Very Early Recurrence After Curative-Intent Surgery for Gastric Adenocarcinoma

Background Recurrence after curative-intent surgery can occur in more than 50% of gastric cancer (GC) patients. We sought to identify predictors of very early recurrence (VER) among GC patients who underwent curative-intent surgery. Methods A multi-institutional database of GC patients undergoing curative-intent surgery between 2000 and 2020 at 8 major institutions was queried. VER was defined as local or distant tumor recurrence within 6 months from surgery. Univariable Cox proportional hazard models were used to evaluate the predictive value of clinical-pathological features on VER. A regularized Cox regression model was employed to build a predictive model of VER and recurrence within 12 months. The discriminant ability of the Cox regularized models was evaluated by reporting a ROC curve together with the calibration plot, considering 200 runs. Results Among 1133 patients, 65 (16.0%) patients experienced a VER. Preoperative symptoms (HR 1.198), comorbidities (HR 1.289), tumor grade (HR 1.043), LNR (HR 4.339) and T stage (HR 1.639) were associated with an increased likelihood of VER. Model performance was very good at predicting VER at 6 months (AUC of 0.722) and 12 months (AUC 0.733). Two nomograms to predict 6-month and 12-month VER were built based on the predictive model. A higher nomogram score was associated with worse prognosis. There was good prediction between observed and estimated VER with minimal evidence of overfitting and good performance on internal bootstrapping validation. Conclusion One in 6 patients experienced VER following curative-intent surgery for GC. Nomograms to predict risk of VER performed well on internal validation, and stratified patients into distinct prognostic groups relative to 6- and 12-months recurrence.

Automated summary

This study aimed to identify predictors of very early recurrence (VER) in gastric cancer patients who underwent curative-intent surgery. The study found that preoperative symptoms, comorbidities, tumor grade, lymph node ratio, and T stage were associated with an increased likelihood of VER. The predictive model performed well in predicting VER and two nomograms were built based on the model.