Clinical features, treatment and prognosis analysis of distant metastatic esophageal cancer

Abstract Backgrounds : Esophageal cancer (EC) is one of the most common malignant tumors in China. EC is characterized by poor clinical prognosis, with many patients being diagnosed at advanced stages. This study utilized data from the the Surveillance, Epidemiology, and End Results (SEER) database. The clinical features, treatment and prognostic factors of patients with distant metastatic esophageal cancer were screened and analyzed, a nomogram was drawn to construct a prognostic model. Methods Eligible patients with distant metastatic esophageal cancer diagnosed from January 2004 to December 2015 were extracted from SEER database. Propensity score matching(PSM)was used to eliminate baseline differences between groups. The data were divided into training cohort (1116 cases) and validation cohort (426 cases) by using R software and random sampling function at the ratio of 7: 3. The baseline table was plotted using x2 or Fisher's exact test. Kaplan-Meier curve, log rank test and Cox regression were used for survival analysis. C index and AUC were used to evaluate the performance of prognosis model. Calibration curve was used to evaluate the calibration of the model. Using the data of the validation cohort, external validation is used to create prediction model. Results After applying the inclusion and exclusion criteria and PSM, a total of 1,542 cases diagnosed between 2004 and 2015 were included in the study. Before and after PSM, we analyzed Kaplan-Meier survival of patients with metastatic esophageal cancer with different treatment methods. The results showed that radiotherapy, chemotherapy or surgical treatment brought significant survival benefits to patients with metastatic esophageal cancer(P < 0.05). Univariate and Multivariate regression analysis showed that T-stage, M-stage, primary site, surgery, chemotherapy and radiotherapy were independent prognostic factors affecting the prognosis of distant metastatic oesophageal cancer (P < 0.05). Evaluating the predictive ability of nomogram, the C index of the training cohort was 0.69(95%CI:0.67–0.71), and the C index of the validation cohort was 0.659 (95% CI:0.627–0.693). The AUC values for the training and validation cohort for the 1-year OS ranged from 0.50 to 0.70, and the AUC for the rest of the training and validation cohort ranged from 0.70 to 0.90, which suggests that the model is moderately discriminating. The calibration curves of 1 year, 2 years and 3 years in the two groups are very close to the 45°reference line, suggesting that the models exhibit a good degree of calibration. The C-index, AUC and calibration curves suggest that the models have good discriminating and calibration. Conclusion The results reveal that T stage, M stage, primary tumor site, surgery, chemotherapy and radiotherapy play an important role in influencing the treatment effect and prognosis of patients. The nomogram prediction model based on the above independent risk factors shows good discriminating and calibration.