Risk factors of lymph node metastasis or lymphovascular invasion for early gastric cancer: a practical and effective predictive model based on international multicenter data

With the advancement of diagnostic techniques and the popularization of health examinations, the diagnostic rate of early gastric cancer (EGC) worldwide has gradually increased [1]. Lymph node metastasis (LNM) is an important disease feature that affects the prognosis of patients with EGC and determines the extent of lymph node dissection. Japanese gastric cancer treatment guidelines suggest that the endoscopic resection (ER) of EGC is feasible [2]. Many scholars are also exploring ER for EGC, and from the earlier endoscopic mucosal resection (EMR) to the current endoscopic submucosal dissection (ESD), there has been an expanding trend of ER indications for EGC. The EMR/ESD indications were categorized as the absolute indication for standard EMR/ESD and the expanded indication for ESD [2, 3]. The absolute indication is mucosal EGC with differentiated type and size ≤2 cm. The following are expanded indications: (1) mucosal differentiated EGC without ulcers and size > 2 cm, (2) mucosal differentiated EGC with ulcers and size ≤3 cm, (3) differentiated EGC with submucosal invasion < 500 mm and size ≤3 cm, and (4) mucosal undifferentiated EGC without ulcers and size ≤2 cm. However, for patients with EGC who already have LNM, such treatment may risk postoperative tumor recurrence [4, 5]. Therefore, to explore the risk factors of LNM in EGC is of great clinical significance. However, most studies on the LNM of EGC include lymphovascular invasion (LVI) as one of the predictors [6, 7]. Since LVI must be confirmed by postoperative pathological examination, it is difficult to know the status of LVI preoperatively. The present of LVI usually indicates lymph node metastasis or micrometastasis [3, 8‐11]. Current predictive models for a simultaneous assessment of LNM/LVI risk in EGC have not been reported. Therefore, this study explored the available preoperative factors for LNM/LVI in EGC and stratified the risk of LNM/LVI based on recursive partitioning analysis (RPA) using international multicenter data.

With the development of endoscopic technology and the increase in population aging in recent years, endoscopic resections, represented by EMR and ESD, to treat EGC have been widely carried out [14, 15]. Compared with traditional abdominal surgery, ER has many advantages such as less trauma and improved postoperative quality of life [16‐19]. However, the endoscopic treatment of gastric cancer itself has its own limitations, namely, the inability of a surgeon to perform a dissection of the lymph nodes around the stomach where metastasis may occur. In addition, there is no effective or accurate way to predict LNM. Thus, endoscopic treatment is faced with a certain degree of a postoperative recurrence risk, and controversy exists [4, 5]. Therefore, a study to investigate the risk factors for potential LNM in EGC is important. Zheng et al. found that age, macroscopic types, tumor size, tumor histological types, the degree of differentiation, ulceration, LVI, and the depth of invasion were closely related to the LNM of EGC and thus established a nomogram to predict the risk of LNM in EGC [7]. Lee et al. conducted a study focusing on early gastric papillary adenocarcinoma and found that LVI was the only risk factor for LNM in EGC [20]. These studies confirm that LVI is an independent risk factor for LNM. Some studies even showed that LVI is associated with poor prognosis in patients with gastric cancer [21, 22]. In our study, the similar 5-year CSS rates of LVI+ and LVI- group may be due to the small proportion of LVI+ in patients without LNM. LVI sometimes indicates that patients have a significant risk of LNM and may lead to a worse prognosis. Moreover, it is suggested in the EMR/ESD-related guidelines that if the pathology confirms the presence of LVI in the resected specimen, the additional radical gastrectomy is needed [2]. From this point of view, LVI is of great significance for patients with gastric cancer, especially those with EGC who may receive ESD/EMR surgery. However, most of the previous studies on EGC LNM include LVI as one of the predictive risk factors, which makes these findings have obvious limitations in clinical application. On the other hand, in Kim et al.’s study, 53.4% (31/58) of EGC patients with LNM had LVI, while the other 46.6% did not have LVI in postoperative pathological examination [23]. In this study, as shown in Fig. 1, there were 45 of EGC patients with LNM had LVI (16.6%, 45/271); not all the EGC patients with LNM had LVI. While there were only 44 cases of N0 patients with LVI in this study. If LVI alone with N0 status was used to be an endpoint, there were 271 EGC patients with LNM will not be analyzed. Therefore, we combined LNM and LVI as an endpoint in order to make a more comprehensive assessment of the status of lymph node metastasis and lymphovascular invasion, and hope to provide more precisely medical evidences for the strategy choices of radical surgery for EGC.

Previous studies on the LNM of EGC mostly report that the tumor size, depth of invasion and tumor histological types have impacts on LNM [7, 24, 25]. However, whether sex has an impact on LNM in EGC remains to be determined. Pyo et al. established a model for predicting LNM in poorly differentiated-type intramucosal gastric cancer. The effect of sex on LNM was comparable with that of tumor size and depth of invasion [6]. However, in analysis of the risk factors of LNM in intramucosal EGC, Kim et al. found sex was not a risk factor [24]. The study by Deng et al. suggests that estrogen receptor-α, a mark highly correlated with LNM, is highly expressed in human gastric cancer [26]. We compared the incidence of LNM/LVI in women ≥60 years old and < 60 years old. The results showed that the incidence of LNM/LVI in women ≥60 years old was lower than < 60 (24.6% vs. 27.6%), but not yet statistically significant (P = 0.484). In addition, we compared the differences in the incidence of LNM/LVI between men and women in different age groups. The results showed that the incidence of LNM/LVI was significantly lower in men < 60 years old than in women (18.5 vs. 27.6, P = 0.004). While in patients ≥60 years old, the incidence of LNM/LVI in men was comparable to that of women (20.6% vs. 24.6%, P = 0.268) (Additional file 2: Table S4). These results indicate that the difference in the incidence of LNM/LVI between men and women disappears with age increasing, which may be caused by the regression of estrogen. Therefore, we hypothesize that the high incidence of LNM/LVI in female patients may be due to higher estrogen levels in female which has a direct or indirect impact on LNM or LVI. Based on the results of the multivariable analysis, a predictive model of LNM/LVI for EGC was established. The AUC of the model in the training set and the validation set were 0.694 and 0.796 respectively, which were no lower than that of the predictive model established by Pyo et al. (the AUC of the training set was 0.70, and the AUC of the internal validation set was 0.68) [27]. The results show that the model has moderate to good accuracy in both the training set and the validation set.

Based on LNM/LVI status, we further performed a RPA to reclassify the patients into different risk groups then proposed strategies for lymph node dissection according to the risk group (Fig. 6). RPA is a statistical method for multivariable analysis, which divides group into subgroups according to the priority of several binary independent variables to correctly classify the members of a group and intuitively generate a concise decision tree to determine decision rules with higher sensitivity or specificity [28]. This approach is widely used in medical decision-making. In 1982 Goldman et al. became the first to use a RPA to establish a decision tree for the diagnosis of patients with acute chest pain [29]. Recently, Fonarow et al. used a RPA to successfully group in-patients with acute decompensated heart failure [30]. In this study, we found that LNM/LVI was present in 5.4% of patients with differentiated EGC in the mucosa, which is relatively safe for receiving ER. LNM/LVI was found more than 12% in male with undifferentiated mucosal EGC, suggesting that these patients should be treated with D1+ lymph node dissection. The incidence of LNM/LVI in the high-risk and extremely high-risk patient groups (female undifferentiated mucosal EGC and submucosa EGC) were above 24%. Therefore, the recommendation is that these groups of patients receive standardized D2 lymph node dissection.

This study had several limitations. First, this was a retrospective study, and bias inevitably existed in the data selection. Second, the tumor size and the depth of invasion in this study were confirmed by postoperative pathology. Although obtaining relatively accurate preoperative data by endoscopy or endoscopic ultrasonography is possible, bias may exist when compared with the postoperative pathological diagnosis. Third, this study lacked subjective diagnostic information, such as the macroscopic types of EGC. Fourth, due to the lack of records of postoperative adjuvant therapy in the SYSUCC and IMIGASTRIC databases, we were unable to provide information about postoperative adjuvant therapy. However, to our knowledge, this study was the first to explore the effect of preoperative factors on potential LNM in patients with EGC using international multicenter data. An effective multivariable prediction model for LNM/LVI was established and validated, and the use of RPA made the model more clinically useful. Based on the findings of this study, we have proposed recommendations for lymph node dissection in EGC for different risk groups, which will be of benefit to surgeons in making better clinical decisions.

In this study, by using PRA to divide EGC patients into different risk groups, we have found that the incidence of LNM/LVI in differentiated mucosal EGC is low, which indicated that ER is a treatment option. The incidences of LNM/LVI in undifferentiated mucosal EGC and submucosa EGC are high and gastrectomy with lymph node dissection is suggested.