenal adenocarcinoma.  The nomogram is easy to use, and it provides clear prognostic superiority over the seventh AJCC-TNM staging system. The nomogram might also help clinicians to make individualized predictions of patient survival and to give improved treatment recommendations.

Materials and Methods
Patient population. We collected two independent Chinese cohorts that consisted of 137 patients with post- Follow up. The primary endpoint of this study was CSS, which was registered as the cause-specific classification of death in the SEER database (alive or dead of other cause or cancer-associated death). CSS represents the survival of a specific cause of death in the absence of other causes of death.
Construction of the nomogram. To construct the effective postoperative CSS nomogram of duodenal adenocarcinoma, we divided the SEER database in two groups randomly. Eighty percent (n = 558) were assigned to the training cohort, and twenty percent (n = 140) were selected as the test cohort. The independent prognostic factors were identified by multivariate Cox proportional hazards regression analysis. Then, a nomogram based on these prognostic factors was constructed by using the train cohort.
Validation of the nomogram. The prognostic performance of the nomogram was evaluated with discrimination and calibration by using the test and external validation cohort (the independent Chinese patient cohort). Discrimination was assessed with the concordance index (C-index). A higher C-index value indicated a better prognostic accuracy. For calibration, the predicted probabilities produced by the nomogram were compared with the actual probabilities. The Kaplan-Meier method and bootstraps with 1000 resamples were used for this purpose 31 . Statistical analysis. Statistical analyses were performed using IBM SPSS version 20.0 and the statistical software package R version 3.4.2. Hazard ratios and their 95% confidence intervals (95% CI) were computed. All P values were 2 sided, and P < 0.05 was considered statistically significant.