Development of a predictive model for extragastric recurrence after curative resection for early gastric cancer

Background Stratification of patients who undergo curative resection for early gastric cancer (EGC) is warranted due to the heterogeneity in the risk of developing extragastric recurrence (EGR). Therefore, we aimed to stratify the need for postoperative surveillance for EGR detection in patients with EGC by developing a model for predicting EGR-free survival. Methods This retrospective cohort study included patients who underwent postoperative surveillance after curative resection of EGC (n = 4149). Cox proportional hazard models were used to identify predictors to build a model for predicting EGR-free survival. Bootstrap-corrected c-index and calibration plots were used for internal and external (n = 2148) validations. Results A risk-scoring system was constructed using variables significantly associated with EGR-free survival: pathologic T stage (pT1b[sm1], hazard ratio [HR] 4.928; pT1b[sm2], HR 5.235; pT1b[sm3], HR 7.748) and N stage (pN1, HR 4.056; pN2, HR 9.075; pN3, HR 30.659). Patients were dichotomized into a very-low-risk group or a low-or-greater-risk group. The 5-year EGR-free survival rates differed between the two groups (99.9 vs. 97.3%). The discriminative performance of the model was 0.851 (Uno's c-index) and 0.751 in the internal and external cohorts, respectively. The calibration slope was 0.916 and 1.131 in the internal and external cohorts, respectively. Conclusions Our model for predicting EGR-free survival based on the pathologic T and N stages may be useful for stratifying patients who have undergone curative surgery for EGC. The results suggest that patients in the very-low-risk group may be spared from postoperative surveillance considering their extremely high EGR-free survival rate.