Integrating nonlinear modeling with pT stage and nodal variables: A multi-center Con-pTN model for prognosis prediction in gastric cancer (pT2–4b)

Lymph node metastasis critically determines prognosis in gastric cancer (GC), yet conventional staging inadequately captures interactions between tumor invasion depth and nodal burden. In this retrospective multi-center study, we analyzed 2,370 patients (pT2-4b) undergoing curative gastrectomy and 724 patients with SEER to develop a contour-like pTN (Con-pTN) model using a Gaussian process-augmented Cox framework that integrates pT stage, retrieved lymph nodes, and positive lymph nodes as continuous variables. Con-pTN showed robust discrimination, with area under the curves (AUCs) of 0.797 (training), 0.804 (internal validation), 0.748 (external validation cohort 1), and 0.813 (external validation cohort 2). Calibration and decision curve analyses demonstrated good agreement with observed outcomes and favorable clinical utility. Compared with pTNM, rN, and LODDS, Con-pTN provided significantly improved prognostic stratification, particularly among high-risk patients. These findings support Con-pTN as a biologically informed, clinically applicable approach for postoperative risk assessment in GC.