A nomogram to predict risk of lymph node metastasis in early gastric cancer

Lymph node (LN) metastasis is known as one of the most important prognostic factors for early gastric cancer (EGC) patients. Patients without LNM normally have better prognosis. However, there is no evaluation criteria to accurately assess the possibility of LN metastasis. Therefore, this study aims to establish an effective nomogram for prognosis prediction. In this study, 285 EGC patients from January 2010 to December 2015 were enrolled. Pearson’s Chi-Square (χ2) test (including continuity correction when appropriate) and logistics regression analyses was used to identify the risk factors for LN metastasis. The independent risk factors identified were then incorporated in a nomogram model. The predictive accuracy and discriminative ability of the nomogram were evaluated by receiver operating characteristic curve (ROC) and calibration curve. LN metastasis occurred in 59 (20.7%) EGC patients. And most of these patients were submucosal cancers (48/59). Chi-square test indicated lymphovascular emboli, carbohydrate antigen 19-9 (CA19-9), ulcer, tumor size, tumor infiltration and histological grade were the risk factors, and multivariate logistics analyses confirmed all these six factors were independent risk factors of LN metastasis, which were selected to construct the nomogram. The nomogram proved well calibrated and had good discriminative ability (C-index value: 0.842). The proposed nomogram could result in more-accurate risk prediction for EGC patients.


Patients and methods
Patients. 285 EGC patients in total who underwent surgical treatment in Sun Yat-sen University Cancer Center from January 2010 to December 2015 were selected into this retrospective study. The inclusion criteria were as follows: (i) diagnosed as early gastric cancer and (ii) underwent gastrectomy and D2 lymphadenectomy, (iii) achieved radical (R0) resection. The exclusion criteria were as follows: (i) distant metastasis, (ii) multiple cancers, (iii) stump cancer or recurrent cancer, (iv) accepted preoperative treatment, and (v) died in the perioperative period. In addition, those patients with indications of postoperative chemotherapy have accepted chemotherapy, either S-1 or XELOX regimen. Follow-up assessments including clinical and laboratory examinations, for the first 2 years were conducted every quarterly. During the 3rd to 5th years, assessments were conducted twice a year, and every year after that until death. The primary endpoint was 5-year OS. OS was calculated from the date of surgery until death or to the last follow-up contact, which was used as a measure of prognosis.
Data collection. We further collected the clinic-pathological data of all included patients. The clinical information included age, gender, surgery type, carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9) and carbohydrate antigen 72-4 (CA72-4) before surgery. The pathological information included depth of tumor infiltration, tumor size, tumor location, Lauren type, ulcer, lymphovascular emboli and histological grade. The depth was measured at the deepest point of penetrated carcinoma cells and tumor size was defined as the maximum diameter 17 . The maximum diameter of the lesions was recorded as the tumor size. The surgery type includes distal, proximal and total resection. The tumor location was separated into three parts (upper third, middle third and lower third part of stomach). Tumor histological grade was divided into two groups: the undifferentiated group, which includes undifferentiated and poorly adenocarcinomas, and the differentiated group, which includes moderately and well differentiated adenocarcinomas 18 .
Statistical analyses. Statistical analyses to identify risk factors were conducted by the SPSS (version 26.0).
The continuous variables were shown as the mean ± standard deviation (SD). The categorical variables were analyzed by Pearson's Chi-Square (χ 2 ) test (including continuity correction when appropriate). The independent risk factors of LN metastasis were assessed by multivariate logistic regression analyses. Moreover, the OS were calculated by the Kaplan-Meier method and compared using the log-rank test. Multivariate analyses were used Cox proportional hazards regression models. The primary end point was 5-year OS.
Then, a nomogram was formulated using the package of rms 19 in R studio (version 1.3.10), based on the results of the multivariate logistic regression model. The performance of the nomogram was assessed by calibration curve as indicator of internal calibration, and ROC as a measure of discriminative ability 20 . Meanwhile, the Hosmer-Lemeshow test was used to evaluate goodness of fit of the nomogram.
P values (two sided) < 0.05 were considered statistical significance in all statistical analyses.
Ethical approval statement. All procedures performed in studies involving human participants were in accordance with the ethical standards of our institutional research committee and with the 1964 Helsinki declaration and its later amendments. This article does not contain any studies with animals performed by any of the authors. Due to the retrospective nature of the study, this research was approved by the Ethics Committee of the Sun Yat-sen University Cancer Center, and inform consent was granted a waiver.

Results
Clinical-pathological factors and non-parametric test. In total, 285 EGC patients had undergone treatment at our institution from January 2010 to December 2015, who were selected into present study. LN metastasis was confirmed pathologically in 59 (20.7%) of those 285 cases. The average number of retrieved lymph nodes was 32.6 ± 14.5(range 2-98). The median age of those EGC patients was 57 ± 11.5(range,  years. Non-parametric test indicated that LN metastasis was associated with lymphovascular emboli, CA19-9, ulcer, tumor size, tumor infiltration and histological grade (all P < 0.05). Patients with lymphovascular emboli, larger size > 2 cm, or ulcer have a higher possibility of LN metastasis (all p < 0.001). Tumors with submucosal invasion was associated with higher LNM metastasis than those with intra-mucosa invasion (p = 0.001). Undifferentiated carcinomas were related to lower lymph node metastases (p = 0.005). Patients with CA19-9 ≤ 35U/ml have a lower probability of LN metastasis than those with CA19-9 > 35U/ml (p = 0.028). But LN metastasis was not correlated with gender, age, CA72-4, CEA, tumor location, surgery type, or Lauren type (Table 1).

Multivariate analyses and risk nomogram for LN metastasis.
We furthermore summarized both univariate and multivariate logistic analyses of LN metastasis (  Fig. 3 and Fig. 4. In addition, multivariate Cox regression analyses indicated that only surgery type, CEA and pN stage were independent prognostic predictors ( Table 3).

Discussion
Since long, gastrectomy with D2 lymphadenectomy was considered as the standard and optimal treatment for EGC patients [21][22][23] . The rate of LN metastasis in patients with EGC ranges from 5 to 22%, which means that approximately 70% to 80% of EGC patients undergo overtreatment with D2 lymphadenectomy 24,25 . The development of less invasive treatments, including endoscopic mucosa resection (EMR) and endoscopic submucosa dissection (ESD), have had an important impact on the treatment strategies revolution in the last few decades 9,10,26 . In this retrospective study, the incidence rate of LN metastasis was 20.7% in EGC patients. The incidence rate of LN metastasis for the patients with one or more risk features, such as lymphovascular emboli, presence of ulcer, tumor size > 2 cm, submucosa infiltration, undifferentiated type, or CA19-9 > 35 U/ml, was higher than those patients without these clinical-pathological features. Multivariate analyses showed that these six features were Figure 1. A nomogram predicting the probability of metastatic lymph node involvement for patients with early gastric cancer. The scores of each variable are added to obtain the total score, and then a vertical line is subtracted from the row of total-points to estimate the probability of lymph node metastasis. www.nature.com/scientificreports/ independent risk factors for LN metastasis, and the presence of lymphovascular emboli was considered the most important factor. Previous studies have confirmed these features were independent risk factors. Lymphovascular emboli is the characteristic of lymphovascular invasion, which is considered a rate-limiting step in the lymph node metastatic process 27 . And therefore, lymphovascular emboli is significantly associated with LN metastasis 28 . CA19-9 > 35U/ml has been reported as a reliable risk factor for predicting LN metastasis previously 29 . Larger size (≥ 20 mm), presence of ulcer, tumor submucosal infiltration and undifferentiated type were related to regional lymph node metastasis [29][30][31][32][33][34][35][36] . The results of our study were consistent with the studies mentioned above. And then, we established a nomogram with these six variables related to LN metastasis, which results in more intuitive and accurate in predicting the risk of LN metastasis. The high AUC value (o.842) and calibration curve indicates a good discriminative ability and universal clinical applicability of our nomogram. LN metastasis has been well considered as one of the most important prognostic factors in both early gastric cancer and advanced gastric cancer 23,37 . For those EGC patients with low probability of LN metastasis, ESD or EMR was the optimal treatment 38 . However, there is still a lack of criteria to evaluate the probability of lymph node metastasis. Our study provides a helpful method to solve this problem. For example, a differentiated mucosal patient without lymphovascular or other features, has a low risk of LN metastasis (less than 5%), and is suitable for EMR. On the contrary, an undifferentiated submucosal patient with lymphovascular emboli, larger tumor size (> 2 cm) and ulcer, has a probability of more than 90% for LN metastasis regardless of other features. This patient is recommended to undergo gastrectomy with D2 lymphadenectomy instead of endoscopic treatment. Thus, we think this quantified and visual predictive model allows clinicians to make more objective decisions on treatment options based on the possibility of LN metastasis.
This study had some potential limitations. Firstly, the final selected sample size is small, so we consider improving the nomogram by expanding the sample size in the future. Secondly, this was a single-center retrospective study. In order to further improve the clinical application environment of the nomogram, additional external validation on different populations is required. Finally, P53, HER-2 or other genetic information was not included in the nomogram, which may provide more individualized evidence of our nomogram. In the future, we intend to incorporate more variables into the nomogram to improve its accuracy, including the information www.nature.com/scientificreports/ of imaging examination and molecular detection. In spite of the above limitations, we believe that our rosette map can be helpful for clinical work. A strength of our nomogram is that it was built from the variables which could be easily obtained before ESD or EMR treatment. It therefore means that this nomogram can be used to preliminarily determine the risk of lymph node metastasis of a patient before surgery, and hence to prescribe a preliminary treatment.

Conclusion
In conclusion, we established a nomogram to help clinicians predict the risk of lymph node metastasis in the patients with early gastric cancer. This tool could help clinicians and patients quantify the potential incidence of lymph node metastasis. Surgery plus lymph node dissection is recommended for certain patients with a high risk of LN metastasis, and endoscopic dissection is suitable for those with a low risk. In the future, we will conduct further research to improve clinical application value of the nomogram.