Pattern and risk factors for postoperative recurrence in borderline resectable pancreatic cancer after upfront surgery

Abstract

BACKGROUND

Postoperative recurrence is the main cause of treatment failure in borderline resectable pancreatic cancer (BRPC) patients. Elaborating patterns and risk factors for tumor recurrence can help clinicians tailor treatment decision, thereby improving the clinical outcome of these patients. However, although upfront surgery serves as an important treatment option for BRPC patients and can lead to distinct recurrence risks and patterns, the exact characteristics of postoperative recurrence in these patients were unclear.

AIM

To describe the pattern of postoperative recurrence and identify its risk factors in BRPC patients receiving upfront surgery.

METHODS

This retrospective study analyzed 216 BRPC patients receiving upfront surgeries. Preoperative data, surgical condition, pathology and postoperative prognosis were analyzed. Kaplan-Meier method was applied for survival comparison. Risk factors for disease-free survival (DFS) and tumor recurrence were identified with Cox and logistic regression. Fine-Gray analysis was applied to adjust for competing event. Restricted cubic splines were used to evaluate the dose-response correlation and receiver operating characteristic analysis was applied for determination of optimal cutoff value.

RESULTS

Of 72.7% patients experienced postoperative recurrence with a medium DFS of 11 months. Most recurrences occurred within postoperative year one, and liver was the most common recurrence site. Patients with early recurrence had decreased post-recurrence survival compared to late recurrence patients (P < 0.001). Preoperative [carcinoembryonic antigen (CEA), hazard risk (HR) = 1.021, 95% confidence interval (CI): 1.002-1.040, P = 0.029], tumor diameter (HR = 1.101, 95%CI: 1.003-1.320, P = 0.044), differentiation (HR = 0.662, 95%CI: 0.473-0.927, P = 0.016) and lymph node metastasis (HR = 1.491, 95%CI: 1.009-2.202, P = 0.045) were independent risk factors for DFS. Preoperative CEA [relative risk (RR) = 1.083, 95%CI: 1.008-1.163, P = 0.029], tumor diameter (RR = 1.236, 95%CI: 1.013-1.507, P = 0.037), differentiation (RR = 0.508, 95%CI: 0.271-0.951, P = 0.034) and lymph node metastasis (RR = 2.151, 95%CI: 1.007-4.297, P = 0.030) were independent risk factors for early recurrence. Vascular invasion extent showed limited association with postoperative recurrence.

CONCLUSION

Most postoperative tumor recurrences occurred in postoperative year one in BRPC patients receiving upfront surgery, and liver was the most common site of recurrence. Preoperative CEA, tumor size, differentiation and lymph node metastasis were risk factors for DFS and early postoperative recurrence.

Key Words: Borderline resectable pancreatic cancer; Upfront surgery; Tumor recurrence; Pattern; Risk factor

Core Tip: This study characterized the recurrence patterns in borderline resectable pancreatic cancer patients undergoing upfront surgery. It identified that early recurrence, predominantly liver metastasis within the first year, is a major challenge. Preoperative carcinoembryonic antigen level, tumor diameter, poor differentiation, and lymph node metastasis were identified as independent risk factors, whereas the extent of vascular invasion showed limited association. These findings improved the understanding of postoperative tumor recurrence in borderline resectable pancreatic cancer patients and would help clinicians make optimal treatment decision so as to improve the prognosis of these patients.

INTRODUCTION

Pancreatic cancer is the most lethal malignant cancer with an increasing disease burden worldwide[1]. Due to the close anatomical correlation between pancreas and several abdominal vessels, invasion to important abdominal vessels is frequently observed in pancreatic cancer patients, causing poorer overall prognosis. Tumors with portomesenteric venous invasion that can be reconstructed, localized contact with hepatic artery, ≤ 180° contact with superior mesenteric artery and celiac axis are defined as borderline resectable pancreatic cancer (BPRC), a subgroup of vascular-invasive pancreatic cancer that may have a chance of radical surgery[2]. Rapid development in surgical techniques and application of neoadjuvant chemotherapy provide BPRC patients the opportunity to receive radical surgery, significantly improving the postoperative prognosis of these patients[3,4].

However, compared to resectable pancreatic cancer, BRPC had a more aggressive biological behavior and advanced tumor stage[5]. Therefore, BRPC patients are at higher risk of postoperative tumor recurrence, which significantly impacting their postoperative prognosis. According to statistics, more than 60.0% of patients develop tumor recurrence, of which 60.0% of these recurrences occur in the first year after surgery, decreasing the 5-year survival rate to only 13%[6]. Understanding the pattern and potential risk factors for postoperative recurrence is crucial for clinicians to better tailor further follow-up schedules and adjuvant therapy in order to improve their overall prognosis. In BRPC patients who receive surgical treatment secondary to neoadjuvant chemotherapy, the pattern of postoperative tumor recurrence has previously been reported[7]. Prior research also identified a series of preoperative and pathological risk factors for postoperative recurrence in BRPC patients who received surgical treatment after neoadjuvant chemotherapy, including carbohydrate antigen (CA) 19-9, CA125, lymph node metastasis, microscopic neural invasion and adjuvant chemotherapy, effectively assisting clinicians to make optimal treatment decisions for these patients[7-9]. Although neoadjuvant chemotherapy is recommended for BRPC patients prior to surgery, some patients are ineligible for neoadjuvant chemotherapy due to their physical condition or personal preference, making upfront surgeries a necessary alternative option for BRPC patients in clinical practice. These patients had a higher proportion of perineural invasion and lymph node metastasis and decreased R0 resection rate, leading to a significantly higher risk of tumor recurrence[10,11]. Thus, elucidating recurrence pattern and risk factors in these patients are also important in clinical practice. Without neoadjuvant chemotherapy, patients may exhibit distinct recurrence characteristics compared to those who receive neoadjuvant therapy, necessitating determination of pattern and risk factors for recurrence in this special population.

Thus, this study was conducted to clearly describe the postoperative recurrence pattern of BRPC patients in a relatively large retrospective cohort to provide references for more precise prognostic assessment and management strategies for BRPC patients undergoing upfront surgery.

MATERIALS AND METHODS

Study population and included data

This retrospective study was conducted in accordance with the principles of the Declaration of Helsinki. Collection and analysis of historical data were reviewed and approved by the Ethics Committee of Beijing Chao-Yang Hospital, Capital Medical University, approval No. 2024-ke-512. The collection and analysis of historical clinical data got the patients’ verbal informed consent. The requirement for written informed consent was waived due to the retrospective design.

Clinical data of patients who were diagnosed as BRPC between January 2011 and December 2024 were retrieved from the electronic medical record system of our hospital. The collected clinical data included demographic data (sex, age, smoking history, diabetes, body mass index, biliary drainage), laboratory examinations (preoperative white blood cell, neutrophil, lymphocyte, hemoglobin, platelet, albumin, alanine aminotransferase, total bilirubin, carcinoembryonic antigen (CEA), CA19-9, imaging examinations (site of vascular invasion, circumference extent of vascular invasion, length of vascular invasion, imaging vascular morphology changes, arterial invasion), surgical variables (surgical procedure, vascular reconstruction procedure, operation time, intraoperative blood loss, intraoperative blood transfusion), pathological data (tumor site, differentiation, diameter, lymph node metastasis) and postoperative follow-up (adjuvant chemotherapy, time and site of tumor recurrence, mortality). All laboratory data were collected from the last tests performed within 1 week before surgery, and all imaging examination results were obtained from the last scan in 1 month before surgery. All clinical data were collected using Microsoft Excel (Microsoft 365).

Patients were further included based on eligibility criteria: Inclusion criteria: (1) Preoperatively diagnosed with BRPC; (2) Received surgical treatment; and (3) Pathological diagnosis of pancreatic adenocarcinoma. Exclusion criteria: (1) Preoperative evaluation indicated surgical contraindication; (2) Received neoadjuvant chemotherapy; (3) Received palliative surgery intraoperatively; (4) Unable to achieve R0 resection during surgery; (5) Perioperative death; and (6) Lost to follow-up after discharge.

Evaluation of vascular invasion and diagnostic criteria for BRPC

In our cohort, all patients with suspected BRPC received abdominal enhanced computed tomography (CT) following admission to evaluate the site, angle, length and contour abnormalities of invaded vessels. The angle of vascular invasion was defined as angle of tumor contact around vascular circumference at the cross-section with the maximum invasion range, while the length of vascular invasion was defined as the distance between the proximal and distal sites of contact between tumor and portomesenteric venous axis. Imaging vascular morphology changes referred to significantly observable stenosis or luminal defects (luminal area decreased more than 25% compared to distal normal portal or mesenteric vein) of the portomesenteric vein caused by the tumor. BRPC was diagnosed according to NCCN guidelines: (1) Angle of portomesenteric vascular invasion ≤ 180° with contour abnormality; (2) Angle of portomesenteric vascular invasion > 180° but can be safely reconstructed; (3) Tumor contact inferior vena cava; (4) Tumor contact common hepatic artery without extending to the celiac axis or branches of the hepatic artery; and (5) Angle of the superior mesenteric artery or celiac axis invasion ≤ 180°[12].

Surgical procedure for BRPC

The surgical procedure for each patient was determined according to their tumor site. Radical pancreaticoduodenectomy (PD) was applied in patients whose tumor was located at the pancreatic head and uncinate process, distal pancreatectomy (DP) was applied in patients whose tumor was located at the pancreatic body and tail. For tumors which may not achieve radical resection by performing PD or DP, total PD (TP) was adopted as an alternative to ensure R0 resection. The extent of surgical resection and lymph node dissection are shown in Supplementary Figure 1 and Supplementary Table 1. PD, DP and TP were performed using the arterial-first procedure.

All patients received vascular resection and reconstruction. Invaded vessels were transected 0.5 cm from the site of tumor invasion, and fast frozen pathology was performed to ensure negative vascular margins. Direct end-to-end anastomosis were utilized for vascular reconstruction if stenosis-free and tension-free anastomosis could be achieved, otherwise allogeneic vascular grafts were used to restore the continuity of vessels (Supplementary Figure 1).

Follow-up schedule

All included patients were first followed up at postoperative months 1 and 3. Follow-up frequency was then set at 3 months in first 2 years and every 6 months thereafter. All selected patients were followed up by an outpatient coordinator via telephone and clinic visits. The primary endpoint of follow-up was postoperative tumor recurrence, and the secondary endpoint was postoperative mortality. Results of imaging examinations (abdominal enhanced CT, lung CT), treatment, tumor recurrence and survival were collected.

Statistical analysis

Normally distributed continuous variables were presented as mean ± SD, whereas non-normally distributed data are shown as median (interquartile range). Differences between two groups were assessed using the Student’s t-test for normally distributed data and the Mann-Whitney U test for non-normally distributed data. One way analysis of variance was used for comparing quantitative data between multiple groups. Kaplan-Meier method and log-rank test was applied for survival comparison using the “survminer” package. Multicollinearity was assessed by the variance inflation factor (VIF) in a linear regression model, and VIF > 5 was considered to have severe multicollinearity. Univariate analysis was used to screen out potential risk factors, and factors which were statistically significant (P < 0.05) were included in multivariate analysis to identify potential risk factors. Independent risk factors for postoperative disease-free survival (DFS) was identified by mulitvariate Cox regression model. Fine-Gray competing-risk analysis was applied to adjust for the competing risk of non-cancer deaths by using “cmprsk” package. A logistic regression model was applied to identify independent risk factors for postoperative early tumor recurrence. Correlation between independent risk factors and risk of postoperative prognosis was evaluated by restricted cubic splines (RCS) analysis, which was conducted using “rcssci” package. The optimal cutoff values were calculated by receiver operating characteristic (ROC) analysis. Subgroup analysis was performed with “jstable” package. SPSS (Version 26.0) and R (Version 4.5.0) were applied for data analysis. Figures were generated by R (Version 4.5.0). A two-tailed P < 0.05 was defined as statistically significant.

RESULTS

Overall condition of enrolled patients

216 patients were retrospectively included for analysis, including 116 males and 100 females. Patient selection process was illustrated in Figure 1. All patients were suspected of BRPC according to preoperative examination before surgery. Radical pancreatic resection including PD, DP and TP combined with vascular resection and reconstruction of invaded vessels were performed in all patients. All patients were pathologically diagnosed with pancreatic malignancy. Patients’ baseline characteristics were summarized in Table 1.

Figure 1 Flow diagram of patient selection BRPC: Borderline resectable pancreatic cancer.

Table 1 Baseline characters of included patients, n (%).

Variables	Results
Sex
Male	116 (53.7)
Female	100 (46.3)
Age (year), mean ± SD	61.3 ± 9.8
Smoking
Yes	59 (27.3)
No	157 (72.7)
Surgical procedure
Pancreaticoduodenectomy	132 (61.1)
Distal pancreatectomy	10 (4.6)
Total pancreaticoduodenectomy	74 (34.3)
Vascular reconstruction procedure
End-to-end anastomosis	59 (31.9)
Allogeneic vascular graft	147 (68.1)
Pathological diagnosis
Adenocarcinoma	190 (88.0)
Adenosquamous carcinoma	23 (10.6)
Mucinous adenocarcinoma	2 (0.9)
Sarcomatoid carcinoma	1 (0.5)
Postoperative adjuvant chemotherapy	118 (54.6)

Timing and sites of postoperative tumor recurrence after surgery

Our follow-up ended in December 2025 with a medium follow-up time of 49.0 [95% confidence interval (CI): 4.66-93.834] months. One hundred and fifty-seven patients developed tumor recurrence after surgery including 10 patients whose recurrence site was not reported during follow-up. The DFS rate of all included patients was 11 months, with 1-year, 3-year and 5-year DFS rates of 44.4%, 27.7% and 18.5%, respectively. Of these recurrences, 114 (72.6%), 26 (16.6%) and 17 (10.8%) occurred in year one, two and three after surgery. The incidence of newly developed tumor recurrence was highest in postoperative year one, reaching 52.8%. However, for patients who survived the first year and remained recurrence-free at the 1-year mark, the probability of developing a new recurrence during the second year dropped to 20.3%. Similarly, the probability of recurrence in the third year for patients recurrence-free at 2 years was 23.3%. These results indicated that postoperative year one was the period with the highest risk of tumor recurrence in BRPC (Figure 2A).

Figure 2 Timing and sites of postoperative tumor recurrence in borderline resectable pancreatic cancer. A: Proportion of newly developed recurrences among at-risk cases in different time intervals; B: Absolute number of newly developed recurrences at different sites in different time intervals; C: Proportion of newly developed site-specific recurrences among at-risk cases in different time intervals.

One hundred and eighteen patients were found to have tumor recurrence in a single organ while 29 patients developed multi-organ recurrence after excluding 10 patients whose recurrence sites were inconclusive. The specific site of tumor recurrence included liver metastasis (n = 101), local recurrence (n = 43), lymph node metastasis (n = 16), peritoneal metastasis (n = 12), lung metastasis (n = 10), and splenic metastasis (n = 1). Upon stratification based on the timing of tumor recurrence, liver metastasis predominated across all intervals, despite a gradual decrease in both case numbers and incidence. Similarly, lymph node and local recurrences were notably concentrated within the first year following surgery, after which a significant reduction in absolute count and incidence was observed. Conversely, the number of lung and peritoneal metastases occurring within one year after surgery was similar to that observed after one year, but their proportion showed a rising trend (Figure 2B and C).

Overall survival of BPRC patients

The median overall survival (OS) of enrolled patients was 15.0 months. Patients who developed postoperative recurrence had a significantly shorter median OS compared to those who did not (13 months vs 25 months; P < 0.001; Figure 3A). The median OS was similar between those with single-organ and multi-organ recurrence (13 months vs 14 months; P = 0.360; Figure 3B). When patients with single-organ recurrence were stratified by the site of recurrence, the median OS time was 37 months for peritoneal recurrence, 20 months for lung recurrence, 15 months for local recurrence, and 11 months for liver recurrence (P = 0.054; Figure 3C). Notably, OS for peritoneal recurrence was significantly longer than that for lung (P = 0.045) or liver recurrence (P = 0.011). The OS among the remaining subgroups didn’t reach statistic significance in this cohort.

Figure 3 Comparison of overall survival time of borderline resectable pancreatic cancer patients in different subgroups. A: Comparison of postoperative overall survival time in patients with and without tumor recurrence; B: Comparison of postoperative overall survival time in patients with single-organ and multi-organ recurrence; C: Comparison of post-recurrence survival time in patients with liver, local, lung and peritoneal recurrence.

Post-recurrence survival in BRPC patients

Although the OS of BRPC patients reached 15.0 months, the median post-recurence OS was only 4 months, indicating a poor prognosis after recurrence. Patients who developed multiple organ recurrence had comparable survival time after recurrence to patients whose tumor recurrence was localized in a single organ or tissue (P = 0.917, Figure 4A). Moreover, the post-recurrence survival times for local recurrence, peritoneal recurrence, liver recurrence and lung recurrence were 3 months, 4 months, 4 months, and 6 months, respectively, with no statistically significant differences (P = 0.347, Figure 4B). These results indicate that the post-recurrence survival had no correlation with tumor recurrence site.

Figure 4 Comparison of post-recurrence survival time of borderline resectable pancreatic cancer patients in different subgroups. A: Comparison of post-recurrence survival time in patients with single-organ and multi-organ recurrence; B: Comparison of post-recurrence survival time in patients with liver, local, lung and peritoneal recurrence; C: Comparison of post-recurrence survival time in patients with early and late tumor recurrence; D: Comparison of post-recurrence survival time in patients with early and late liver recurrence; E: Comparison of post-recurrence survival time in patients with early and late local recurrence.

The patients were then categorized into early and late recurrence group based on the time of recurrence and compared their post-recurrence survival. According to previous research, tumor recurrence within postoperative 1-year is usually defined as early postoperative tumor recurrence, thus we adopted this criterion to define early recurrence in our research[6,13,14]. Post-recurrence survival time in patients who developed recurrence after postoperative year one was 8 months, significantly exceeding that of those who developed tumor recurrence in postoperative year one (P < 0.001, Figure 4C). After categorizing patients according to their site of recurrence, it was found that patients who developed liver (P = 0.010, Figure 4D) and local recurrence (P < 0.001, Figure 4E) within the first year after surgery had significantly decreased post-recurrence survival than those whose tumor recurrence was after postoperative year one. The median post-recurrence survival of patients who developed liver and local recurrence after postoperative year one was 8 months and 9 months, respectively, much longer than patients who developed recurrence in postoperative year one, which was 4 and 3 months, respectively. This difference was not observed in patients with lymph node recurrence (P = 0.138), peritoneal recurrence (P = 0.200), and lung recurrence (P = 0.763). Collectively, these results confirmed that the time of recurrence, especially early tumor recurrence within one year, negatively influenced the post-recurrence survival of BRPC patients.

Potential risk factors for postoperative DFS

According to univariate Cox regression, preoperative CEA, CA19-9, imaging vascular morphology changes, length of vascular invasion, tumor diameter, tumor differentiation and lymph node metastasis were correlated with postoperative tumor recurrence (P < 0.05, Table 2). VIF of preoperative CEA, CA19-9, imaging vascular morphology changes, length of vascular invasion, tumor diameter, tumor differentiation and lymph node metastasis were 1.112, 1.171, 1.042, 1.100, 1.075, 1.282 and 1.230, respectively, indicating no severe multicollinearity. Multivariate Cox regression showed that preoperative CEA, tumor diameter, tumor differentiation and lymph node metastasis were independent risk factors for postoperative DFS (Figure 5A). After including preoperative total bilirubin into multivariate Cox analysis to adjust for potential impact of obstructive jaundice, these factors remained statistically significant, indicating the prognostic value of these factors was independent of obstructive jaundice (Supplementary Table 2). As competing risk existed between non-cancer death and tumor recurrence in the cohort, we also employed the Fine-Gray competing risks model for analysis. In univariate analysis, preoperative white blood cell count, neutrophil count, CEA, CA19-9, imaging vascular morphology changes, length of vascular invasion, tumor diameter, tumor differentiation and lymph node metastasis were significantly correlated with postoperative tumor recurrence (Table 2). After including these factors in multivariate analysis, only preoperative CEA and tumor diameter were independent risk factors for DFS (Figure 5B). Moreover, RCS analysis also showed a significant relationship between postoperative tumor recurrence and CEA as well as tumor diameter, confirming their correlation with postoperative DFS (Figure 5C and D). ROC analysis confirmed that the optimal cutoff value for preoperative CEA and tumor diameter were 3.285 ng/mL and 3.15 cm. Patients with CEA > 3.285 ng/mL had a 56.5% increased risk of postoperative tumor recurrence (hazard risk = 1.565, 95%CI: 1.140-2.147, P = 0.006), while patients whose tumor diameter was > 3.15 cm had a 57.5% increased risk of postoperative tumor recurrence (hazard risk = 1.575, 95%CI: 1.141-2.173, P = 0.006).

Figure 5 Risk factors for postoperative disease-free survival in borderline resectable pancreatic cancer. A: Results of multivariate Cox regression for postoperative disease-free survival; B: Results of multivariate Fine-Gray competing risk analysis for postoperative disease-free survival; C: Restricted cubic splines analysis of carcinoembryonic antigen and postoperative disease-free survival; D: Restricted cubic splines analysis of tumor diameter and postoperative disease-free survival. CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CI: Confidence interval; HR: Hazard risk; sHR: Subdistribution hazard ratio.

Table 2 Result of univariate Cox regression and Fine-Gray competing risk analysis for postoperative disease-free survival.

Variables	Cox regression			Fine-Gray competing risks model
	HR	95%CI	P value	sHR	95%CI	P value
Age (years)	1.016	0.999-1.032	0.058	1.011	0.996-1.027	0.154
Sex
Male	Reference	-	-	Reference	-	-
Female	1.005	0.770-1.445	0.738	0.952	0.706-1.284	0.748
Smoking history
No	Reference	-	-	Reference	-	-
Yes	1.113	0.783-1.583	0.550	1.013	0.726-1.416	0.937
Diabetes
No	Reference	-	-	Reference	-	-
Yes	0.938	0.665-1.322	0.714	0.936	0.672-1.304	0.697
BMI (kg/m2)	1.011	0.969-1.056	0.606	1.015	0.975-1.056	0.472
WBC count (× 109/L)	1.045	0.977-1.118	0.199	1.057	1.004-1.112	0.034a
NEUT count (× 109/L)	1.043	0.972-1.119	0.244	1.052	1.000-1.106	0.049a
LYM count (× 109/L)	1.162	0.855-1.578	0.338	1.222	0.914-1.634	0.176
Hemoglobin (g/L)	1.004	0.994-1.013	0.429	1.001	0.992-1.010	0.833
Platelet count (× 109/L)	1.000	0.998-1.002	0.989	1.000	0.999-1.003	0.566
Serum albumin (g/L)	0.992	0.957-1.029	0.682	0.995	0.960-1.031	0.764
Serum ALT (U/L)	1.000	0.998-1.001	0.965	1.000	0.999-1.002	0.803
Serum total bilirubin (μmol/L)	1.001	0.999-1.003	0.232	1.001	0.999-1.003	0.179
Serum CEA (ng/mL)	1.029	1.012-1.047	0.001a	1.031	1.020-1.041	< 0.001a
Serum CA19-9 (U/mL)	1.203	1.100-1.317	< 0.001a	1.150	1.039-1.272	0.007a
Biliary drainage
No	Reference	-	-	Reference	-	-
Yes	1.082	0.726-1.611	0.700	1.108	0.763-1.608	0.591
Arterial invasion
No	Reference	-	-	Reference	-	-
Yes	1.090	0.693-1.713	0.710	0.976	0.590-1.614	0.924
Vascular invasion site
PV	Reference	-	-	Reference	-	-
Confluence of PV and SMV	1.077	0.697-1.662	0.739	0.875	0.592-1.293	0.502
SMV	0.910	0.571-1.448	0.690	0.763	0.503-1.157	0.202
Imaging vascular morphology changes
No	Reference	-	-	Reference	-	-
Yes	1.463	1.066-2.009	0.018a	1.429	1.054-1.936	0.022a
Length of vascular invasion (cm)	1.114	1.013-1.226	0.026a	1.104	1.005-1.213	0.039a
Angle of vascular invasion
≤ 180	Reference	-	-	Reference	-	-
> 180	1.005	0.718-1.407	0.976	1.030	0.743-1.427	0.862
Surgical procedure
Pancreaticoduodenectomy	Reference	-	-	Reference	-	-
Distal pancreatectomy	1.297	0.627-2.680	0.483	1.137	0.571-2.265	0.716
Total pancreaticoduodenectomy	1.058	0.746-1.500	0.751	0.933	0.664-1.311	0.690
Vascular reconstruction procedure
End-to-end anastomosis	Reference	-	-	Reference	-	-
Allogeneic vascular graft	1.335	0.949-1.877	0.097	1.254	0.922-1.704	0.149
Operation time (hours)	1.025	0.973-1.080	0.350	1.008	0.957-1.061	0.773
Intraoperative blood loss (mL)	1.000	1.000-1.000	0.398	1.000	0.999-1.000	0.800
Intraoperative blood transfusion
No	Reference	-	-	Reference	-	-
Yes	1.087	0.793-1.489	0.604	1.034	0.768-1.391	0.827
Tumor site
Pancreatic head and uncinate process	Reference	-	-	Reference	-	-
Pancreatic neck	1.089	0.794-1.559	0.642	0.966	0.704-2.169	0.462
Pancreatic body and tail	1.340	0.715-2.511	0.361	1.235	0.681-1.371	0.848
Tumor differentiation
Low	Reference	-	-	Reference	-	-
Moderate and high	0.605	0.437-0.838	0.003a	0.704	0.505-0.981	0.038a
Tumor diameter (cm)	1.159	1.069-1.256	< 0.001a	1.146	1.074-1.223	< 0.001a
Lymph node metastasis
No	Reference	-	-	Reference	-	-
Yes	1.702	1.176-2.463	0.005a	1.430	1.015-2.014	0.041a
Postoperative adjuvant chemotherapy
No	Reference	-	-	Reference	-	-
Yes	0.735	0.536-1.007	0.055	0.876	0.645-1.189	0.396

^aP < 0.05.

HR: Hazard risk; sHR: Subdistribution hazard ratio; CI: Confidence interval; BMI: Body mass index; WBC: White blood cell; NEUT: Neutrophil; LYM: Lymphocyte; ALT: Alanine transaminase; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; PV: Portal vein; SMV: Superior mesenteric vein.

Potential risk factors for early postoperative recurrence

As early tumor recurrence was associated with poor post-recurrence survival, we further compared patients who developed tumor recurrence within one year with those who did not develop early tumor recurrence to find out potential risk factors for early postoperative recurrence. Preoperative absolute white blood cell count, neutrophil count, CEA, CA19-9, imaging vascular morphology changes, length of vascular invasion, tumor differentiation, tumor diameter and lymph node metastasis were significantly different in intergroup comparison (P < 0.05, Table 3). Further multivariate analysis in these variables identified preoperative CEA, tumor diameter, tumor differentiation and lymph node metastasis as independent risk factors for early postoperative recurrence (Figure 6A). These factors were statistically significant after adjusting for preoperative total bilirubin, confirming that the independent prognostic role of identified risk factors was not confounded by obstructive jaundice (Supplementary Table 3). RCS analysis confirmed a linear correlation between the risk of early tumor recurrence and preoperative CEA and tumor diameter, the higher the preoperative CEA level and large tumor size, the higher the risk of developing early tumor recurrence (Figure 6B and C). Further ROC analysis confirmed that 2.185 ng/mL and 3.9 cm were the optimal cutoff values for preoperative CEA and tumor diameter. Patients with preoperative CEA > 2.35 ng/mL and tumor diameter > 3.9 cm had a 2.435-fold (95%CI: 1.403-4.225, P = 0.002) and 2.222-fold (95%CI: 1.280-3.859, P = 0.005) higher risk of early tumor recurrence compared to the control group.

Figure 6 Risk factors for postoperative early tumor recurrence in borderline resectable pancreatic cancer. A: Results of multivariate logistic regression for postoperative early tumor recurrence; B: Restricted cubic splines analysis of carcinoembryonic antigen and early tumor recurrence; C: Restricted cubic splines analysis of tumor diameter and early tumor recurrence. WBC: White blood cell; NEUT: Neutrophil; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CI: Confidence interval; OR: Odds ratio.

Table 3 Comparison of general data of patients that developed tumor recurrence within and after postoperative 1-year, n (%).

Variables	Early recurrence group (n = 114)	Late recurrence group (n = 102)	P value
Age (years), median (Q1, Q3)	63.00 (57.00, 68.75)	60.50 (53.25, 66.75)	0.054
Sex			0.302
Male	65 (57.02)	51 (50.00)
Female	49 (42.98)	51 (50.00)
Smoking history			0.966
Yes	31 (27.19)	28 (27.45)
No	83 (72.81)	74 (72.55)
Diabetes			0.423
Yes	37 (32.46)	28 (27.45)
No	77 (67.54)	74 (72.55)
BMI (kg/m2), median (Q1, Q3)	23.14 (20.77, 24.97)	22.74 (20.82, 24.46)	0.529
WBC count (× 109/L), median (Q1, Q3)	5.80 (4.82, 7.07)	5.40 (4.37, 6.30)	0.024a
NEUT count (× 109/L), median (Q1, Q3)	3.62 (2.87, 4.65)	3.15 (2.59, 4.20)	0.023a
LYM count (× 109/L), median (Q1, Q3)	1.51 (1.17, 1.90)	1.39 (1.08, 1.80)	0.209
Hemoglobin (g/L), mean ± SD	123.47 ± 16.25	122.77 ± 16.68	0.756
Platelet count (× 109/L), median (Q1, Q3)	192.00 (164.00, 235.50)	190.00 (143.25, 248.75)	0.498
Serum albumin (g/L), mean ± SD	38.11 ± 4.31	38.77 ± 4.20	0.260
Serum ALT (U/L), median (Q1, Q3)	32.50 (16.00, 84.00)	40.00 (19.25, 97.00)	0.193
Serum total bilirubin (μmol/L), median (Q1, Q3)	15.85 (10.22, 111.20)	15.55 (9.62, 79.90)	0.597
Serum CEA (ng/mL), median (Q1, Q3)	3.25 (1.70, 5.80)	2.07 (1.40, 3.76)	0.003a
Serum CA19-9 (U/mL), median (Q1, Q3)	250.25 (39.23, 1000.73)	113.10 (28.63, 565.90)	0.020a
Biliary drainage			0.392
Yes	23 (20.18)	16 (15.69)
No	91 (79.82)	86 (84.31)
Vascular invasion site			0.961
PV	17 (14.91)	16 (15.69)
Confluence of PV and SMV	57 (50.00)	52 (50.98)
SMV	40 (35.09)	34 (33.33)
Imaging vascular morphology changes			0.014a
Yes	66 (57.89)	42 (41.18)
No	48 (42.11)	60 (58.82)
Length of vascular invasion (cm), median (Q1, Q3)	2.75 (2.15, 3.88)	2.40 (1.80, 3.27)	0.009a
Extent of vascular invasion			0.679
≤ 180	79 (69.30)	68 (66.67)
> 180	35 (30.70)	34 (33.33)
Arterial invasion			0.983
Yes	18 (15.79)	16 (15.69)
No	96 (84.21)	86 (84.31)
Surgical procedure			0.895
Pancreaticoduodenectomy	69 (60.53)	63 (61.76)
Distal pancreatectomy	6 (5.26)	4 (3.92)
Total Pancreaticoduodenectomy	39 (34.21)	35 (34.31)
Vascular reconstruction procedure			0.197
End-to-end anastomosis	32 (28.07)	37 (36.27)
Allogeneic vascular graft	82 (71.93)	65 (63.73)
Operation time (hours), median (Q1, Q3)	12.00 (11.00, 14.75)	13.00 (11.00, 15.00)	0.719
Intraoperative blood loss (mL), median (Q1, Q3)	750.00 (500.00, 1000.00)	600.00 (400.00, 1000.00)	0.372
Intraoperative blood transfusion			0.158
Yes	69 (60.53)	52 (50.98)
No	45 (39.47)	50 (49.02)
Tumor site			0.939
Pancreatic head and unicorn process	70 (61.40)	65 (63.73)
Pancreatic neck	37 (32.46)	31 (30.39)
Pancreatic body and tail	7 (6.14)	6 (5.88)
Tumor differentiation			0.003a
Low	54 (47.37)	28 (27.45)
Moderate and high	60 (52.63)	74 (72.55)
Tumor diameter (cm), median (Q1, Q3)	4.00 (3.00, 5.00)	3.00 (2.50, 4.00)	0.002a
Lymph node metastasis			0.001a
Yes	94 (82.46)	64 (62.75)
No	20 (17.54)	38 (37.25)
Postoperative adjuvant chemotherapy			0.242
Yes	58 (50.88)	60 (58.82)
No	56 (49.12)	42 (41.18)

^aP < 0.05.

BMI: Body mass index; WBC: White blood cell; NEUT: Neutrophil; LYM: Lymphocyte; ALT: Alanine transaminase; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; PV: Portal vein; SMV: Superior mesenteric vein.

Influence of vascular invasion on postoperative tumor recurrence

BRPC patients with different extents of vascular invasion, especially in the portomesenteric vein, received different reconstruction procedures, but their impact on postoperative recurrence is unknown. Therefore, we evaluated the correlation between portomesenteric vascular invasion extent and postoperative recurrence. Imaging vascular morphology changes were correlated with decreased DFS and higher early recurrence risk, while length of vascular invasion was positively correlated with decreased DFS (Figure 7A and B). Given the close anatomical relationship between the portomesenteric venous axis and liver, we further analyzed the correlation between vascular invasion and liver recurrence. Univariate analysis indicated no correlation between vascular invasion status and postoperative liver recurrence (Figure 7C and D). We then applied to subgroup analysis to reveal the influence of vascular invasion extent on tumor recurrence across different patterns of invasion and reconstruction procedures (Supplementary Figures 2-5). Among patients who underwent end-to-end anastomosis, those with imaging vascular morphology changes had significantly worse DFS and a higher risk of liver recurrence (P < 0.05). Furthermore, a greater length of vascular invasion was significantly correlated with reduced DFS in patients with portal vein- superior mesenteric vein confluence invasion (P = 0.019), and an increased risk of early recurrence in patients with superior mesenteric vein invasion (P = 0.019). These results collectively indicated that the extent of portomesenteric vascular invasion had a limited effect on postoperative tumor recurrence generally, especially on liver recurrence. However, in patients undergoing end-to-end anastomosis, imaging vascular morphology changes may identify a subgroup at higher risk of tumor recurrence.

Figure 7 Impact of vascular invasion extent on postoperative recurrence. A: Impact of vascular invasion extent on disease-free survival; B: Impact of vascular invasion extent on liver-recurrence free survival; C: Impact of vascular invasion extent on early tumor recurrence; D: Impact of vascular invasion extent on early liver recurrence. HR: Hazard ratio; CI: Confidence interval; PV: Portal vein; SMV: Superior mesenteric vein; OR: Odds ratio.

DISCUSSION

In this retrospective study of BRPC patients undergoing upfront surgery, we found that 72.7% of patients experienced postoperative recurrence, with significantly decreased OS. Postoperative tumor recurrence mainly occurred within the first year after surgery and liver was the predominant site of recurrence throughout the entire postoperative period. Furthermore, preoperative CEA, tumor diameter, differentiation, and lymph node metastasis were independently correlated with both DFS and early postoperative recurrence. In addition, subgroup analysis further demonstrated the limited impact of portomesenteric vascular invasion on DFS and early tumor recurrence. As far as we know, this was the first research that systemically reported the pattern and risk factors for postoperative recurrence in BRPC patients receiving upfront surgery. These findings enhanced the understanding of tumor recurrence in this patient population and may help clinicians tailor treatment strategies to improve prognosis.

It was observed that more than half of patients who survived surgery developed recurrence in the first year, and recurrence in the first year accounted for 72.6% of all recurrences. The observed time pattern of tumor recurrence was similar to that in resectable pancreatic cancer and BRPC after neoadjuvant therapy. Brunner et al[15] found that 61.7% of recurrences were observed in the first year after surgery in a resectable pancreatic cancer patients cohort, and Groot et al[7] further broadened this finding in BRPC patients after neoadjuvant chemotherapy. Interestingly, both of these studies observed a significant decreasing risk of newly developed recurrence with increasing duration of follow-up, which was also similar to our findings[7,15]. Therefore, we believe that the time patterns of postoperative tumor recurrence are similar among all resectable pancreatic cancers and BRPC regardless of whether receiving neoadjuvant chemotherapy, and postoperative year one is the highest-risk period for recurrence. Given the aggressive biological behavior of pancreatic cancer, micro or occult metastasis may already occur in distant organs which cannot be resected even after R0 resection, and can quickly develop into recurrences, partly explaining this time pattern of recurrence[16]. In addition, the current assessment of resectability in pancreatic cancer primarily relies on local tumor features without capturing the systemic tumor burden or the biological aggressiveness of the disease, partly explaining the similar time pattern of recurrence across pancreatic cancers with different resectability. This finding indicates the necessity of conducting a more frequent follow-up schedule during postoperative year one and the feasibility of reducing subsequent follow-up frequency as performed in resectable pancreatic cancer.

In terms of the pattern of tumor recurrence site, single-organ recurrence was significantly greater than multi-organ recurrence in our cohort, and liver recurrence constituted the majority of recurrences followed by local recurrence. Interestingly, the characteristics of recurrence site in our cohort was contrary to BRPC patients after neoadjuvant chemotherapy according to previous research. Following neoadjuvant chemotherapy in BRPC patients, the majority of recurrences were multi-organ recurrences, while local recurrence was emerged as the dominant site of recurrence[7]. The influence of neoadjuvant chemotherapy and distinct surgical procedures may explain this discrepancy. Most patients who had multi-organ recurrence developed both local and distant recurrence in Groot’s cohort. However, all BRPC patients received extended lymphadenectomy removing nearly all lymph nodes and fat tissue in front of the abdominal aorta and inferior vena cava, thereby leading to a decreased local and multi-organ recurrence rate by diminishing the local recurrence rate. Our surgical procedure may also have caused underestimation of local recurrence and lymph node recurrence rates, necessitating further validation in further research. In addition, the relatively lower proportion of liver recurrence may be attributed to the application of neoadjuvant chemotherapy as it can effectively deplete existing micrometastasis in the liver[17]. These findings indicated that the pattern of recurrence was different in patients with and without neoadjuvant chemotherapy. Clinicians should pay more attention to the liver especially in postoperative year one in order to enable early detection of potential recurrence. Furthermore, since magnetic resonance imaging (MRI) have higher resolution and diagnostic ability in liver metastasis, we consider that incorporating abdominal MRI into the surveillance protocol may facilitate earlier detection of recurrence and improve the prognosis of these patients[18].

Since postoperative tumor recurrence adversely affects patient survival, we further investigated the potential effect of different recurrence patterns on OS and post-recurrence survival. In resectable pancreatic cancer, previous confirmed that both multi-organ recurrence and liver-specific recurrence could significantly impact postoperative outcomes[19,20]. However, this finding is not always found in BRPC. Postoperative OS and post-recurrence survival were comparable in patients who had single and multi-organ recurrence in our cohort. Furthermore, although patients who developed peritoneal recurrence had prolonged postoperative survival, this finding was not conclusive due to the relatively limited number of peritoneal recurrences in our research. Therefore, we believe that the site of tumor recurrence had little impact on survival and should be paid equal attention when observed in clinical practice. With regard to the reason for this discrepancy in relation to previous research, we attribute this to the more aggressive biological behavior in BRPC patients compared to resectable lesions. Single-organ recurrence may already be a sign of systemic and uncontrolled disease status in this aggressive disease, and may also indicate the existence of potential micrometastasis that cannot be detected using current imaging techniques, making the survival of these patients similar to those with multi-organ recurrence.

In contrast, time of recurrence appeared to be a more important factor impacting postoperative survival in our cohort, consistent with previous research. According to Ono et al[21], the OS of patients who developed early postoperative recurrence was only 8.6 months, significantly lower than that of patients who developed late tumor recurrence. Zhang et al[20] further confirmed the negative impact of early tumor recurrence on post-recurrence survival time. This finding emphasized the importance of identifying patients with higher risk of early postoperative recurrence. Therefore, we further determined the risk factors for tumor recurrence in our cohort. Previous research have identified a series of potential risk factors for DFS including pathological, laboratory and imaging indices. In that study, imaging features including larger tumor diameter, and extrapancreatic organ infiltration were confirmed as important preoperative indices predicting postoperative DFS[22]. In addition, medical history (acute pancreatitis, diabetes), perioperative conditions (postoperative pancreatic fistula) and pathological findings (lymph node metastasis, R1 resection, tumor location) were also reported as risk factors for postoperative and early tumor recurrence[15,23,24]. As all these factors were closely correlated with poor tumor biological behavior, their predictive value for postoperative tumor recurrence is reasonable. Similarly, we found that CEA, tumor diameter, lymph node metastasis and tumor differentiation were independent risk factors for recurrence, further broadening the predictive value of these identified risk factors to BRPC patients after upfront surgery. Since CEA, tumor diameter and lymph node metastasis can be obtained from preoperative laboratory test and CT scan, these finding enables early prediction of postoperative recurrence in these patients. The proposed the optimal cutoff value of CEA and tumor diameter in predicting DFS and early recurrence also improved its clinical application value. Clinicians should cautiously balance the pros and cons before performing upfront surgeries in those patients with these risk factors, thereby reducing the unnecessary surgery that impair patients’ quality of life without offering a survival benefit. A comprehensive prediction model incorporating risk factors can further improve the predictive accuracy and may be a potential direction for further research.

Currently, CA19-9 is widely adopted as an effective biomarker in pancreatic cancer and serves as an effective predictor of postoperative recurrence[25,26]. However, we found that preoperative CEA also had predictive value for tumor recurrence in our cohort. CEA is a widely used biomarker in clinical practice and can predict postoperative postoperative recurrence in pancreatic cancer regardless of its resectability[27,28]. In a clinical study involving 260 pancreatic patients, CEA > 4.45 ng/mL emerged as a significant predictors of early postoperative recurrence[29]. Kato et al[30] also showed similar findings in localized pancreatic cancer, confirming that elevated CEA was the most significant independent risk factor for decreased postoperative DFS. All localized pancreatic cancer patients whose preoperative CEA exceeded 7.2 ng/mL developed tumor recurrence within the first year after surgery, and their median DFS time was only 5.4 months[30]. Our findings reinforced and broadened the applicability of CEA as an effective index in postoperative recurrence prediction in BRPC. However, whether neoadjuvant chemotherapy or upfront surgery can better benefit BPRC patients with elevated CEA levels remains unclear. Jacover et al[31] found that elevated CEA level at diagnosis could effectively predict the failure of bridging to surgical treatment after neoadjuvant chemotherapy mainly due to its correlation with local tumor progression. Notably, they also found that such tumor progression did not lead to early tumor metastasis at least in the first 6 months after diagnosis. Thus, they proposed upfront surgery as the optimal treatment for these patients instead of neoadjuvant chemotherapy to avoid losing chance for surgery[31]. However, patients with elevated CEA who received upfront surgery still faced a higher risk of early postoperative recurrence and shorter DFS in our cohort, indicating that upfront surgery may not be beneficial in all such patients. Therefore, it is crucial to combine CEA with other markers for optimal treatment decisions in current clinical practice. Future studies are warranted to explore optimal treatment plan for these patients to improve their clinical outcomes.

According to previous research, whether the extent of vascular invasion impact postoperative recurrence in BRPC patients remained controversial. Some researchers reported that extended vascular invasion, such as increased tumor-vein circumferential interface, had close correlation with decreased postoperative DFS, while others found that DFS was mainly determined by oncological characteristics instead of vascular invasion extent[32-34]. According to our findings, although increased length of vascular invasion and imaging vascular morphology changes tend to result in decreased DFS and higher early recurrence risk, none of these factors emerged as independent risk factors in multivariate analysis. These factors were also not statistically significant when predicting liver recurrence. Therefore, our result further confirmed that the extent of vascular invasion had a limited impact on postoperative recurrence. Interestingly, subgroup analysis showed that significant vascular morphology changes on CT significantly increased the risk of recurrence and liver-specific recurrence in patients receiving end-to-end anastomosis intraoperatively. Significant vascular morphology changes may indicate a higher risk of transmural invasion, increasing the risk of dissemination. Furthermore, morphological changes also correlate with a tighter tumor-vein adhesion, increasing the technical difficulty of vascular dissection and reconstruction. The prolonged reconstruction time during portomesenteric vascular reconstruction exacerbates hepatic ischemia, thereby promoting a pro-metastatic microenvironment in the liver[35]. Thus, significant vascular morphology changes may be a potential risk factor for recurrence in patients receiving end-to-end anastomosis and requires further attention.

The clinical implications of this study are multifaceted, offering potential guidance for clinical management of these patients. Preoperatively, clinicians can stratify patients with recurrence risk based on those identified risk factors and choose the patients may benefits from upfront surgery. For those with low risk of recurrence, extended vascular invasion do not increase postoperative recurrence risk and should not be regarded as a surgical contraindication as long as it can be safely reconstructed. For those with high risk of recurrence, palliative care may be a more appropriate treatment option compared to upfront surgery. Intraoperatively, reconstruction with vascular grafts may benefits patients for those with vascular morphology changes in imaging examination. Postoperatively, adopting a more intensive surveillance protocol and liver-targeted imaging method like MRI during the first postoperative year may enable earlier detection of tumor recurrence, while reducing the follow-up frequency thereafter is also feasible. Our research helps clinicians to better tailor treatment strategies and follow-up schedule and can better improve the prognosis of BRPC patients after upfront surgery.

This study has several limitations. First, the identified risk factors require further validation in multi-center cohorts to ensure their generalizability. Second, the long enrollment period from 2011 to 2024 may have introduced chronological bias due to evolving surgical techniques, perioperative care and systemic oncologic treatments, which may potentially influence DFS and the recurrence pattern over time. Third, our exclusion criteria may also have resulted in potential bias in estimating recurrences. As R1/R2 resections and perioperative death may indicate advanced tumor stage that requires extended surgery, excluding these patients may underestimate the total recurrence burden, introducing potential selection and survival bias. Fourth, the small number of cases with lung and peritoneal recurrence restricted our ability to accurately characterize these specific patterns. Fifth, all patients with lymph node recurrence developed multi-organ recurrence, therefore DFS and post-recurrence survival could not be clearly evaluated in our cohort, which requires further investigation. Sixth, this study primarily relied on traditional clinical and pathological parameters without integrating molecular and genomic data, limiting our ability to evaluate the biological heterogeneity that might drive early recurrence beyond traditional clinical markers. Finally, due to incomplete records on postoperative adjuvant chemotherapy and post-recurrence salvage therapies, especially time to treatment initiation, regimen, course and reason for omission, we were unable to evaluate their impact on survival. Further prospective research incorporating larger cohorts and molecular profiling is necessary to address these issues.

CONCLUSION

BRPC patients who receive upfront surgery tend to have the highest risk of tumor recurrence in the 12-month period after surgery, and liver recurrence is the most common site of recurrence followed. Preoperative CEA, tumor diameter, tumor differentiation and lymph node metastasis were confirmed as independent risk factors for postoperative DFS and early postoperative recurrence.