A population-based predictive model predicting candidate for primary tumor surgery in patients with metastatic esophageal cancer

Background: The survival benefit of primary tumor surgery for metastatic esophageal cancer (mEC) patients has been observed, but methods for discriminating which individual patients would benefit from surgery have been poorly defined. Herein, a predictive model was developed to test the hypothesis that only certain metastatic patients would gain a survival benefit from primary tumor surgery. Methods: Clinical data for patients with mEC were extracted from the Surveillance, Epidemiology and End Results (SEER) database [2004-2016] and then divided into surgery and no-surgery groups according to whether surgery was performed on the primary tumor. Propensity-score-matching (PSM) was performed to balance the confounding factors. We hypothesized that the patients who had undergone surgery and lived longer than the median cancer-specific-survival (CSS) of the no-surgery group could benefit from surgery. We constructed a nomogram to predict surgery benefit potential based on multivariable logistic-regression analysis using preoperative factors. The predictive performance of the nomogram was evaluated by the area under the receiver operating characteristic (AUC) and calibration curves. The clinical application value of the nomogram was estimated with decision curve analysis (DCA). Results: A total of 5,250 eligible patients with mEC were identified, and 9.4% [492] received primary tumor surgery. After PSM, CSS for the surgery group was significantly longer [median: 19 vs. 9 months; hazard ratio (HR) 0.52, P 0.001] compared with the no-surgery group. Among the surgery group, 69.3% [327] survived 9 months (surgery-beneficial group). The prediction nomogram showed good discrimination both in training and validation sets (AUC: 0.72 and 0.70, respectively), and the calibration curves indicated a good consistency. DCA demonstrated that the nomogram was clinically useful. According to this nomogram, surgery patients were classified into two groups: no-benefit-candidate and benefit-candidate. The benefit candidate group was associated with longer survival than the no-benefit-candidate group (median CSS: 19 vs. 6.5 months, P<0.001). Additionally, there was no difference in survival between the no-benefit-candidate and no-surgery groups (median CSS: 6.5 vs. 9 months, P=0.070). Conclusions: A predictive model was created for the selection of candidates for surgical treatment among mEC patients. This predictive model might be used to select patients who may benefit from primary tumor surgery.