If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Corresponding author. Department of Cancer Control and Population Science, Graduate School of Cancer Science and policy & Gastric Cancer Branch, Research Institute and Hospital, National Cancer Center, Ilsan-ro 323, Ilsandong-gu, Goyang-si, 10408, Korea.
Center for Gastric Cancer, National Cancer Center, KoreaDepartment of Cancer Control and Population Science, Graduate School of Cancer Science and Policy, National Cancer Center, Korea
Selective diagnostic laparoscopy in gastric cancer patients at high risk of peritoneal metastasis is essential for optimal treatment planning.
In this study available clinicopathologic factors predictive of peritoneal seeding in advanced gastric cancer (AGC) were identified, and this information was translated into a clinically useful tool.
Totally 2833 patients underwent surgery for AGC between 2003 and 2013. The study identified clinicopathologic factors associated with the risk of peritoneal seeding for constructing nomograms using a multivariate logistic regression model with backward elimination. A nomogram was constructed to generate a numerical value indicating risk. Accuracy was validated using bootstrapping and cross-validation.
The proportion of seeding positive was 12.7% in females and 9.6% in males. Of 2833 patients who underwent surgery for AGC, 300 (10.6%) were intraoperatively identified with peritoneal seeding. Multivariate analysis revealed the following factors associated with peritoneal seeding: high American Society of Anesthesiologists score, fibrinogen, Borrmann type 3 or 4 tumors, the involvement of the middle, anterior, and greater curvature, cT3 or cT4cN1 or cN2 or cN3, cM1, and the presence of ascites or peritoneal thickening or plaque or a nodule on the peritoneal wall on computed tomography. The bootstrap analysis revealed a robust concordance between mean and final parameter estimates. The area under the ROC curve for the final model was 0.856 (95% CI, 0.835–0.877), which implies good performance.
This nomogram provides effective risk estimates of peritoneal seeding from gastric cancer and can facilitate individualized decision-making regarding the selective use of diagnostic laparoscopy.
]. However, countries such as Korea and Japan selectively administer diagnostic laparoscopy for four main reasons: a high proportion of early-stage patients due to nationwide screenings for gastric cancer, avoidance of a procedure that may be associated with some morbidity especially in older comorbid patients [
]. Traditionally, computed tomography scans (CT) readings have facilitated decisions to pursue diagnostic laparoscopy, but this modality fails to consistently deliver accurate readings of peritoneal seeding [
]. However, combining CT findings, independently associated with P+, with other independent clinical risk factors may augment the clinician's ability to assess the peritoneum for seeding from primary gastric cancer. On the other side of the spectrum, routine diagnostic laparoscopy – as recommended in Western countries – can lead to similar issues. However, when done on hurry or by non-experienced surgeon may lead to false negatives results that lead to unnecessary laparotomies, can delay definitive treatment, increase costs, and introduce adverse oncologic effects. In the past 15 years, several retrospective studies that recognized this issue have found several clinicopathologic factors associated with increased rates of peritoneal seeding, such as regional lymph node spread, signet ring cell status, and tumor size, depth, and location [
]. However, a lack of integrative models bridging these results to a clinically useable form perpetuates the gap between knowledge and practice.
In this study, a large cohort was analyzed to develop a risk assessment tool comprised of routinely measured and validated parameters. This decision-making aid will allow clinicians and advanced gastric cancer patients to generate a personalized score regarding the risk of metastasis to the peritoneum, and a more effective role can be defined for the use of diagnostic laparoscopy in the management of gastric cancer.
2.1 Patient selection
The institutional review board approved the study (IRB No. NCC2016-0079) and waived the need for patients’ informed consent.
Using information provided by the National Cancer Center, Korea, 7354 patients who received surgical intervention for primary gastric adenocarcinoma from 1 January 2003 to 31 December 2013 were retrospectively identified. The exclusion criteria were as follows: history of malignant diseases, existence of other malignant diseases, history of gastrectomy, preoperative chemotherapy or radiotherapy for gastric cancer, and diagnosis of clinical T1 or grossly early gastric cancer type tumor. Of the original 7354 patients, 2833 met all the above criteria and were included in our analysis (Fig. 1). Data analyses were performed from 5 April 2016 to 30 September 2019.
We defined the peritoneal seeding as any macroscopic dissemination that found during the surgery. We excluded cytology-positive cases from the analysis because cytology results were unavailable for immediate analysis in the operating room.
2.2 Preoperative clinicopathologic predictors of peritoneal seeding
All factors were examined and obtained preoperatively within a ten-year analysis period. Tumor type and TNM were classified according to the Japanese Gastric Cancer Classification (JGCC) 4th edition. Tumor type and size were determined by esophagogastroduodenoscopy. Tumor location and wall involvement were classified as an upper third (U), middle third (M), lower third (L), anterior wall (Ant), posterior wall (Post), lesser curvature (Less), or greater curvature (Gre) based on the JGCC classification.
Classified factors also included sex, age, American Society of Anesthesiologists (ASA) score (1–2/more than 3), cT (2/3/4), cN (0/1/2/3), cM (0/1), cH (0/1), cP (0/1), histopathologic classification from biopsy specimen (well-differentiated; WD/moderately differentiated; MD/poorly differentiated; PD/Signet Ring Cell; SRC/Others), tumor type (2/3/4), tumor size, multiplicity of cancer (single/multiple), tumor location in the stomach (U/M/L), involvement of gastric wall (Ant/Post/Less/Gre), complete blood count, platelet lymphocytes ratio (PLR), serum fibrinogen, tumor markers (CEA, AFP, CA19-9), and CT findings.
CEA, AFP, and CA19-9 were categorized by each value based on the reference value of the institution. All tumor markers were divided into four categories: 1) within reference value, 2) more than but less than double that of the reference value, 3) more than double but less than triple, and 4) more than triple.
For the staging, the 8th edition TNM classification for gastric carcinoma of the American Joint Committee on Cancer (AJCC) was used. The clinical TNM staging were based on preoperative CT reports. CT findings were extracted from preoperative CT reports, which were read and described by experienced radiologists. The CTs were categorized by the presence of each finding as follows: ascites, omental cake, thickening of the peritoneal wall, plaques or nodules in the peritoneal cavity, or infiltration or stranding in abdominal fatty tissue.
2.3 Statistical analysis
Patients were divided into two groups according to the status of peritoneal seeding: seeding positive (+) or seeding negative (−). We excluded cytology-positive cases in the analysis because we cannot discriminate with bear eyes and cytology results were unavailable for immediate analysis in the operating room. The distribution of age, blood examination results, and tumor size in the two seeding groups were compared using the Wilcoxon Mann-Whitney U test. Associations between the existence of peritoneal seeding and ASA score, tumor markers, CT imaging, and TNM stage were evaluated by the chi-square test and the Mantel-Haenszel chi-square test. See Table 1 for details. To model the probability of peritoneal seeding, univariate and multivariate logistic regression analyses were conducted on associative factors. In the final model, all covariates included were selected using backward elimination.
Table 1Demographic information, blood test and imaging characteristics, tumor markers, TNM Stage and histology distribution of the study cohort.
Percentage of total patients (n = 2833). For some variables sample size is smaller than 2833 due to missing observations (AFP, n = 2767; CEA, n = 2801; CA19-9, n = 2378; Fibrinogen, n = 2799; Plt/Lymph, n = 2828; Abdominal CT finding, n = 2831; Tumor size, n = 2751; Multiplicity, n = 2828).
2 (Ulcerated tumors with raised margins surrounded by a thickened gastric wall with clear margins)
3 (Ulcerated tumors with raised margins, surrounded by a thickened gastric wall without clear margins)
4 (Tumors without marked ulceration or raised margins, the gastric wall is thickened and indurated and the margin is unclear)
Tumor size, median (IQR), cm
Circumferential tumor location
Abdominal CT finding
Plaque or Nodule
Stranding or Infiltration
cT (primary tumor)
2 (Tumor invades the muscularis propria)
3 (Tumor invades the subserosa)
4 (Tumor invasion is contiguous to or exposed beyond the serosa or tumor invades adjacent structures)
cN (regional metastasis)
0 (No regional lymph node metastasis)
1 (Metastasis in 1–2 regional lymph nodes)
2 (Metastasis in 3–6 regional lymph nodes)
3 (Metastasis in 7 or more regional lymph nodes)
cM (distant metastasis)
cH (liver metastasis)
cP (peritoneal metastasis)
otherM (other distant metastasis)
Signet ring cell carcinoma
Abbreviation: ASA, American Society of Anesthesiologist; EGD, Esophagogastroduodenoscopy.
c P value without any mark is from Chi-Square test.
d P value from Wilcoxon Mann-Whitney U test.
-: Less than 10 cases.
a Percentage of total patients (n = 2833). For some variables sample size is smaller than 2833 due to missing observations (AFP, n = 2767; CEA, n = 2801; CA19-9, n = 2378; Fibrinogen, n = 2799; Plt/Lymph, n = 2828; Abdominal CT finding, n = 2831; Tumor size, n = 2751; Multiplicity, n = 2828).
b Row percentage of each characteristic.
e Low, ASA Physiological Status category 1 and 2; High, ASA Physiological Status category 3.
To evaluate the performance of the final model, receiver operating characteristic (ROC) curve analysis was done. The area under the curve (AUC) was calculated, and diagnostic cut-off values were derived. Youden's index [
] was used to select an optimal cut-off. For internal validation, 10-fold cross-validation and the bootstrapping technique with 1000 resamples from the study data were used. Based on the final model, a nomogram was developed to predict the individual risk of peritoneal seeding occurrence. Statistical results were considered significant if the p-value was less than 0.05. All statistical analyses were performed with SAS (9.4; SAS Institute Inc., Cary, North Carolina) and R software (3.2.2; www.R-project.org).
Of the 2833 patients with advanced gastric cancer in our study cohort, 300 (10.6%) were intraoperatively identified with peritoneal seeding.
3.1 Factors predictive of peritoneal seeding in patients with gastric cancer
Baseline and distribution of clinical patient information, blood and imaging results, tumor markers, biopsy histology, and TNM stage can be found in Table 1. The study population consisted of 69.1% male and 30.9% female patients, and the proportion of seeding positive was 12.7% in females and 9.6% in males (p = 0.01). The median age of all patients was 60 (interquartile range; IQR 49–68).
Table 2 lists the results of the logistic regression analyses. In multivariate logistic regression, covariates selected via backward elimination were as follows: decrease in age (OR = 0.99, 95% CI 0.98–1.00, p = 0.02), high ASA score (OR = 2.13, 95% CI 1.21–3.73, p = 0.01), increase in fibrinogen (OR = 1.20, 95% CI 1.05–1.38, p = 0.01), tumor type 3 (OR = 2.69, 95% CI 1.62–4.47, p < 0.001) or 4 (OR = 8.71, 95% CI 4.61–16.42, p < 0.001) rather than type 2, involvement of middle (OR = 1.95, 95% CI 1.45–2.64, p < 0.001), anterior (OR = 0.60, 95% CI 0.43–0.83, p = 0.002), greater curvature (OR = 1.88, 95% CI 1.39–2.55, p < 0.001), CT finding of ascites (OR = 3.51, 95% CI 2.54–4.86, p < 0.001), peritoneal thickening (OR = 2.30, 95% CI 1.20–4.41, p = 0.01), plaques or nodules (OR = 7.03, 95% CI 4.14–11.93, p < 0.001), clinical T stage cT3 (OR = 3.22, 95% CI 1.94–5.32, p < 0.001) or cT4 (OR = 3.44, 95% CI 1.91–6.18, p < 0.001) rather than cT2, clinical N stage cN1 (OR = 4.21, 95% CI 2.35–7.54, p < 0.001), cN2 (OR = 3.76, 95% CI 2.06–6.84, p < 0.001), cN3 (OR = 3.35, 95% CI 1.66–6.78, p < 0.001) rather than cN0, and cM1 (OR = 2.29, 95% CI 1.38–3.82, p = 0.002) rather than cM0. No significant difference was found in other covariates.
Table 2Result of logistic regression analysis for the probability of peritoneal seeding.
2, Ulcerated tumors with raised margins surrounded by a thickened gastric wall with clear margins; 3, Ulcerated tumors with raised margins, surrounded by a thickened gastric wall without clear margins; 4, Tumors without marked ulceration or raised margins, the gastric wall is thickened and indurated and the margin is unclear.
0, No regional lymph node metastasis; 1, Metastasis in 1–2 regional lymph nodes; 2, Metastasis in 3–6 regional lymph nodes; 3, Metastasis in 7 or more regional lymph nodes.
Signet ring cell carcinoma
Abbreviation: OR, Odds ratio; CI, Confidence interval, ASA, American Society of Anesthesiologist; EGD, Esophagogastroduodenoscopy.
a Covariates are from the best fitting multivariate logistic regression model.
b Low, ASA Physiological Status category 1 and 2; High, ASA Physiological Status category 3.
c Standardized variable.
d 2, Ulcerated tumors with raised margins surrounded by a thickened gastric wall with clear margins; 3, Ulcerated tumors with raised margins, surrounded by a thickened gastric wall without clear margins; 4, Tumors without marked ulceration or raised margins, the gastric wall is thickened and indurated and the margin is unclear.
e 2, tumor invades the muscularis propria; 3, tumor invades the subserosa; 4, Tumor invasion is contiguous to or exposed beyond the serosa or tumor invades adjacent structures.
f 0, No regional lymph node metastasis; 1, Metastasis in 1–2 regional lymph nodes; 2, Metastasis in 3–6 regional lymph nodes; 3, Metastasis in 7 or more regional lymph nodes.
3.2 Internal validation of performance and accuracy
The AUC under the ROC curve was 0.86, which implies a good performance of the developed nomogram in predicting seeding occurrence (Fig. 2). Sensitivity and 1-specificity, which represent the performance of a logistic model in different values of diagnostic cut-offs, were derived and summarized in eTable 2. An optimal cut-off probability level maximizes Youden's index, which was 7.5% with 84.6% sensitivity and 67.5% specificity. A fitted multivariate logistic regression model from each of the 1000 bootstrap resamples was also drawn. Mean parameter estimates obtained from the bootstrap samples were very close to the estimates in the final model, as biases for all selected covariates were less than 0.005 (eTable 1). Average AUC from 10-fold cross-validation was 0.85 (eFig. 1).
3.3 Resulting nomogram
A nomogram consisting of 13 variables was constructed (Fig. 3A) based on the multivariate logistic analysis results. As compared to the current standard of radiologic examination, this nomogram outperformed CT scans in the ability to identify peritoneal seeding. False-positive readings of peritoneal seeding using CT scans were high, even with interpretation by highly experienced radiologists (ascites: 67.8%, peritoneal thickening: 59.4%, plaque or nodule: 43.8%, stranding or infiltration of fat: 69.6%), whereas the nomogram produced an overall sensitivity rate of 84.6%. Fig. 3B juxtaposes two cases to demonstrate this nomogram's superior diagnostic capabilities compared to CT scans. Both patients were diagnosed with advanced gastric cancer without suspicion of seeding upon review of CT scans, and peritoneal deposits were subsequently discovered during exploratory laparotomy. However, risk estimates derived from the nomogram retrospectively predicted the presence of peritoneal seeding.
In this large, single-institution study, a nomogram was generated after the investigation of clinicopathologic factors predictive of peritoneal seeding in advanced gastric cancer. Due to the highly heterogeneous nature of gastric cancer, personalized medicine tailored to the unique indexes of each patient is a necessary but unmet need in this oncologic field [
]. Nomograms, which can be understood as an individualized and non-invasive diagnostic medium, have shown superior prognostic abilities compared to traditional staging systems in gastric cancer as well as other cancer types [
], and comparison to other decision aids (including risk groups, artificial neural networks, probability tables, and classification and regression tree analyses) demonstrated its superior ability to assess risk [
The inability to accurately identify peritoneal deposits in the preoperative setting may reveal unexpected and unresectable tumors during exploratory laparotomy – resulting in longer hospital stay, increased risk of postoperative complications, increased medical costs, and decreased likelihood of receiving systemic therapy in unresectable cases [
]. Diagnostic laparoscopy is, therefore, critical in the staging of gastric cancer patients. However, overutilization of this resource to detect seeding leads to several disadvantages. While there is a slight increase in morbidity associated with diagnostic laparoscopy, this does not necessarily outweigh its benefits, diagnostic laparoscopy has the potential to differentiate between localized and widespread peritoneal metastasis. Also in the localized type, documentation of the metastatic area before neoadjuvant chemotherapy is essential to include it in the resection specimen at the time of gastric resection, especially if the metastasis have disappeared after chemotherapy. However, only a small proportion of patients eventually gain clinically useful information [
], as complications of this procedure include hemodynamic instability, posterior penetrating trauma with highly potential bowel injury, and intraabdominal injury. Second, routine diagnostic laparoscopy results in inefficient economics when done in cases whose risk of peritoneal seeding is low. But this procedure still has important role in staging and decision making in the selective cases. In a study of Korea's healthcare delivery system, 85% of patients received unnecessary outpatient hospital services [
]. Selective diagnostic laparoscopy ameliorates the disadvantages found with its over- and underutilization, and this nomogram can be used to identify candidates who can benefit from this procedure.
To construct the nomogram, routinely tested, preoperative factors were identified in a comprehensive manner, and aggressive statistical analysis of specific variables was performed on these wide-ranging categories (patient characteristics; blood examinations; tumor biomarkers, type, size, and location; patterns of stomach wall involvement; CT scans; and histopathologic and TNM classifications).
Concerning elevated levels of fibrinogen, there are two theories to explain this event. The first is that fibrinogen is directly involved with the pathogenesis of peritoneal seeding, where its increased interaction with fibroblast growth factor-2 (FGF-2) promotes cell growth and angiogenesis [
]. To note, previous studies theorized that fibroblasts at peritoneal metastatic sites stimulate tumor progression by promoting adhesion of tumor cells to the sub-mesothelium, which is a critical step in peritoneal dissemination [
As for stomach wall involvement, the results indicated that localized invasion of the greater curvature increased the probability of peritoneal metastasis. It might be due to lack of obstructive symptoms which allow tumors to grow large until diagnosis. A previous study found similar results, implicating the anterior wall as a site correlated with increased peritoneal metastasis [
], which generally falls in line with results associating peritoneal seeding with the upper and middle portions of the stomach.
CT indicators of peritoneal seeding – involving ascites, peritoneal thickening, and plaque/nodule in peritoneal cavity – have also been associated with peritoneal metastasis. However, accurate staging of peritoneal metastasis with traditional radiologic modalities is challenging because microscopic malignancies can develop unnoticed [
]. In a recent study assessing the role of CTs in identifying peritoneal disease in 52 patients with potentially curable gastric cancer, the overall sensitivity was found to be 25%, with one false positive and six false negative patients whose seeding was identified during laparoscopy, and nine patients whose seeding was identified during surgery [
]. Even after a multimodal radiologic approach using CTs, ultrasound, and PET scans, unexpected and unresectable tumors were found during exploratory laparotomy.
Poorly differentiated and SRC histological classifications also showed an independent association with peritoneal metastasis in advanced gastric cancer patients. A study of 662 patients with advanced SRC gastric adenocarcinoma showed a trend toward deeper tumor invasion of the gastric wall, greater lymph node spread, and peritoneal metastasis compared to the non-SRC group [
Internal validation of this nomogram revealed robust performance in terms of accuracy and the balance between specificity and sensitivity. A 1000-sample bootstrapping method designed to estimate the sensitivity of each factor showed a bias of less than 0.005 between parameter estimates derived from the final multivariate logistic model and mean model parameter estimates obtained from random sampling; this means there was hardly any deviation between observed and sampled pools. Testing of the selected covariates using a ROC curve revealed an AUC indicative of a statistically good correlation between specificity (84.6%) and sensitivity (67.5%) with a cut-off probability of 7.5%. Future studies from external sources will provide further validation.
This study has several limitations. The use of retrospective data can potentially incorporate selection bias into the results despite efforts to adjust for confounding variables. Also, patients with advanced stage gastric cancer who received neoadjuvant chemotherapy were excluded, so the role of staging laparoscopy nomogram in those group of patients should be further evaluated. Despite the robust results of this nomogram, solely relying on this prognostic methodology will produce incidences of false positive and negative cases and therefore this tool should serve as an adjunct to judicious, multimodal clinical decision-making. However, deciding cut off value in nomogram depends on different situations like the neoadjuvant treatment setting or the upfront surgery setting, and therefore should be adjusted according to varying purposes.
Nomograms would be useful to assist clinicians in planning diagnostic laparoscopy in different situations. This nomogram may be used to minimize unnecessary laparotomies by detecting patients unqualified for surgery due to peritoneal seeding, whose diagnostic laparoscopy did not indicate otherwise. This, in turn, may reduce the large volume of patients who receive upfront surgery in Korea and Japan. In this case, cut-off criteria with higher sensitivity would be preferred. Considering the aforementioned purposes, we evaluated the performance of the nomogram. Among 2560 patients who did not receive diagnostic laparoscopy, 215 had peritoneal seeding. Among them, our nomogram correctly predicted 179 patients (83.3%) to be seeding positive with and optimal cut-off probability at 7.5%. If diagnostic laparoscopy was conducted after nomogram prediction, this means one could identify peritoneal seeding approximately 83% correctly and avoid unnecessary laparotomy.
On the other hand, in the situation in which neoadjuvant chemotherapy is popular, like in many western countries, our nomogram can be utilized to select the patients who can get chemotherapy without diagnostic laparoscopy to minimize treatment delay, medical cost, and morbidity. Thus, in this case, cut-off criteria with higher specificity would be preferred. In real the nomogram had better ability to detect the peritoneal metastasis if compared to the computer tomography. Finally, we believe that there is no magic procedure or imaging, or nomogram and all the mentioned evaluation methods should be discussed and used as needed dependent on the circumstance of the cases.
In conclusion, this study demonstrated the independent association of various factors with peritoneal seeding. This nomogram could have different clinical utilities in different treatment strategies and finally contribute to patient-centered decision-making in any situation.
Source of funding
There are no declared conflicts of interest that could lead to bias for any of the authors of this manuscript. This work was supported by grant NCC-1410130, 1710120, 2010090, and 2310210 from the National Cancer Center, Korea.
This study was approved by the Institutional Review Board of the National Cancer Center (IRB No. NCC2016-0079).
Consent to participate
The IRB waived the need for informed consent for this retrospective study.
Data available on request due to privacy/ethical restrictions: The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
CRediT authorship contribution statement
Norihito Kubo: Data accrual, Design of the study, manuscript writing, critical review. Hyunsoon Cho: Statistical, Formal analysis, manuscript writing, critical revie. Dahhay Lee: Statistical analysis, manuscript writing, critical review. Hannah Yang: manuscript writing, critical review. Youngsook Kim: Data acquisition, critical review. Harbi Khalayleh: manuscript writing, critical review and revision. Hong Man Yoon: Data acquisition, critical review. Keun Won Ryu: Data acquisition, critical review. George B. Hanna: Critical Review, approval of the study. Daniel G. Coit: Critical Review, approval of the study. Kenichi Hakamada: Critical Review, approval of the study. Young-Woo Kim: design of the study, Data acquisition, Formal analysis, manuscript writing, critical review.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
This work was supported by grant NCC-1410130, 1710120, 2010090, 2310210 from the National Cancer Center, Korea. There are no declared conflicts of interest that could lead to bias for any of the authors of this manuscript.
Appendix A. Supplementary data
The following is the Supplementary data to this article: