From January 1986 to December 2013, 161 patients (median, 65; range 38–88), including 85 male and 76 female patients, with histologically confirmed PDAC underwent a pancreatectomy with D0-D3 lymph node dissection at the Department of Surgery, Kitasato University Hospital (Kanagawa, Japan). Patients with histological variants, including mucinous cystic adenocarcinoma, intraductal papillary adenocarcinoma, acinar cell carcinoma and endocrine carcinoma were excluded from the study. Preoperative workup included an ultrasonography, computed tomography, endoscopic retrograde pancreatography, endoscopic ultrasonography and magnetic resonance cholangiopancreatography to evaluate primary and metastatic tumor sites.

The preoperative serum CA19-9 levels were investigated immediately prior to surgery, avoiding the effects of obstructive jaundice and/or inflammation of the biliary duct. The recommended upper limit of a normal CA19-9 level is 37 U/ml, when CA19-9 levels are determined and defined from the standard deviations of healthy individuals. CA19-9 levels were determined using a chemiluminescent enzyme immunoassay kit (LUMIPLSE G; cat. no., CA19-9-N) obtained from Fujirebio US, Inc. (Malvern, PA, USA).

The surgical resection tissues were obtained from pancreaticoduodenectomy (n=39), pylorus-preserving pancreaticoduodenectomy (n=74), distal pancreatectomy (n=43) and total pancreatectomy (n=5). Partial resection of the portal vein was performed if the surgeon observed tumor invasion of the portal vein. Intraoperative pathologic assessment of the proximal or distal pancreatic margins was performed using frozen-tissue sections. If the pancreatic margin was positive for cancerous cells, further resection of the pancreas was performed. The present study was approved by the Institutional Review Board for Observation and Epidemiological Study, Kitasato University Medical Ethics Organization (approval no. B18-017; Kanagawa, Japan). All patients agreed to the use of their samples in scientific research and written informed consent was obtained from all patients.

All histological and other clinicopathological factors were judged independently and blindly by histopathologists, and all histopathologic factors from the Japanese classification system (JCS) version 6 (17) were obtained from medical records. In the current study, stage was determined according to the 6th edition of the International Union Against Cancer (UICC) tumor-node-metastasis classification (18). Assessment of cancer infiltration at the surgical margin was evaluated according to the JCS, as DPM, pancreatic cut end margin or bile duct cut end margin. According to the National Comprehensive Cancer Network guideline (19), the surgical margin was defined as a superior mesenteric artery (SMA) margin, posterior margin, portal vein groove margin, portal vein margin, pancreatic margin or pancreatic surface and bile duct margin. Details of the DPM included the SMA margin, posterior margin and portal vein groove margin. Surgical margin-positive was defined as tumor cells present in the dissection surface, as judged by histopathologists. When requested by a pathologist, the gene mutation was investigated using non-radioisotopic single-strand conformation polymorphism (SSCP) from each DNA sample (n=97).

For simultaneous DNA analysis, the previously described protocols were conducted (20,21). Briefly, this consisted of five steps. Firstly, to prevent cross-contamination, small sections of fresh solid tissue were sampled by scraping with disposable bamboo combs, which are 3×3×120 mm rods made of bamboo with a spatula like ends. Secondly, one-step DNA extraction with lysis buffer containing proteinase K, Nonidet P-40 and Tween 20 was performed. Subsequently. polymerase chain reaction (PCR) was performed with each gene primer presented in Table I and the conditions presented in Table II. Next, SSCP analysis with polyacrylamide gels was performed, and detection was achieved by silver staining. In this analysis, mutated bands were evident at 1:64 dilutions of the mutated alleles.

A total of 108 patients (67%) received empirical adjuvant therapy; this consisted of 105 patients who received chemotherapy and 3 patients who received radiotherapy. Another 23 patients (14%) received therapy for remnant tumor; specifically 22 patients who received chemotherapy and 1 patient who received radiotherapy. A number of chemotherapy regimens consisted of 5-fluorouracil (5-FU)-based chemotherapy, including 5-FU only (n=3), 5-FU/cisplatin (n=9) or 5-FU/Adriamycin/mitomycin-C/nimustine hydrochloride (n=2) by venous, portal or arterial infusion. Other chemotherapy regimens included tegafur/uracil (n=11), 5′-deoxy-5-fluorouridine (n=3) or TS-1 (n=3) as oral therapy, or mitomycin-C (n=7). In addition to gemcitabine (n=83) or gemcitabine+TS-1 (n=6) by venous or arterial infusion. Radiotherapy consisted of 1.8 Gy per day to a total dose of 50 Gy. Routine follow-up consisted of physical examination, laboratory studies and computed tomography imaging at 3- to 4-month intervals for the first 2 years, at 6-month intervals for years 3 to 5, and followed by annual follow-ups from thereon. Only the first sites of recurrence were recorded.

Continuous variables were compared using the Student's t-test and nominal variables were compared using Pearson's chi-square test. The Kaplan-Meier method was used for disease-specific survival (DSS) analysis and the difference in survival rate was assessed by log-rank test (22). DSS was measured from the date of surgery to the date of mortality or last follow-up. Mortality associated with causes other than PDAC were not counted in this measurement. The variables that demonstrated prognostic potential suggested by univariate analysis (P<0.05) were subjected to multivariate analysis with a Cox proportional hazards model. Propensity scores were matched using a caliper width of 0.2 multiplied by the standard deviation of values that was calculated by a logistic regression analysis. A receiver operating characteristic (ROC) curve was generated using JMP software v.11.0 (SAS institute, Cary, NC, USA) and the area under the curve (AUC) was used to optimize the best cut-off value. The associations between K-ras gene mutation and prognostic factors were examined with the Student's t-test. P<0.05 was considered to indicate a statistically significant difference. All statistical analysis was performed using JMP software v.11.0.

The first aim of the current study was to validate the prognostic relevance of preCA19-9 and DPM, which have previously been identified as independent prognostic factors in PDAC (12). The characteristics of 161 patients with PDAC and the univariate prognostic factors are summarized in Table III. Significant prognostic factors included lymphatic permeation factor (ly; P=0.036), vascular permeation factor (v; P=0.0087), tumor differentiation (P=0.0037), intrapancreatic nerve invasion factor (ne; P=0.00025), retropancreatic tissue invasion factor (RP; P=0.0053), portal venous system invasion factor (PV; P=0.0133), extrapancreatic nerve plexus invasion factor (PL; P=0.0004), arterial system invasion factor (A; P=0.0019), preCA19-9 level (P<0.0001), DPM (P<0.0001) and UICC 6th stage (P<0.0001). Arterial invasion was composed of pT4 (celiac artery or superior mesenteric artery; n=2) and non-pT4 (splenic artery; n=3). By increasing the number of patients from the previous pilot study (13), several pathological factors, which are not UICC staging factors, including ly, v, ne, RP, PV and A, were newly identified as univariate prognostic factors in PDAC.

The ten variables that exhibited prognostic potential identified by univariate prognostic analysis, including ly, v, tumor differentiation, ne, RP, PV, PL, A, preCA19-9, DPM and UICC 6th stage, were subjected to multivariate analysis. This analysis revealed that preCA19-9 [P=0.0002, relative risk (RR)=2.8], DPM positive (P=0.0002, RR=2.4), PV (P=0.01, RR=2.0) and A (P=0.035, RR=3.3) were the remaining prognostic factors independent of UICC stage (P=0.0015; Table III).

Similar to previous studies (23,24), preCA19-9 and DPM were identified as independent prognostic factors. The Kaplan-Meier curves of each factor are presented in Fig. 1A for UICC stage, Fig. 1B for PreCA19-9 and Fig. 1C for DPM. Notably, all DPM-positive patients succumbed to the disease within 5 years, while a number of patients with preCA19-9 levels >37 U/ml were alive with no recurrence after five years.

In addition, survival outcomes for the combination of preCA19-9 and DPM are presented in Table III and Fig. 1D. The prognosis of patients with PDAC was sub-classified into four groups (A, B, C and D) according to preCA19-9 and DPM (group A, preCA19-9 >37 U/ml and DPM-positive; group B, preCA19-9 >37 U/ml and DPM-negative; group C, preCA19-9 <37 U/ml and DPM-positive; and group D, preCA19-9 <37 U/ml and DPM-negative). Of note, group D demonstrated the best clinical outcome and included long-term survivors (10/30) of PDAC (Fig. 1D).

A DPM-positive status may be the best indicator of highly aggressive tumor characteristics in PDAC; therefore, a risk model for DPM prediction was generated by logistic regression analysis. The clinicopathological factors analyzed were as follows: age, gender, ly, v, ne, RP, PV, A, PL, UICC stage and preCA19-9. Using this prediction model, an ROC curve was generated and presented in Fig. 2A (AUC=0.80) and the propensity score (PS) to predict DPM was calculated. The density distribution of PS was analyzed between DPM-positive and DPM-negative patients, and a notable difference in the PS distribution was identified between the groups (Fig. 2B). The matching of PS was performed by logit exchange. The logit exchange scores of the matched cases between both DPM-positive and DPM-negative cases were presented linearly in a scatter plot (Fig. 2C). The logit exchange scores of the matched cases between DPM-positive and DPM-negative were distributed in Kernel density estimation differently from the unmatched cases (Fig. 2D, left panel), as compared with the differential ranges prior to matching (Fig. 2D, right panel). Following matching, no significant differences were identified for factors potentially affecting DPM between DPM-positive and DPM-negative cases (Table IV). Notably, the factors identified to be associated with DPM positive were RP and RL, and these factors were corrected by score matching. As a result, DPM was demonstrated as a robust prognostic factor in PDAC (P=0.0008; Fig. 2E).

Subsequently, initial recurrent sites were investigated in DPM-positive patients with PDAC. Among the 60 DPM-positive patients, 52 recurrences were identified. Liver metastases were the most prevalent in 21 patients (35%), followed by lymph node metastases in 13 patients (22%), peritoneal dissemination in 5 patients (8%), lung metastasis in 1 patient (35%), nerve plexus of the SMA in 9 patients (15%) and remnant pancreas in 3 patients (5%).

Subsequently, the associations of DPM with the mutational status of K-ras and TP53 genes in PDAC were investigated. For K-ras gene mutation, 96 patients with PDAC were investigated. K-ras gene mutation was identified in 42 patients (44%), among whom 34 patients possessed mutations in exon1 (codon 12 or 13). The Kaplan-Meier curve was generated according to K-ras gene mutation status (P=0.043 by the log-rank test and P=0.0047 by the Wilcoxon test; Fig. 3A). Since DPM has been cited as a poor prognostic factor in the present results, univariate analysis was performed for factors associated with DPM status (Table V). RP (P<0.001), PL (P=0.0003) and K-ras gene mutation status (P=0.0002) were significantly associated with DPM-positive status. Multivariate analysis confirmed that RP (P=0.037), PL (P=0.026) and K-ras gene mutation status (P=0.0004) remained as independent factors of DPM-positive statue. A significant association was identified between DPM and K-ras gene mutation status (Fig. 3B). DPM-positive patients had a K-ras gene mutation in 66.7% (26/39) of cases, while DPM-negative patients exhibited a K-ras gene mutation in 28.1% (16/57) of cases. For the TP53 gene mutation, 67 patients with PDAC were examined and a TP53 gene mutation was identified in 14 (20.9%). Of these, 6 patients were DPM-positive, and no significant difference was identified with regard to DPM status (P=0.9275).