Abstract
OBJECTIVE: To analyze the relationship between tumor regression patterns and ypN positivity and explore their implications for postoperative nodal-risk stratification after neoadjuvant or conversion therapy in advanced gastric cancer, while generating hypotheses for future individualized lymphadenectomy research. METHODS: Tumor regression patterns were classified as centripetal, diffuse/mixed, or centrifugal. Clinical and pathological characteristics were compared using the Kruskal-Wallis and χ² tests. Using ypN positivity as the outcome, a multivariable logistic regression model was constructed. Sensitivity analyses were performed in the subgroup with ≥16 retrieved lymph nodes, after additional adjustment for ypT and Becker tumor regression grade (TRG), and in the non-pCR subgroup. Internal validation was performed using a 7:3 stratified random split and 10-fold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), 95% confidence intervals, calibration, and the Brier score. We additionally compared a baseline clinicopathological model with a combined model incorporating regression pattern to assess incremental predictive value. RESULTS: Among 195 patients, 74 (38.0%) exhibited centripetal regression, 43 (22.1%) had diffuse/mixed regression, and 78 (40.0%) demonstrated centrifugal regression. Centripetal regression was characterized by low PRI, higher LRI and CER, and a very low ypN positivity rate (5.4%), whereas centrifugal regression showed the opposite pattern and the highest ypN positivity rate (75.6%); diffuse/mixed regression showed intermediate features (all p < 0.001). Multivariable analysis identified diffuse/mixed and centrifugal regression as the strongest independent predictors of ypN positivity. The apparent full-cohort model demonstrated an AUC of 0.875 (95% CI 0.826-0.922) with good calibration and a Brier score of 0.137. These associations remained robust after additional adjustment for ypT and Becker TRG and in the non-pCR subgroup. Internal validation showed acceptable performance, with a validation AUC of 0.826 in the 7:3 split-sample analysis and a pooled AUC of 0.822 in 10-fold cross-validation. Addition of regression pattern to the baseline clinicopathological model improved discrimination and reduced prediction error. CONCLUSION: Pathological regression patterns provide effective stratification of residual lymph node metastasis after neoadjuvant or conversion therapy. Centripetal regression indicates a very low residual nodal-risk phenotype, whereas centrifugal regression is associated with a heavier nodal burden. At present, regression patterns may be most appropriately used for postoperative risk assessment and multidisciplinary stratification. Their potential role in individualized lymphadenectomy should be viewed as a future translational direction requiring validated preoperative or intraoperative surrogate markers and prospective confirmation.