Abstract
OBJECTIVE: This study evaluated the 8th edition American Joint Committee on Cancer (AJCC) staging system for esophageal squamous cell carcinoma (ESCC) and developed an improved staging framework using automated recursive partitioning analysis (autoRPA). METHODS: This retrospective study included 2773 ESCC patients treated with definitive intensity-modulated radiation therapy (IMRT) or chemo-IMRT across 8 Chinese centers (2001-2019). Kaplan‒Meier curves and log-rank tests were used to assess overall survival (OS). AutoRPA-derived stage groupings were optimized via the revised T/N criteria. The proposed staging was compared with the 8th edition AJCC staging via hazard discrimination, consistency, sample balance, and predictive accuracy. RESULTS: The 3-year, and 5-year rates for the entire cohort were 43.5%, and 34.0%, respectively. The AJCC T4a/T4b stages exhibited overlapping OS curves, prompting their consolidation into a single T4 stage. While the AJCC N2 and N3 stages showed overlapping OS curves, supraclavicular lymph node (SLN) metastasis independently predicted worse OS than N2, with outcomes similar to those of N3. Location-based SLN classification further refined nodal staging, with cervical esophageal-SLN metastasis classified as N1, upper thoracic-SLN metastasis as N2, and middle or lower thoracic-SLN metastasis as N3, yielding distinct OS stratification. The autoRPA-derived staging outperformed the 8th edition AJCC staging in hazard consistency, sample balance, and predictive accuracy, with RPA-I exhibiting distinctly sharper OS curves than other stages. CONCLUSION: Combining T4a/T4b and SLN subclassification enhanced prognostic precision in ESCC, with the autoRPA staging demonstrating superior hazard consistency, sample balance, and predictive accuracy compared to the 8th edition AJCC staging, thereby guiding therapeutic strategies.