Integrating tumor mutation profiles, preoperative circulating tumor DNA (ctDNA), and clinical features to a predicting model for early peritoneal recurrence in stage II-III gastric cancer following curative gastrectomy.

Abstract

3056 Background: Peritoneal recurrence is the common failure pattern after curative gastrectomy and is most consequential within the first 2 years. Early risk identification can alter treatment strategy. Conventional clinicopathologic models are inadequate for individualized early detection. Tumor-informed circulating tumor DNA (ctDNA) enables detection of minimal residual disease (MRD) and may help refine risk stratification. Methods: We conducted a prospective cohort study of 134 patients with stage II–III gastric adenocarcinoma undergoing curative-intent gastrectomy (MRD cohort). Clinicopathologic variables, tumor mutation profiles, and ctDNA metrics were integrated to train and internally validate machine-learning models using logistic regression, random forest, and XGBoost for the primary endpoint of peritoneal recurrence within 12 months and 24 months after surgery. External validation was performed using the Yang et al. dataset. Results: Among 134 MRD patients with complete molecular and clinical data, 19 (14.2%) and 35 (26.1%) developed peritoneal recurrence within 12 months and 24 months, respectively. ctDNA positivity and higher max variant allele frequency (max VAF) were strong predictors of early peritoneal relapse, with additional contributions from pathologic stage and preoperative CEA, PI3K–AKT or EMT pathway genes. Across algorithms, ctDNA-integrated models consistently outperformed clinical-only baselines, with XGBoost showing the best overall discrimination on cross-validation. Findings generalized on external validation with the Yang et al. dataset, supporting model transportability. Conclusions: Integrating tumor-informed ctDNA with clinical and genomic features enables actionable prediction of early peritoneal recurrence after curative gastrectomy. This tool can trigger strategy modification of either stricter follow-up, adjuvant therapy intensification, or consideration of neoadjuvant approaches, in selected high-risk patients. Multicenter validation and prospective utility studies are warranted. Clinical trial information: NCT05029869 .