Validation of the 7th AJCC/UICC staging system for gastric cancer and a proposal for a new TNM system based on a prognostic score: a retrospective multicenter study

Gastric cancer is one of the leading causes of cancer-related deaths in the world [1]. Similar to other malignancies, we can predict the survival of gastric cancer patients and make plans for multidisciplinary treatments on the basis of the stage of disease; thus, the staging system is crucial in the management of patients with gastric cancer. We adopted the 7th American Joint Committee on Cancer/Union for International Cancer Control (AJCC/UICC) staging system starting in 2010, and it contained major changes compared with the 6th AJCC/UICC TNM staging system for gastric cancer. These changes included reclassification of T2a and T2b to T2 and T3, respectively; division of the previous N1 stage (up to 6 metastatic regional lymph nodes) into N1 (1 or 2 metastatic regional lymph nodes) and N2 (3 to 6 metastatic regional lymph nodes); and reclassification of the previous N2 (7 to 15 metastatic regional lymph nodes) and N3 stages (more than 15 metastatic regional lymph nodes) to N3a (7 to 15 metastatic regional lymph nodes) and N3b (more than 15 metastatic regional lymph nodes) [2, 3, 4]. Although the 7th AJCC/UICC staging system for gastric cancer has been reported to be more accurate or valid by some authors [3, 5, 6, 7], the system is not without controversy. One of the most common controversies is classification of the N category, such that the discriminative power of the N category in the 7th UICC/AJCC staging system has been decreased relative to that of the 6th AJCC/UICC staging system [8, 9]. In addition, different prognostic significances of the subgroups of N3 (N3a and N3b) was not reflected in the final staging [10, 11]. The metastatic lymph node ratio was suggested to be a complementary measure by many authors [12, 13, 14, 15], and others noted that there was heterogeneity in the survival of subgroups in a certain stage of the 7th AJCC/UICC staging system [16]. These controversies have highlighted the need for a new staging system [17]. In this study, we validated the 7th AJCC/UICC staging system with data from a large number of patients at 4 university hospitals and propose a new staging system based on basic concepts of the TNM staging system that reflects the different prognostic significance of each T and N category using a prognostic score (P-score).

Included patients were those who had undergone surgery for gastric cancer from January 2000 to December 2007 at 4 university hospitals in Daegu Metropolitan city in Korea. Patients who met the following criteria were excluded from the analysis: (1) nonresective surgery, (2) gastric cancer located in the cardia, (3) less than 16 total retrieved lymph nodes, (4) recurrent gastric cancer, (5) coexisting malignancy in other organs, (6) history of preoperative chemotherapy, and (7) death within 90 days of surgery that might have been caused by complications of the surgery. After excluding 274 out of 6,241 patients, data from 5,967 patients were analyzed retrospectively. The mean follow up time was 66.2 months including early deaths. Roux-en-Y esophagojejunostomy was performed after total gastrectomy except for one case of jejunal interposition. Gastroduodenostomy (Billroth I), gastrojejunostomy (Billroth II) or Roux-en-Y gastrojejunostomy was performed after distal gastrectomy. Gastrogastrostomy was performed after pylorus preserving gastrectomy. The standard extent of lymph node dissection was D2 and D1+ in patients with early gastric cancer. Follow-up data were collected on the basis of medical records and telephone interviews. Survival duration was calculated from the day of surgery to the day of the last follow-up at outpatient department or by telephone, or the day of the patient's death.

The process for generating the new staging system was as follows: (1) estimation of the hazard ratio (HR) of each T and N category; (2) transformation of HR into a weighting; (3) summation of the weightings of the T and N categories (P-score); and (4) homogeneity testing and grouping as well as testing for the appropriateness of the 5-year survival gap between adjacent stages.

The estimated survival probability of the 7th edition of the TNM staging system and respective combinations of T and N categories were estimated using the Kaplan-Meier method and were compared by the log-rank test. The HR within the Cox's proportional hazards model was used to assess the weighting of prognostic significance of the T and N categories. At first, the variables of age and sex were included to evaluate the interaction effect with the T and N categories, but the interactions were not significant. Thus, only data from the T and N categories were included in the model, and the HR was estimated. These HRs were simplified to the nearest number ending in 0.5 to meet both convenience and accuracy. We named this number the weighting. All the weightings of the T and N categories were confirmed to be within the 95% confidential intervals of their respective HRs and summated to make P-scores. The Cox's proportional hazards model was applied using P-scores to estimate the regression coefficient. This regression coefficient was applied to assess the homogeneity and grouping of P-scores using an estimated covariance matrix and Wald test. With sufficient numbers of patients in each stage, the Z-test was used to examine the discrimination of the 5-year survival rates between adjacent stages. The log-rank score within the Cox's proportional hazards model was used to demonstrate the discriminative ability of the new staging system in comparison with the 7th AJCC/UICC TNM classification. Higher log-rank scores were regarded to have better discriminative power in terms of survival. A P-value of less than 0.05 was considered statistically significant. No adjustments were made.

The purpose of a cancer staging system is stated in the AJCC cancer staging manual. The extent or stage of cancer at the time of diagnosis is a key factor that defines prognosis and is a critical element in determining an appropriate treatment based on the experience and outcomes of groups of prior patients at a similar stage. In addition, accurate staging is necessary to evaluate the results of treatments and clinical trials, to facilitate the exchange and comparison of information among treatment centers, and to serve as a basis for clinical and translational cancer research [2]. Until recently, many appraisals of the 7th AJCC/UICC staging system for gastric cancer and proposals for new staging systems have been reported [18, 19, 20]. In our proposed staging system, the increments in weightings varied from 0.5 to 2.5, and this feature might disclose the different prognostic significance of each component of the T and N categories. More weighting was placed on categories with more prognostic significance, depending on their HR. Some authors have reported that the subgroups of N3 (N3a and N3b) should be included in the final stage separately because the prognosis of the patients with N3a and N3b differs significantly [10, 21]. In our proposed staging system, the weightings of N3a and N3b were 3.5 and 6, respectively. This difference in the prognosis was reflected in the proposed staging system without modifications. Some authors insisted that heterogeneity existed within certain groups in the 7th AJCC/UICC staging system [16]. In the proposed staging system, the 5-year survival rate of T1N3aM0 was 66.0% and 22% for T1N3bM0. T1N3a remained in stage IIB, and T1N3bM0 moved to stage IIIC. According to prognostic significance, T2N2M0 moved from IIB to stage IIA, T3N2M0 moved from stage IIIA to stage IIB and T3N3bM0 moved from stage IIIB to stage IIIC. Some authors reported poor discrimination of survival between stage IB and IIA and between IIB and IIIA in the 7th AJCC/UICC staging system [4, 8, 19, 20, 22]. In the present study, poor discrimination of survival was noted between stage IIB and IIIA with intersection of the survival curves by the 7th AJCC/UICC staging system. However, in the staging system proposed here, the survival curves did not intersect, and a significant survival difference was observed. Dikken et al. [23] reported increased complexity without prognostic significance in the 7th AJCC/UICC staging system. Our proposed staging system is composed of seven stages that fall in the middle of the 6th and 7th AJCC/UICC staging system with regard to complexity. Warneke et al. [19] stressed that there was no consideration for different biologic behavior between the primary tumor and lymph node metastasis. The 7th AJCC/UICC staging system reflects a mathematical model where the addition of the values of the T and N categories equals the same sum total in each subgroup. This feature is not an adequate categorization of the tumor's biologic potential or prognosis. In the proposed system, we convert the HRs of each T and N category to weightings, which reflect the prognosis by providing more weight to categories with worse biological behavior. Then, the weightings of the T and N categories are added to make P-scores. Thus, the different biological behavior between the status of the primary tumor and lymph node metastasis is simultaneously considered. Several benchmarks for comparing the performance of two staging systems have been suggested [23]. First, there should be homogeneity within stage groups. Second, there should be discrimination between stage groups; patients in different stage groups should have larger differences in survival. Third, the staging system should have good predictive accuracy. Fourth, the staging system should be as simple and intuitive as possible in clinical practice. Our proposed system, for the most part, appears to satisfy these four benchmarks. As for discrimination, in this study, the survival difference between stage IA and IB in the 7th AJCC/UICC system was small and unified to achieve relatively good discrimination of survival rates between neighboring stages. This unification of stage IA and IB was statistically verified to be rational by the result of Z-test presented in Table 5. Conversion of the T and N categories to weightings is simple and easy to perform. The staging process is easier and more scientific compared with the 7th AJCC/UICC staging system because the stage is determined by the P-scores.

Furthermore, this study is the first to propose a staging system generated on the basis of prognostic outcome of HR of respective T and N categories to reflect different prognostic impacts of respective T and N categories overcoming the limitations of anatomical and mathematical model.

In conclusion, the newly proposed staging system preserves the basic concepts of TNM classification, reflecting different prognostic significance for the T and N categories using a P-score, dividing N3 into N3a and N3b in stage grouping, and unifying stages IA and IB into a single stage I. This system demonstrates homogeneity of each stage group with respect to survival and a distinct difference in the survival rates between stages. Therefore, the TNM system for gastric cancer based on this P-score appears to be more scientifically accurate than the 7th AJCC/UICC staging system for gastric cancer.