Postoperative weight loss trajectories and early risk stratification in gastric cancer patients receiving adjuvant chemotherapy after gastrectomy.

PURPOSE: To identify postoperative weight-loss trajectories during the first six months after gastrectomy and develop an interpretable machine-learning model to predict rapid decline before adjuvant chemotherapy. METHODS: This single-centre retrospective cohort included 546 patients undergoing radical gastrectomy followed by adjuvant chemotherapy. Patients were divided into a training cohort (n = 435) and an independent test cohort (n = 111). Group-based trajectory modelling was fitted exclusively in the training cohort, with fixed parameters applied to the test cohort. Six machine-learning models used predictors available by admission for the first chemotherapy cycle. Discrimination, calibration, and decision curve analysis assessed performance, and SHapley Additive exPlanations interpreted the final model. RESULTS: Three trajectories were identified: stable (n = 177, 32.4%), moderate decline (n = 279, 51.1%), and rapid decline (n = 90, 16.5%). The rapid-decline trajectory reached approximately 22% weight loss by postoperative month 6. CatBoost showed the best overall performance in the test cohort, with an area under the receiver operating characteristic curve of 0.808 (95% confidence interval, 0.695-0.902), a Brier score of 0.112, and a calibration slope of 0.954. Decision curve analysis indicated greater net benefit than the treat-all and treat-none strategies across threshold probabilities of approximately 5%-30%. Preoperative body mass index, histological subtype, and gastrointestinal reconstruction were the leading predictors, while psychosocial variables provided complementary information. CONCLUSIONS: Postoperative weight loss followed distinct trajectories. CatBoost may identify patients at risk of rapid, sustained weight loss when adjuvant chemotherapy begins. External multicentre validation is required before clinical implementation.