The medical records of 1,525 patients with biopsy-confirmed gastric cancer at the Department of General Surgery of Nanfang Hospital (Guangzhou, China) between June 2004 and June 2014 were retrospectively reviewed from a prospectively collected database (5). Of the 1,525 patients, 19 were excluded, including 10 patients with carcinoma in situ, 5 patients with synchronous malignancies (1 with prostate cancer, 1 with Hodgkin's lymphoma, 1 with colon cancer, 1 with bladder cancer and 1 with nasopharyngeal cancer), 3 patients who died of postoperative complications, and 1 patient aged <18 years. Finally, 1,506 eligible patients were included in the subsequent analysis. The study was approved by the Ethics Committee of Nanfang Hospital of Southern Medical University. Written informed consent was obtained from all patients prior to entering their information into the database.

Surgical resection with a curative intent was performed for stage I–III patients and attempted for those with peritoneal metastasis, if allowed by preoperative assessment and intraoperative observation. Personalized adjuvant chemotherapy regimens were prescribed for patients, mainly using capecitabine, oxaliplatin, leucovorin and irinotecan. For patients with T3-T4 lesions or lymph node metastases, adjuvant chemotherapy was generally recommended. In this analyzed cohort, 656 (62.1%) received adjuvant chemotherapy, whereas no patients received radiotherapy or chemotherapy as neoadjuvant treatment.

The patient demographic and clinicopathological data were analyzed, including gender (male or female), age (<60 or ≥60 years), history of abdominal surgery (yes or no), classification of comorbidities (0, 1 or ≥2), Eastern Cooperative Oncology Group performance status (0, 1 or ≥2), primary tumor size (≤5 or >5 cm), level of serum carcinoembryonic antigen (CEA) and cancer antigen (CA) 19-9 (normal or elevated), type of gastrectomy (subtotal or total), resection extent (radical or palliative), differentiation degree (high/moderate or poor/signet-ring cell carcinoma) and positive resection margin (yes or no). Tumor classification was determined according to the 7th and 8th editions of the AJCC TNM staging system (referred to as 7th TNM and 8th TNM, respectively). For those patients with distant metastasis who were not considered eligible for surgical treatment, the pathological stage and T and N status were unavailable.

The aim of the present study was to evaluate whether the 8th TNM classification was superior to the 7th TNM classification in the prognostic prediction of patient OS. The OS time was defined as the duration from the date of surgery (or date of first diagnosis for those patients who did not undergo surgery) to the date of death or the last follow-up visit. The 5-year OS rate and median OS time were calculated, and survival curves were estimated using the Kaplan-Meier method. As mentioned above, the main modification in the 8th TNM staging system was the incorporation of N3 subdivisions (N3a and N3b) into the final pathological staging, which contributed to the shift of the final TNM classification. The discriminatory ability and monotonicity of gradient assessments were measured with the linear trend χ² test of survival curves. Among the four aforementioned groups with migration, three groups (T3N3bM0, T2N3bM0 and T1N3bM0) included only a small number of patients, and the survival outcomes could not be statistically assessed. Since another group (T4aN3aM0) was a subgroup of the T4 stage, it was integrated with the other three T4 subgroups (T4bN0M0, T4aN2M0 and T4bN2M0) for analysis. The presence of the stage migration phenomenon mainly occurred in stage III, namely T4bN0M0 and T4aN2M0 of stage IIIB in the 7th TNM migrating to stage IIIA in the 8th TNM, and T4aN3aM0 and T4bN2M0 of stage IIIC in the 7th TNM migrating to stage IIIB in the 8th TNM. To investigate the suitability of these stage migrations, survival comparison was performed between patients with stage migration and those remaining in the same stage category according to both editions. The log-rank test and hazard ratio (HR) were used during analysis of univariate risk factors. Variables associated with OS were included in a multivariable Cox proportional regression model to identify independent risk factors of OS in the two staging systems. Variables highly related to others were excluded from the final model. Finally, two Cox regression models were built based on the 7th and 8th TNM to compare their prognostic prediction abilities. The variables in the two models were identical, except for TNM classification; in other words, the model based on the 8th TNM was constructed by replacing the 7th TNM stage by the 8th TNM stage. The statistics, −2log likelihood of the Cox regression model, were calculated to compare the predictive efficacy of the two models. A smaller value indicated a better model for predicting outcome.

For the majority of the patients, the postoperative follow-up was assessed at 3-month intervals for the first 2 years, 6-month intervals for the next 3 years, and annually thereafter until the endpoints were reached. Physical examination, serum tumor biomarker levels, chest X-ray, abdominal ultrasonography, gastroendoscopy and positron emission tomography/computed tomography were selected for accurate assessment at each follow-up. The last follow-up was performed in July 2016.

All statistical analyses were performed using SPSS version 17.0 for Windows (SPSS Inc., Chicago, IL, USA). A P-value of <0.05 (two-tailed) was considered to indicate statistically significant differences.

Of the 1,506 patients included, 1,018 (67.6%) were male and 488 (32.4%) were female. The mean age of the entire cohort was 55.6 years (range, 19–90 years). The cohort was followed up for a mean duration of 44.1 months (range, 1–145 months). The demographic characteristics, pathological characteristics and clinical outcomes of the patients are summarized in Table I. The total number of dissected lymph nodes was 32,439, with a mean ± standard deviation of 23.8±19.3. The mean number of metastatic lymph nodes was 5.2±7.7 (median, 2; range, 0–65) in all patients and 8.0±8.3 (median, 5; range, 1–65) in patients with lymph node involvement.

The main modification in the new TNM staging system was subdivision of the N3 category into N3a and N3b, and the resultant stage migration in the final TNM classification. The necessity of these modifications was unknown. In the present study, the OS curves based on the 7th TNM and 8th TNM were separately constructed and compared (Fig. 1). A log-rank test for trend revealed good discriminatory abilities in both N classifications (both P<0.001), but the 8th TNM classification exhibited a better monotonicity of gradient compared with the 7th TNM classification, as evidenced by a higher χ² value in the former classification (χ² for log-rank trend test, 311.783 vs. 297.588).

A total of 1,484 patients were included for analysis of TNM staging, and 22 patients with distant metastasis who did not undergo surgical resection were excluded due to unavailable data on T and N status. The distribution of patients according to the 7th and 8th editions of the AJCC TNM staging system is summarized in Table II. Among 1,484 patients, 1,145 (77.2%) remained in the same stage category and 339 (22.8%) migrated to a different stage, including 325 (21.9%) patients who migrated to a lower tier and 14 (0.9%) who migrated to a higher tier, compared with the 7th TNM. Given the small number of patients that migrated to a higher tier, including 2 patients in T1N3bM0 migrating from stage IIB to IIIB, 4 patients in T2N3bM0 migrating from stage IIIA to IIIB, and 8 patients in T3N3bM0 migrating from stage IIIB to IIIC, survival comparisons could not be made for these patients. Interestingly, all patients with stage migration to a lower tier were in stage III, including 177 (54.5%) patients migrating from stage IIIB to IIIA, and 148 (45.5%) patients migrating from stage IIIC to IIIB. To evaluate the suitability of these migrations, the OS of patients with stage migration was further compared to those remaining in the same stage category (Table III). For patients migrating from stage IIIB to IIIA, the survival curves had no significant difference from those that remained in stage IIIB or IIIA (P=0.342). However, patients with stage migration from IIIB to IIIA yielded a median OS time and 5-year OS rate closer to those remaining in stage IIIA (discrepancy of 8 months in median OS and 4.5% in 5-year OS rate), compared with those remaining in stage IIIB (discrepancy of 21 months in median OS and 14.1% in 5-year OS rate). Similarly, for patients who migrated from stage IIIC to IIIB, the median OS and 5-year OS rate were closer to those remaining in stage IIIB, with a discrepancy of 4 months in median OS and 2.3% in 5-year OS rate, exhibiting a significant difference from those in stage IIIC (discrepancy of 13 months in median OS and 14.3% in 5-year OS rate; overall P=0.014). The discriminatory ability in stage III was improved following stage migrations. The survival curves for stage IIIA and IIIB based on the 7th TNM were close (P=0.260), with improved stratification in the 8th edition (P<0.001). The survival curves of patients based on the 7th and 8th editions of the AJCC TNM staging system are shown in Fig. 2.

Univariate analysis was performed to evaluate the risk factors of OS (Table I). Variables including age, classification of comorbidities, primary tumor size, serum CEA, serum CA 19-9, type of gastrectomy, resection extent, differentiation degree, positive resection margin, pathological T and N stage, 7th TNM stage and 8th TNM stage were identified as prognostic factors of OS. After elimination of the variables highly related to others, two Cox proportional regression models were constructed, and included age, classification of comorbidities, primary tumor size, serum CEA, serum CA 19-9, type of gastrectomy, tumor differentiation, positive resection margin and TNM stage in the two staging systems to predict the prognosis of gastric cancer patients (Table IV). In both models, an elevated level of serum CA 19-9, total gastrectomy, positive resection margin and high TNM stage were independent risk factors for predicting an unfavorable OS. A smaller value of −2log likelihood was calculated in the model for the 8th edition of the staging system (7th edition 4,738.859 vs. 8th edition 4,736.683), which indicated a better predictive capability compared with the 7th edition.