Methods
Between 1996 and 2015, a total of 2722 patients with primary gastric cancer, excluding patients with cancer of the remnant stomach, underwent surgery at the Tokyo Women’s Medical University Hospital (Tokyo, Japan). Our institutional review board waived the need for informed consent because this was a retrospective study. TNM categories were determined according to the Japanese classification of gastric carcinoma [
3] (Table
1), which is widely used. In this classification, the depth of tumor invasion is recorded as the T category, lymph node metastasis is recorded as the N category, and the presence/absence and sites of distant metastasis are recoded as the M category.
Table 1
Patient characteristics
Age | 63.7 ± 11.8 |
Sex |
M | 1849 |
F | 873 |
T | |
M | 647 |
SM | 581 |
MP | 283 |
SS | 173 |
SE | 858 |
SI | 180 |
N |
N0 | 1498 |
N1 | 580 |
N2 | 417 |
N3 | 227 |
M |
M0 | 100 |
M0 | 2622 |
P1 | 233 |
P0 | 2485 |
Various pretreatment factors were investigated as potential predictors of tumor stage, including the age, sex, tumor markers (serum carcinoembryonic antigen (CEA) (≤ 5.0 ng/ml vs. ≥ 5.1 ng/ml) and serum cancer antigen 19-9 (CA19-9) (≤ 37 U/ml vs. ≥ 38 U/ml)), and tumor diameter (≤ 59 mm vs. ≥ 60 mm). To investigate the predictive value of tumor location and extent, the stomach was divided into thirds (upper third, middle third, and lower third) and the gastric circumference was divided into four equal parts for assessment of circumferential involvement (lesser curvature, greater curvature, anterior wall, and posterior wall; a circumferential category was also added). Extension of the tumor into the esophagus or duodenum was also assessed. Finally, the histological type and the macroscopic type (Types 0–4) were investigated. Macroscopic and histological types were determined according to the Japanese classification of gastric carcinoma [
3].
Statistical analysis
We initially investigated the association between pretreatment factors and tumor stage (I/II vs. III/IV) by univariate logistic regression analysis. Then, we entered the candidate factors identified by univariate analysis as explanatory variables for multivariate logistic regression analysis. Subsequently, the factors that predicted tumor stage were weighted according to the relative magnitude of the β-coefficient in logistic regression analysis to develop the CSP score.
Receiver operating characteristics (ROC) analysis was performed, and the cut-off value of the CSP score was calculated from the ROC curve by determining the Youden index. The accuracy of the CSP score for staging gastric cancer (I/II vs. III/IV) was evaluated in our patient cohort by the chi-square test. All analyses were performed with JMP software ver. 12 (SAS Institute, Cary, NC).
Discussion
We devised a new staging score for gastric cancer (the CSP score) by analysis of pretreatment factors in 2722 patients, and we demonstrated that this score could effectively discriminate between stage I/II disease and stage III/IV disease before initiation of treatment. It has been reported that preoperative chemotherapy may improve outcomes for stage III/IV disease, suggesting that an accurate method of predicting the stage before starting treatment could be useful. Numerous prognostic factors for gastric cancer have been reported, including the depth of tumor invasion [
4], site of lymph node metastasis [
4], number of metastatic lymph nodes [
5‐
7] lymph node metastasis ratio [
7‐
9], distant metastasis [
4], results of peritoneal lavage cytodiagnosis [
10,
11], tumor diameter [
12], macroscopic type [
13,
14], tumor location [
15,
16], age [
17,
18], sex [
17], lymphatic invasion [
19], venous invasion [
20], histologic type [
21], macroscopic serosal invasion [
22], tumor markers (CEA and CA19-9) [
23,
24], and extent of lymphadenectomy [
24‐
26]. Most of these factors can be assessed before initiation of treatment, apart from those related to lymph nodes, lymphatic invasion, and venous invasion.
In the present study, both the tumor size and macroscopic type were confirmed to be useful for identifying advanced disease. Mucinous carcinoma was also a significant factor on multivariate analysis, which is reasonable since 90% of mucinous tumors are advanced because early cancers release most of their mucin into the gastric lumen [
27].
Various modalities are employed for diagnosis and staging of gastric cancer, with each method being influenced by inherent characteristics, observation conditions, instrument performance, and differences between institutions. Therefore, clinical staging of gastric cancer has a relatively low accuracy (60–70%) compared with pathological staging [
28‐
33]. Also, endoscopic staging is often based on clinical experience because of the lack of objective criteria for assessing the depth of invasion. While endoscopic ultrasonography (EUS) is useful, it is difficult to determine the depth of ulcerated lesions and the accuracy is no better than that of standard endoscopic diagnosis [
29]. Accordingly, we investigated preoperative factors related to tumor stage and we devised the CSP score by weighting each factor to obtain a useful predictor of gastric cancer stage. The depth of invasion is assessed preoperatively by endoscopic observation, EUS, and abdominal ultrasound. It has been reported that NBI observation achieves 92% accuracy for determining the depth of invasion of early gastric cancer, but this decreases to 57–86% with white light observation and is around 74% for EUS [
29].
EUS, CT, positron emission tomography, and abdominal ultrasound can all be used to assess lymph node metastasis, but the reported accuracy varies widely from 50 to 92% [
30‐
32]. Accuracy of CT is comparatively high among these modalities, with lymph nodes > 10 mm in diameter being detectable and visible nodes likely to be metastatic [
33]. However, many metastatic nodes are not enlarged, and it is impossible to predict the presence/absence of metastasis from size alone. Also, tumor progression is judged by the number of nodal metastases, but it is difficult to separate N1 (1–2 nodes involved) from N2 (3–5 nodes involved). On the other hand, imaging is useful for assessing the tumor diameter and distant metastasis, except for remote lymph node metastasis or micrometastases, while laparoscopic examination can be performed to identify peritoneal metastases [
33].
Various scoring systems for gastric cancer have been reported that predict the prognosis, complications, or risks for elderly patients. A depth prediction score that separates M-SM1 disease from SM2 disease based on endoscopic findings such as tumor location, macroscopic type, and tumor size has also been reported. However, there have been few reports about diagnostic scores or methods that can judge the applicability of endoscopic treatment for early gastric cancer [
34].
A risk score system has also been reported for preoperative prediction of the TNM stage of gastric cancer based on four factors: serum albumin, tumor size, and the T and N categories determined by helical computed tomography. When that score was investigated in 108 curatively resected patients, it showed poor versatility due to the large influence of the computed tomography findings [
35].
While staging of gastric cancer is based on T, N, and M factors, the pretreatment factors in the CSP score are also associated with the prognosis or stage, and T, N, and M factors were specifically excluded from the CSP score to increase its objectivity and versatility.
Moreover, there have been no previous reports of a staging score that can be used to decide whether a patient should receive preoperative chemotherapy, which is the function of the CPS score developed in this study.
While we found that the CPS score was effective for discriminating between low- and high-stage disease, its sensitivity was somewhat low (78.9%). It is possible that the sensitivity of the CSP score could be improved by adding factors related to the nutritional status (albumin or prealbumin) and tumor markers (CA125 or AFP).