A novel grade-lymph node ratio model predicts the prognosis of the advanced gastric cancer patients after neoadjuvant radiotherapy

Although local advanced gastric cancer (AGC) could benefit from neoadjuvant radiotherapy (NRT), there are few studies evaluating patients survival after NRT. In current study, we aimed to investigate the value of prognostic factors in AGC patients after NRT and to evaluate whether post-therapy pathological characteristics were predictive factors in these patients. We retrospectively analyzed AGC patients who underwent NRT from Surveillance, Epidemiology, and End Results (SEER) Database. The patients clinical and post-therapy pathological characteristics were analyzed. The best cutoff points for continuous variables were identified by X-tile. The discrimination of risk factors were compared by receiver operating characteristic (ROC) curve. As a result, 1,429 AGC patients were included into this study. In the multivariate analysis, the lymph nodes status and histology grade were significant risk factors for DSS (disease special survival). Then, we propose a novel Grade-lymph node Ratio (G-R) staging system for the AGC patients survival prognosis. Clearly, the new G-R staging system has a more-accurate 3-year and 5-year DSS prediction than the AJCC staging system (p = 0.001, 0.007, respectively). In conclusions, the current large, general population-based study demonstrated that the G-R staging system resulting in more-accurate DSS prediction. It could be regarded as a reliable classification for AGC patients after NRT in future.


INTRODUCTION
Gastric cancer is the fourth most common malignancy and second leading cause of cancer-related death worldwide. [1] In 2014, approximately 22,220 new cases were diagnosed and 10,990 deaths attributed to gastric cancer in United States. Although the incidence has declined recently, the 5-year survival was less than 30%. [2,3] Completed resection with lymph node dissection is the only potential curative treatment for resectable cancer. [4] However, the cancer symptoms usually were not obvious in early stage, most patients were diagnosed with advanced stage in most countries. Recently, the MAGIC trials demonstrated that the neoadjuvant chemotherapy improved survival for AGC patients. [5] Meanwhile, several phase I/II trails also demonstrated a survival benefit for AGC patients after NRT. [6][7][8] Indeed, the neoadjuvant therapy can achieve a clinical downstaging before resection and increase the possibility of R0 resection for AGC patients. [9][10][11] As a result, the neoadjuvant therapy were widely used for potentially resectable AGC before surgery.
Based on those promising results, the neoadjuvant Clinical Research Paper www.impactjournals.com/oncotarget therapy may be a potential standard treatments for AGC patients. However, the prognostic value of post-therapy pathologic characteristic were still unclear, especially in patients underwent NRT. [10,[12][13][14][15] In current study, we aimed to investigate the value of prognostic factors in AGC patients and to evaluate whether post-therapy pathological characteristics were predictive factors of survival in these patients.

Patients and demographics
A total of 1,429 patients between January 1998 and November 2013 in the SEER database who met all the inclusion criteria were analyzed in current study. The patients' characteristics were listed in Table 1. Overall, there were 1,206 male patients and 223 female patients, and the median age was 60.8 years old. There were also 60 (4.2%) patients who received adjuvant radiotherapy after curative resection. The mean number of positive lymph node was 1.8±3.3, and the mean total examined number of lymph node was 14.3±9.9. With a median follow-up of 31.9 months, there were 760 (53.2%) patients died before the analysis of the present study, and 102 (13.4%) patients of them were died because of other causes.

Survival and lymph node ratio categories
The median DSS for all patients was 35 months, and the 1-year DSS, 3-year DSS, 5-year DSS were 83.0%, 48.9% and 39.0%, respectively. Lymph node ratio was  defined as the number of positive nodes divided by the total examined nodes. As shown in the Figure 1, the best cutoff points for mLNR were 17.0% and 38.0%. Therefore, the mLNR were classified into four groups, as the following intervals. The mLNRs 0: mLNR = 0%;

Analysis of post-therapy prognostic factors
In the univariate analysis, the depth of invasion, grade and mLNRs were significant risk factors for AGC after NRT, whereas, other factors such as age, gender, race, location, total examined lymph node (TLN) were not correlated with DSS. However, in the multivariate analysis, the depth of invasion lost statistical significance as a prognostic factor (p = 0.638).
According to the Cox regression analysis, we divided the patients into 8 groups (Group 1: mLNRs 0 and Grade I-II; Group 2: mLNRs 0 and Grade III-IV; Group 3: mLNRs 1 and Grade I-II; Group 4: mLNRs 1 and Grade III-IV; Group 5: mLNRs 2 and Grade I-II; Group 6: mLNRs 2 and Grade III-IV; Group 7: mLNRs 3 and Grade I-II; Group 8: mLNRs 3 and Grade III-IV) based on grade and mLNRs (Figure 2A). However, as shown in Figure  2A, there was no significant difference between Group 3 and 4 (log rank p = 0.768), and also have insignificant difference in Group 5, 6 and 7 (log rank p = 0.955). As a result, we propose a novel Grade-lymph node Ratio (G-R) staging system for all the AGC patients after NRT (As shown in Figure 2B). The G-R 1 was defined as mLNR equal to 0% and Grade I-II; G-R 2 as mLNR equal to 0% and Grade III-IV; G-R 3 as mLNRs 1; G-R 4 as mLNRs 2, or mLNRs 3 with Grade I-II; G-R 5 as mLNRs 3 and Grade III-IV. The observed 5-year DSS curves based G-R staging was shown in Figure 2B.

Comparison of predictive accuracy for G-R staging system single independent factors and AJCC staging system
The Area Under the ROC Curve (AUC) values were used to compare the discrimination for the G-R staging system and other prognostic models (mLNRs, AJCC staging system). The statistics power for discrimination between G-R staging system and AJCC staging system were compared in each time points. As shown in Table 3 and Figure 3, the AUC of G-R staging system were higher than mLNRs and AJCC staging system in both 3-year and 5-year time points (p = 0.001, 0.007, respectively). Whereas, the AUC of G-R staging system were higher than AJCC staging system 1-year DSS but did not reach statistical significance (p = 0.282).

DISCUSSION
In the present study, we developed a novel G-R staging system for AGC patients after NRT. A total of 1,429 AGC patients who received surgery following NRT from SEER database between 1998 and 2012 were analyzed. We demonstrated that the new G-R staging system has a more-accurate 3-year and 5-year DSS prediction than the AJCC staging system (p value, < 0.001, 0.008, respectively).
The survival benefit of neoadjuvant therapy had been investigated in various tumors, including esophageal, pancreatic and rectal cancers. [16][17][18] However, the prognostic value of post-therapy pathological  the authors identified that the post-therapy nodal status is significantly associated with patients' overall survival. [12] In fact, indicators that independently and optimally reflect AGC patients' survival who received NRT had not been discovered. In 2005, Ajani JA et al conducted a study to investigate prognosis factors for overall survival for AGC patients after neoadjuvant chemoradiotherapy. [14] A total of 41 AGC patients from M.D. Anderson Cancer Center were analyzed. The authors found that the posttherapy T-N-M stage was associated with overall survival. The findings were interesting, however, the samples were small, and the patients' histological type had not been analyzed. In addition, all of the AGC patients in those studies were form tertiary care institutions that specialize in the treatment of malignancies, and the patients may have different characteristics from community institutions.
Currently, our study first review of a large data from national cancer registry, and established a novel G-R staging system for AGC patients after NRT. Although the conventional AJCC staging system represents an optimal tool in the field of gastric cancer, it is unclear whether this staging system suitable for gastric cancer patients after neoadjuvant therapy. [19] In fact, the known factor posttherapy depth of invasion after surgery held statistical significance in the univariate analysis, however, it did not maintain significance in the multivariate analysis (p = 0.638) which was consistent with previous studies. [12,13] Therefore, the AJCC staging system based on depth of invasion, nodal status and metastatic status may not be applicable for AGC patients after NRT. In contrast, the histology grade and mLNR were the only two prognosis factors correlated with DSS both in univariate and multivariate analysis. Based on this findings, we established a novel classifier for AGC patients after NRT. The G-R staging system appeared to have a more-accurate DSS prediction than AJCC staging system or single independent factors. Since approximately 50% of the gastrectomy cases included in the SEER database came from community institutions, it is inconceivable that the G-R staging system has a broader applicability. [20] The completed (R0) resection with D2 lymphadenectomy was the only potential curative for GC patients. [4] Since resection with D2 lymphadenectomy had not been widely performed in United States, the variable whether the patient received D2 lymphadenectomy was not included in this study. Additionally, although the previous studies demonstrated that the R0 resection was a significant risk for patients' survival, such direct evidence is unavailable because that SEER database did not collect the information of gastrectomy surgical margin status. [21] Given that the current study was a retrospective study, only the patients who had complete information were included in present analysis, there may be a selection bias. In addition, in the univariate analysis and multivariate analysis, the number of positive lymph node was also associated with patient's survival. The discrimination of mLNRs was better than that of positive lymph node number, but the difference was insignificant (data not shown). Given that the positive lymph nodes number identified depends on the pathologic procedure and surgical scope, we used mLNRs instead of positive lymph nodes number in the new classifier. [22] Since the G-R staging system have a more-accurate survival prediction, it would be favor for designing postoperative treatment, particularly, which stage patients can obtain survival benefit from adjuvant radiotherapy after surgery. However, the difference of adjuvant radiotherapy were not correlated with DSS in each G-R stages (detailed data unshown). However, these findings should be interpreted with caution. The reasons are as follows: Firstly, there is missing information regarding the use of adjuvant chemotherapy. It is possible that patients may have received adjuvant chemotherapy after surgery, resulting in a potential confounder in this study. Secondly, the possibility of selection bias could be acknowledged, since the current analysis was not based on a randomized patient population. Thirdly, since there were only 60 patients underwent adjuvant radiotherapy included in this study, it may lead to a bias result.
In conclusion, the current large, general populationbased study demonstrated that the post-therapy pathological characteristics were associated with the survival of AGC patients after NRT. The post-therapy grade and mLNR were the only two predictors for those patients. Based on this finding, the novel G-R staging system showed optimal accuracy of survival prediction for AGC patients after neoadjuvant therapy. Through this classifier, clinicians can estimate the survival of the AGC patients after NRT more precisely.

Surveillance, epidemiology, and end results (SEER) database
The SEER program, a large population-based collaboration program, was surveyed by the National Cancer Institute. A total of 18 population-based cancer registries were participated in this program, and the data are updated annually. It collects and provides approximately 26% American population's cancer incidence and survival data.

Statistical methods
The data of patients' clinicpathological characteristics such as age at diagnosis, gender, race, marital status, surgery, tumor site, size, histology, grade, depth of invasion, number of positive lymph node (PLN) and TLN were collected. The primary endpoint was DSS, which was defined as the time form surgery to cancerrelated death or the last follow-up. The pathological characteristics depth of invasion, lymph nodal stage and tumor stage were restaged according to the 7th edition AJCC staging system.
Tumor grade is a measurement of how closely the tumor cells resemble the normal gastric cancer. Welldifferentiated (Grade I) and moderately-differentiated (Grade II) tumor cells closely resemble the tissue from normal gastric, whereas, poorly-differentiated (Grade III) and undifferentiated (Grade IV) tumor cells are disorganized and abnormal looking.
The mLNR was defined as the number of metastatic lymph nodes divided by the number of examined lymph nodes. A unique category mLNRs 0 was defined for patients with no regional positive lymph node, since those patients had significantly better prognosis than other patients (HR = 0.48, p < 0.001). The patients with mLNR higher than 0% were categorized by X-tile (http:// www.tissuearray.org/rimmlab/). By this statistic tool, the continuous variables can divide into three groups. Associations between each group can be calculated by various standard statistical tests, including the log-rank test for survival and means tests for associations between other marker data. The X-tile can provide the optimal division of the data by p values obtained from a lookup table [23] Independent risk factors were identified by the Cox regression analysis. DSS estimation and survival curves were performed by the Kaplan-Meier method and valided by the log-rank test. The Discrimination between the proposed G-R staging system and AJCC staging system was performed with ROC curve.
All analyses were performed by the software statistical package for social sciences (SPSS) version 19.0 (Chicago, IL), X-tile and the R software version 3.13 (http://www.r-project.org/). All the statistical tests were two sided. p value < 0.05 was considered to be statistically significant.