预测晚期胃癌根治性手术的疗效：一种基于生物学信息的分期方法

Background: Treatment strategies for advanced gastric cancer (GC) largely depend on the presence or absence of metastasis. However, a subset of patients with limited metastasis may still derive meaningful benefit from curative-intent surgery, challenging the traditional binary staging paradigm. Current assessments of metastasis rely on lesion volume and anatomical distribution, which oversimplify tumor biology. The degree to which patients with metastatic GC can benefit from surgery remains unclear. We therefore propose a serum-derived staging model that integrates systemic tumor-host interactions to define biological tumor stage and inform surgical decision-making in a biologically informed manner. Methods: Advanced GC was stratified by metastatic burden into locally advanced GC (LAGC), limited metastatic GC (LMGC), and widely metastatic GC (WMGC). Cohort 1 included 179 patients receiving curative-intent surgery following preoperative systemic therapy (100 LAGC, 48 LMGC, 31 WMGC). Serum samples were obtained at baseline and preoperatively. Cohort 2, an independent dataset collected during a different period, included 149 patients (109 LAGC, 25 LMGC, 15 WMGC) with baseline serum samples. Systemic inflammatory profiles were characterized via Olink platform. Baseline samples from Cohort 1 served as the training set. Feature selection was performed using ordinal logistic regression, XGBoost, and SVM-RFE, followed by Elastic Net regression for model construction. Preoperative samples from Cohort 1 were used for internal validation, and Cohort 2 for external validation. Results: Three inflammatory proteins, IL-22 RA1, HGF, and 4E-BP1, were significantly associated with metastatic burden and were incorporated into the tumor-induced perturbation score (TIPscore). Baseline TIPscore distinguished WMGC, the subgroup least likely to benefit from surgery, from LAGC and LMGC with an AUC of 0.812. Lower TIPscore was associated with improved overall survival (OS) (p = 0.049). In preoperative samples from Cohort 1, TIPscore outperformed conventional M0/M1 staging, increasing 1-year OS prediction AUC from 0.683 to 0.850 in advanced GC overall, and reaching 0.875 in the LMGC subgroup. Among LMGC patients receiving curative surgery, a lower post-treatment TIPscore predicted better OS (p = 0.034) and event-free survival (EFS) (p = 0.015). TIPscore was further validated in Cohort 2, achieving an AUC of 0.810 for distinguishing WMGC from LAGC and LMGC, and 0.711 for distinguishing WMGC from LMGC. Conclusions: TIPscore is a biologically informed staging model that reframes metastatic burden as a continuous biological spectrum rather than a categorical variable. By precisely situating individual patients along this spectrum, TIPscore provides prognostic insight to guide surgical decision-making and may be particularly valuable in ambiguous clinical contexts such as LMGC.