A machine learning model predicting candidates for surgical treatment modality in patients with distant metastatic esophageal adenocarcinoma: A propensity score-matched analysis

Abstract

ObjectiveTo explore the role of surgical treatment modality on prognosis of metastatic esophageal adenocarcinoma (mEAC), as well as to construct a machine learning model to predict suitable candidates.MethodAll mEAC patients pathologically diagnosed between January 2010 and December 2018 were extracted from the Surveillance, Epidemiology, and End Results (SEER) database. A 1:4 propensity score-matched analysis and a multivariate Cox analysis were performed to verify the prognostic value of surgical treatment modality. To identify suitable candidates, a machine learning model, classification and regression tree (CART), was constructed, and its predictive performance was evaluated by the area under receiver operating characteristic curve (AUC).ResultsOf 4520 mEAC patients, 2901 (64.2%) were aged over 60 years and 4012 (88.8%) were males. There were 411 (9.1%) patients receiving surgical treatment modality. In the propensity score-matched analysis, surgical treatment modality was significantly associated with a decreased risk of death (HR: 0.47, 95% CI: 0.40-0.55); surgical patients had almost twice as much median survival time (MST) as those without resection (MST with 95% CI: 23 [17-27] months vs. 11 [11-12] months, P <0.0001). The similar association was also observed in the multivariate Cox analysis (HR: 0.47, 95% CI: 0.41-0.53). Then, a CART was constructed to identify suitable candidates for surgical treatment modality, with a relatively good discrimination ability (AUC with 95% CI: 0.710 [0.648-0.771]).ConclusionSurgical treatment modality may be a promising strategy to prolong survival of mEAC patients. The CART in our study could serve as a useful tool to predict suitable candidates for surgical treatment modality. Further creditable studies are warranted to confirm our findings.