A clinically applicable toolkit for cervical gastric-type adenocarcinoma: Diagnostic model, predictive survival nomograms, and risk stratification.

Abstract

e17509 Background: Cervical gastric-type adenocarcinoma (GAS), an aggressive non-HPV-associated malignancy, is often misdiagnosed and correlated with unsatisfactory treatment efficacy and poor prognosis. The lack of diagnostic and prognostic tools poses challenges to clinical management. We aimed to develop a comprehensive toolkit for accurate preoperative diagnosis, individualized prediction of recurrence and survival as well as postoperative risk stratification for GAS patients. Methods: In this retrospective, multicenter study, 160 GAS and 194 non-GAS patients from four tertiary centers in Shanghai, China were included. Clinicopathological data, including symptoms, HPV status, imaging findings, tumor markers, postoperative pathological features, adjuvant treatment and survival outcomes, were collected. A multivariate logistic regression model was developed for preoperative diagnosis of GAS. Prognostic nomograms for recurrence-free survival (RFS) and overall survival (OS) were established using least absolute shrinkage and selection operator (LASSO) Cox regression and multivariate Cox analysis. Besides, recursive partitioning analysis (RPA) was applied for simplified risk stratification for RFS of GAS. Model performance was evaluated using area under the curve (AUC) and concordance index (C-index). Results: The preoperative diagnostic model, incorporating symptoms, HPV status, imaging features, and CA199 levels, achieved an AUC of 0.93 in the training cohort and 0.84 in the validation cohort. Prognostic nomograms for RFS and OS were developed manifesting robust predictive accuracy, with 2-year, 5-year C-indexes ranging from 0.829 to 0.844 in the training set and 0.624 to 0.728 upon external validation. Notably, the RFS model incorporated ovarian involvement, maximum tumor diameter, vaginal involvement, positive vaginal resection margin, parametrial metastasis, and pelvic lymph node metastasis, whereas the OS model included ovarian involvement, positive vaginal resection margin, parametrial metastasis, and pelvic lymph node metastasis. Moreover, the RPA-based risk stratification model classified GAS patients into three distinct groups based on parametrial involvement and pelvic lymph node metastasis, with significant differences in 5-year RFS (Low-risk: 78.86%, 95% CI: 64.84%-95.90%; Intermediate-risk: 38.90%, 95% CI: 21.30%-71.00%; High-risk: 24.79%, 95% CI: 13.79%-44.57%; p<0.001), potentially guiding adjuvant therapeutic strategies. Conclusions: This study presents a clinically applicable toolkit for GAS, comprising an accurate diagnostic model for preoperative identification, survival nomograms for individualized prognosis prediction, and a simplified risk stratification system for postoperative risk assessment, offering a novel framework to optimize the clinical management of GAS.