Development and validation of a new staging system for node‐negative gastric cancer based on recursive partitioning analysis: An international multi‐institutional study

BACKGROUND: Whether the tumor‐node‐metastasis (TNM) staging system is appropriate for patients with node‐negative gastric cancer (GC) is still inconclusive. The modified staging system developed by recursive partitioning analysis (RPA) showed good prognostic performance in a variety of cancers. The application of RPA has not been reported in the prognostic prediction of GC. METHODS: Node‐negative GC patients who underwent radical resection at Fujian Medical University Union Hospital (n = 862) and Sun Yat‐sen University Cancer Center (n = 311) with at least 5 years of follow‐up were selected as the training set. RPA was used to develop a modified staging system. Patients from the Surveillance, Epidemiology, and End Results database (n = 1415) were selected as the validation set. RESULTS: The 5‐year overall survival (OS) rates of patients with 8th AJCC‐TNM stage IA‐IIIA in the training set were IA 95.2%, IB 87.1%, IIA 78.3%, IIB 75.8%, and IIIA 72.6%. Multivariate analysis (MVA) showed that larger tumor size, elder age, and deeper depth of invasion were independent predictors for OS in patients with node‐negative GC (all P < 0.05). Patients were reclassified into RPA I, RPA II, RPA III, and RPA IV stages based on RPA; the 5‐year OS rates were 96.1%, 87.2%, 81.0%, and 64.3%, respectively, with significant difference (P < 0.05). Two‐step MVA showed that the RPA staging system was an independent predictor of OS (P < 0.05). Compared with the 8th AJCC‐TNM staging system, the RPA staging system had a smaller AIC value (2544.9 vs 2576.2), higher χ(2) score (104.2 vs 69.6) and higher Harrell's C‐index (0.697 vs 0.669, P = 0.007). The similar results were found in the validation set. CONCLUSIONS: A new prognostic predictive system based on RPA was successfully developed and validated, which may be suggested for staging node‐negative GC in future.