Pathological tumor infiltrative pattern and sites of initial recurrence in stage II/III gastric cancer: Propensity score matching analysis of a multi‐institutional dataset

Abstract Background Advanced gastric cancer frequently recurs even after radical resection followed by adjuvant chemotherapy. The aim of this study was to evaluate the relationship between pathological infiltrative pattern (INF) and initial recurrence patterns in patients with stage II/III gastric cancer using a large multicenter database. Methods We retrospectively analyzed 1098 eligible patients who underwent curative gastrectomy for stage II/III gastric cancer at nine institutions between 2010 and 2014. Patients were categorized into the INF‐a/b and INF‐c groups and adjusted using propensity score matching. Results After propensity score matching, 686 patients (343 for each) were classified in the INF‐a/b and INF‐c groups. There were no significant differences in overall and disease‐free survival between the two groups. In the INF‐a/b group, frequencies of recurrence at the peritoneum, lymph node, and liver were equivalent. In contrast, the peritoneum was the most frequent site and accounted for 60% of the total recurrences in the INF‐c group. The cumulative peritoneal recurrence rate was significantly higher in the INF‐c group than in the INF‐a/b group (hazard ratio 2.47). INF‐c was a significant risk factor for peritoneal recurrences in most subgroups including age, sex, macroscopic type, tumor differentiation, and disease stage, and whether the postoperative treatment was given. Multivariate analysis identified INF‐c as an independent risk factor for peritoneal recurrences. The cumulative liver recurrence rate was significantly higher in the INF‐a/b group than in the INF‐c group (hazard ratio 3.44). Conclusions INF may represent an important predictor of recurrence patterns after curative resection of stage II/III gastric cancer.


| INTRODUCTION
Gastric cancer is a common malignant tumor that is the second cause of all cancer deaths worldwide. 1 Patients with stage I gastric cancer who undergo endoscopic or surgical resection can expect an excellent prognosis. 2 On the contrary, individuals with stage II/III gastric cancer suffer more frequently from recurrences even if they undergo radical resection and adjuvant treatment. 3,4 To detect recurrences early and commence treatment, it is important to predict the sites of recurrences. If physicians predict recurrence sites accurately, the schedule and methods of postoperative surveillance can be optimized.
We recently reported that the pathological infiltrative pattern (INF) was closely related to sites of initial recurrence after curative resection of gastric cancer. 5 In that report, patients with the infiltrative growth type had a significantly high risk of peritoneal recurrences, whereas those with the noninfiltrative type had a significantly high risk of hepatic recurrences. 5 However, the study suffered from several limitations including being a single institution study with a small sample size, using patient data obtained over a prolonged period, and clinicopathologic differences between the patient groups that were compared.
To overcome these problems, we analyzed data from a multicenter integrated database of patients operated during the 5 years between 2010 and 2014 and made comparisons after propensity score matching. The aim of this study was to verify our findings regarding the relationship between the INF and recurrence patterns in patients with stage II/III gastric cancer.

| Patients
Clinical data of 3484 patients who underwent gastrectomy for gastric cancer between January 2010 and December 2014 were retrospectively collected from medical records at nine institutions. Of these, we selected 1098 patients for analysis according to the following inclusion criteria: no preoperative treatment, R0 gastrectomy with systematic lymphadenectomy performed according to the Japanese Gastric Cancer Treatment Guidelines, 6 pathologically diagnosed as stage II or III gastric cancer according to the TNM Classification of Malignant Tumors, 8th Edition, 7 and sufficient data for analysis ( Figure 1A). Patients with gastric stump cancer and those who underwent extended surgery (eg, pancreaticoduodenectomy and Appleby's procedure) were excluded. This study conformed to the ethical guidelines of the World Medical Association Declaration of Helsinki-Ethical Principles for Medical Research Involving Human Subjects. Patients provided written informed consent for surgery and use of clinical data as required by the Institutional Review Board at each participating institute. Pathological diagnosis was determined by two institutional  pathologists using paraffin sections stained with hematoxylin  and eosin. The pathological INF types were classified into  either INF-a (expansive growth having a distinct border with  the surrounding tissues), INF-b (intermediate type), or INF-c (infiltrative growth having no distinct border with the surrounding tissues), according to the Japanese Classification of Gastric Carcinoma as shown in Figure S1 and in previous reports. 5,8

| Surgery and postoperative management
Patients underwent gastrectomy with systematic lymphadenectomy according to the Japanese Gastric Cancer Treatment Guidelines, 6 and the reconstruction method was selected at the surgeon's discretion. Patients received postoperative follow-up for 5 years or until recurrence that included physical examinations and laboratory tests including serum tumor markers every 3 months, and enhanced computed tomography (chest and abdominal cavity) once every 6 months, and upper gastrointestinal endoscopy at 1, 3, and 5 postoperative years as described in the Japanese Gastric Cancer Treatment Guidelines 6 . Disease recurrences were diagnosed based on radiological or pathological findings, with serum tumor markers playing an adjunctive role. 6 Twelve months of S-1 (an oral fluoropyrimidine derivative) monotherapy or 6 months of capecitabine plus oxaliplatin has been recommended to all patients as postoperative adjuvant treatment unless contraindicated by a patient's condition or patient refusal. 9-11 Treatment after recurrences was determined according to the evidence available at the time of treatment, according to the patient's condition, and with the patient's consent.

| Propensity score matching
We employed propensity score matching to balance more strictly essential variables for the comparison analyses between INF-c (invasive growth type) and INF-a/b (noninvasive type). Propensity scores were estimated using a logistic regression model based on age, sex, tumor location, type of gastrectomy, disease stage, and adjuvant chemotherapy. Age and sex were included in the variables for the matching as the most fundamental demographics. Since tumor location affects biological properties of gastric cancer and extent of resection, it also was included. Type of gastrectomy should be balanced because it can influence on postoperative nutritional status and chemotherapy tolerability. Lastly, disease stage and adjuvant chemotherapy were major relevant factors to the main point of analysis in the present study (postoperative prognosis and recurrences), and thus, they were used for the matching. One-to-one matching without replacement was performed using a 0.1 caliper width, and the resulting scorematched pairs were used in subsequent analyses.

| Statistical analysis
The chi-square and Mann-Whitney tests were used to compare the two groups. Overall and disease-free survival rates were calculated using the Kaplan-Meier method, and the difference between survival curves was evaluated using the log-rank test. Risk factors for peritoneal recurrences were evaluated using binomial logistic regression analysis. P < 0.05 was considered statistically significant. All statistical analyses were performed using JMP 13 software (SAS Institute Inc, Cary, NC, USA).

| INF and clinical signatures before matching
Before propensity score matching, 707 and 391 patients were classified in the INF-a/b and INF-c groups, respectively. As shown in Table 1, there were significant differences between the INF-a/b and INF-c groups in age, sex distribution, tumor location, type of gastrectomy, pathological stage, and administration rate of postoperative adjuvant chemotherapy. Overall, survival time was significantly shorter in the INF-c group than in the INF-a/b group (hazard ratio [HR] 1.91, 95% confidence interval [CI] 1.46-2.50, P < 0.0001; Figure  S2A). Patients in the INF-c group were more likely to have shorter disease-free survival times compared with those in the INF-a/b group (HR 1.64, 95% CI 1.32-2.05, P < 0.0001; Figure S2B). In the multivariate analysis, INF-c was identified as an independent risk factor for peritoneal recurrence (odds ratio, 1.98; 95% CI 1.33-2.99; P = 0.0007) along with Borrmann type 4/5 tumor, pT4, lymphatic involvement, lymph node metastasis, and stage III (Table S1).

| Patient characteristics after matching
After propensity score matching, 686 patients (343 for each) were classified in the INF-a/b and INF-c groups, and age, sex distribution, tumor location, type of gastrectomy, pathological stage, and administration rate of postoperative adjuvant chemotherapy were well balanced (Table 1)    preoperative serum CEA levels, vascular invasion, and pathological lymph node metastasis (Table S2).

| Prognostic impact of INF in stage II/III gastric cancer after matching
After propensity score matching, survival differences between the INF-a/b and INF-c groups were reduced. There were no significant differences in overall survival and disease-free survival between the INF-a/b and INF-c groups ( Figure 1B).

| Association between INF and recurrence patterns
The frequency of initial recurrence sites is depicted in Figure 2A. The overall recurrence rates of the INF-a/b and INF-c groups were equivalent. In the INF-a/b group, the frequency of recurrences at the peritoneum, lymph node, and liver was similar and each accounted for approximately 30% of the total recurrences. In the INF-c group, the peritoneum was the most frequent site and it accounted for 60% of the total recurrences. The prevalence of peritoneal recurrences was significantly greater in the INF-c group than in the INF-a/b group. On the contrary, liver recurrences were more commonly observed in the INF-a/b group. The cumulative peritoneal recurrence rate was significantly higher in the INF-c group than in the INF-a/b group (HR 2.47, 95% CI 1.62-3.88, P < 0.0001; Figure 2B). In univariate analysis, macroscopic tumor size ≥50 mm, Borrmann type 4/5 tumor, pT4, undifferentiated tumor, lymphatic involvement, vascular invasion, INF-c, and stage III were identified as significant risk factors for peritoneal recurrence. Multivariate analysis identified INF-c as an independent risk factor for peritoneal recurrence after curative gastrectomy (odds ratio, 1.70; 95% CI 1.06-2.78; P = 0.0270) along with Borrmann type 4/5 tumor, pT4, and stage III ( Table 2). In contrast, the cumulative liver recurrence rate was significantly higher in the INF-a/b group than in the INF-c group (HR 3.44, 95% CI 1.71-7.65, P = 0.0003; Figure 2C).

| Further evaluation of INF-c as a risk factor of peritoneal recurrences
A forest plot to evaluate the impact of INF-c on peritoneal recurrences is shown in Figure 3. INF-c was a significant risk factor for peritoneal recurrences in most subgroups including age, sex, macroscopic type, tumor differentiation, and disease stage. Of note, INF-c had a significant influence on peritoneal recurrences both in patients who underwent surgery alone (n = 225, HR 2.90, 95% CI 1.28-7.38, P = 0.0010) and in those who underwent postoperative adjuvant chemotherapy (n = 461, HR 2.34, 95% CI 1.43-3.95, P = 0.0006).

| DISCUSSION
In East Asia, pathological INF has long been routinely evaluated in surgically resected specimens. 12,13 INF can be easily determined using only hematoxylin and eosin-stained sections. 12,14 However, there have been a few recent studies focusing on the clinical significance of INF in gastric cancer. [13][14][15] In the present study, we revalidated the impact of pathological INF on prognosis and recurrence patterns after curative gastrectomy in patients with stage II/III gastric cancer using a large multicenter dataset. Using propensity score matching, it was revealed that there was no difference in overall survival time, disease-free survival time, and overall recurrence rates between the INF-a/b and INF-c groups. In contrast, significant differences in recurrence patterns were detected between the INF-a/b and INF-c groups, and INF-c was found to be an independent risk factor for peritoneal recurrences.
Our findings suggest that INF-c type gastric cancer tends to grow out of the stomach wall directly by skipping the lymphatic and blood vascular systems, and that INF-a/b type gastric cancer causes lymphatic involvement and vascular invasion in parallel with growth outside the stomach wall. It has been reported that INF-c is closely linked to other risk factors for peritoneal metastasis such as Borrmann type 4 tumors, poorly differentiated tumors, and serosal invasion. [13][14][15][16] However, our multivariate analysis revealed that INF-c was an independent risk factor for peritoneal recurrences. Since there is a certain correlation between the macroscopic type and pathological INF, clinicians believe that INF-c groups are mostly type 4/5 tumors and associate with peritoneal recurrences. However, physicians sometimes experience cases of INF-c gastric cancer with non-linitis plastica type macroscopic appearance and/or well-differentiated type histology. 2 17,18 In our patient cohort, S-1 monotherapy comprised the majority (90%) of treatment regimens for adjuvant therapy. It has been suggested that S-1 adjuvant chemotherapy mainly suppresses peritoneal recurrences based on results of the Adjuvant Chemotherapy Trial of TS-1 for Gastric Cancer (ACTS-GC) phase III clinical trial. 19 However, results of a subgroup analysis showed that INF-c was a significant risk factor for peritoneal recurrences irrespective of whether the adjuvant chemotherapy was given, and highlighted the utility of INF-c as an indicator to screen peritoneal recurrences even in patients who underwent S-1 adjuvant treatment. Development of postoperative treatments that excel in controlling peritoneal recurrences shed a new light on patients with INF-c gastric cancer.
Accurate prediction of recurrence sites is extremely important in postoperative follow-up because early detection of recurrences will be possible by conducting appropriate surveillance. 20,21 By detecting recurrent lesions early, the first-line treatment for recurrences can be initiated early. In the current Japanese Treatment Guideline of Gastric Cancer, follow-up methods after curative resection are recommended uniformly according to only pathological disease stage. 6 In the case of INF-c, patients who are at high risk for a peritoneal recurrence but at low risk for a hematogenous metastasis, a regular abdominal computed tomography (CT), ultrasound, or digital rectal examination is advisable to detect malignant ascites fluids and peritoneal nodules. For patients with suspected peritoneal recurrences, staging laparoscopy or ascites puncture cytology is considered accordingly. Meanwhile, for patients with INF-a/b type gastric cancer, who are at higher risk of hematogenous and nodal metastasis, broad range (cervical to abdominal) contrast CT scan or positron-emission tomography might be given preference to detect recurrences early. Once evidence for selecting appropriate anticancer drugs or treatments based on the pattern of metastasis can be established, INF can be a candidate selection factor of treatment methods.
This study also has some limitations. It was a retrospective study. It was difficult to completely eliminate the pathologists' subjectivity in the evaluation of INF. Furthermore, lack of information on postrecurrence treatment restricted the discussion. Nevertheless, our preceding thesis was successfully reproduced using a large multicenter database, indicating that the value of INF in predicting recurrence patterns was enhanced.
In conclusion, pathological INF represents an important predictive factor for recurrence patterns after radical resection of stage II/III gastric cancer and may guide clinicians in providing appropriate postoperative management.

RESEARCH INVOLVING HUMAN PARTICIPANTS INFORMED CONSENT
This study conforms to the ethical guidelines of the World Medical Association Declaration of Helsinki-Ethical Principles for Medical Research Involving Human Subjects, and written informed consent for surgery and the use of clinical data were obtained from all patients as required by the Institutional Review Board of all participating institutes.