Association Between Pancreatic Fistula and Long-term Survival in the Era of Neoadjuvant Chemotherapy

Key Points

Question  Have the frequency and long-term implications of clinically relevant pancreatic fistula after resection for pancreatic cancer changed in the era of neoadjuvant therapy?

Findings  In this cohort study of 346 patients receiving neoadjuvant therapy of overall 753 individuals with resected pancreatic cancers, the occurrence of clinically relevant pancreatic fistula was markedly reduced (3.6 times) compared with upfront resections, and classic clinically relevant pancreatic fistula factors—except for soft pancreatic texture—are no longer applicable. Yet, if clinically relevant pancreatic fistula occurs in this setting, long-term overall survival may be impaired.

Meaning  Despite being uncommon, clinically relevant pancreatic fistula following neoadjuvant therapy for pancreatic cancer may have a major and independent association with long-term survival, and patients might benefit from closer follow-up and further systemic interventions.

Importance  In the past decade, the use of neoadjuvant therapy (NAT) has increased for patients with borderline and locally advanced pancreatic ductal adenocarcinoma (PDAC). Data on pancreatic fistula and related overall survival (OS) in this setting are limited.

Objective  To compare postoperative complications in patients undergoing either upfront resection or pancreatectomy following NAT, focusing on clinically relevant postoperative pancreatic fistula (CR-POPF) and potential associations with OS.

Design, Setting, and Participants  This retrospective cohort study was conducted on data from patients who underwent pancreatic resection for PDAC at the Massachusetts General Hospital from January 1, 2007, to December 31, 2017,

Exposures  Pancreatic cancer surgery with or without NAT.

Main Outcomes and Measures  Overall morbidity and CR-POPF rates were compared between NAT and upfront resection. Factors associated with CR-POPF were assessed with univariate and multivariate analysis. Survival data were analyzed by Kaplan-Meier curves and a Cox proportional hazards regression model.

Results  Of 753 patients, 364 were men (48.3%); median (interquartile range) age was 68 (61-75) years. A total of 346 patients (45.9%) received NAT and 407 patients (54.1%) underwent upfront resection. At pathologic examination, NAT was associated with smaller tumor size (mean [SD], 26.0 [15.3] mm vs 32.7 [14.4] mm; P < .001), reduced nodal involvement (102 [25.1%] vs 191 [55.2%]; P < .001), and higher R0 rates (257 [74.3%] vs 239 [58.7%]; P < .001). There were no significant differences in severe complication rate or 90-day mortality. The rate of CR-POPF was 3.6-fold lower in patients receiving NAT vs upfront resection (13 [3.8%] vs 56 [13.8%]; P < .001). In addition, factors associated with CR-POPF changed after NAT, and only soft pancreatic texture was associated with a higher risk of CR-POPF (38.5% vs 6.3%; P < .001). Survival analysis showed no differences between patients with or without CR-POPF after upfront resection (26 vs 25 months; P = .66), but after NAT, a worse overall survival rate was observed in patients with CR-POPF (17 vs 34 months; P = .002). This association was independent of other established predictors of overall survival by multivariate analysis (hazard ratio, 2.80; 95% CI, 1.44-5.45; P < .002).

Conclusions and Relevance  Neoadjuvant therapy may be associated with a significant reduction in the rate of CR-POPF. In addition, standard factors associated with CR-POPF appear to be no longer applicable following NAT. However, once CR-POPF occurs, it is associated with a significant reduction in long-term survival. Patients with CR-POPF may require closer follow-up and could benefit from additional therapy.

In the past decade, the use of neoadjuvant treatment (NAT) has markedly increased for patients with borderline resectable and locally advanced pancreatic ductal adenocarcinoma (PDAC).¹ This increase has enabled a large number of patients whose tumors were considered inoperable in the past to undergo pancreatic resection,² and several studies have shown encouraging surgical outcomes,^2-4 as well as improvement in pathologic findings.^5,6

Postoperative pancreatic fistula (POPF) remains one of the most frequent and dangerous complications after pancreatic surgery.⁷ In the setting of upfront resection, POPF rates vary between 5% and 15% for proximal resections and are higher after distal pancreatectomy.^8-10 Clinically relevant grade B and C fistulas (CR-POPF)^11,12 in turn are associated with a significant increase of postpancreatectomy hemorrhage, severe infectious complications, and mortality rates.^13-15 Several risk factors have been described to estimate the risk of POPF, such as soft pancreatic texture, small pancreatic duct diameter, and high intraoperative blood loss.^16,17 However, these data have been validated almost exclusively in the context of upfront resection, and studies specifically addressing rate and factors associated with POPF in NAT cohorts are lacking.

The aim of this study was to assess the rate of CR-POPF in the era of NAT at our institution. The relevance of classic CR-POPF in patients receiving NAT was also assessed, as well as whether CR-POPF–related morbidity affected long-term outcomes after pancreatic cancer surgery.

Adult patients who underwent a pancreatoduodenectomy or a distal pancreatectomy over an 11-year period (January 1, 2007, to December 31, 2017) at Massachusetts General Hospital were identified from a prospectively maintained institutional database. All patients with pathologic diagnosis of PDAC who underwent either neoadjuvant therapy and subsequent resection or upfront resection were included. Exclusion criteria were evidence of distant metastasis; PDAC arising from intraductal papillary mucinous neoplasms; other benign or malignant tumor diseases of the pancreas, including neuroendocrine carcinoma, acinar cell carcinoma, and distal cholangiocarcinoma; and total pancreatectomies. The study protocol was approved by the institutional review board at the Massachusetts General Hospital, which waived informed patient consent because of deidentification of data.

For most patients, FOLFIRINOX (folinic acid, fluorouracil, irinotecan, and oxaliplatin) was administered as the standard of care in those with locally advanced and borderline resectable PDAC, either followed, or not, by chemoradiotherapy with 50.4 Gy photon or 25 Gy proton radiotherapy.² In addition, some clinical trials were ongoing during the study period, including different NAT regimens, such as short-course proton beam radiation with capecitabine or fluorouracil.¹⁸ Also included were patients who were referred to Massachusetts General Hospital after receiving different NAT regimens at outside institutions.

Patients with locally advanced and borderline resectable PDAC that did not progress during NAT were taken to the operating room for open exploration and surgical resection.¹⁹ A standard lymphadenectomy was performed according to the International Study Group on Pancreatic Surgery guidelines, including peripancreatic lymph-node stations as well as lymph-node clearance of the tumor facing the side of the superior mesenteric artery or the celiac trunk.²⁰ Frozen sections were taken routinely from tumor-surrounding areas and additional vascular resections were performed when necessary, with the goal of achieving an R0 resection. Surgical drains were routinely left in all patients.

For patients with upfront resection, a gemcitabine-based standard adjuvant chemotherapy was routinely administered.²¹ In addition, most patients who underwent upfront resection received radiotherapy as well, especially if an R1 resection was noted on the final pathologic report. After NAT, with the exception of the subgroup of patients who received only short-course chemoradiotherapy, there was no general recommendation for additional adjuvant chemotherapy, and most patients were referred for routine oncologic surveillance.

Patients’ clinicopathologic data were collected retrospectively from a prospective maintained database and from electronical medical records, including demographics, body mass index (BMI), preoperative (post-NAT) carbohydrate antigen (CA) 19-9 levels, placement of preoperative biliary stent, type and extent of surgery, intraoperative pancreatic texture and main pancreatic duct diameter, estimated intraoperative blood loss, intraoperative radiotherapy, and duration of surgery. Postoperative outcomes and treatment included the occurrence of any complications graded according to the Clavien-Dindo classification, 30-day and 90-day mortality, duration of postoperative hospital stay, readmission rate, and receipt of postoperative radiotherapy and/or chemotherapy. Postoperative pancreatic fistula was defined according to the updated International Study Group on Pancreatic Surgery guidelines into biochemical leakage if any measurable drain amylase was higher than 300 U/L after postoperative day 2, or grade B or C fistula depending on the duration of drainage (≥3 weeks), the clinical outcome of POPF, or the type of reintervention.¹² Pathologic data included tumor size, tumor stage according to the 8th edition of the TNM classification,²² R-status based on a 1-mm tumor-free margin,²³ the presence of perineural invasion, and lymphovascular invasion.

Follow-up was updated until August 31, 2018, through the medical records or phone interviews. Patients were followed up from the date of operation until their last oncologic surveillance or death.

Dichotomous data are expressed as absolute numbers (relative percentages) and continuous data as mean (SD) or median and interquartile range (IQR). Continuous variables were assessed for normal distribution with the Shapiro-Wilk test. The t test or the Mann-Whitney test was used to compare parametric and nonparametric continuous variables, respectively. Nonrandom association for categorical variables was analyzed with the Fisher exact χ² test.

The survival analysis was performed with the Kaplan-Meier method, and patients who did and did not experience CR-POPF were compared using the log-rank (Mantel-Cox) test. If a death was not reported during the follow-up period, patients were censored at the last available contact date. To assess the independent association between the occurrence of CR-POPF and overall survival (OS), a Cox proportional hazards regression model was constructed. Hazard ratios (HRs) are reported with 95% CIs. Positive resection margins, size of the tumor, degree of differentiation, nodal status, CA19-9 levels and receipt of adjuvant treatment were included in the multivariate model, together with the occurrence and severity of CR-POPF.

For each test, a 2-sided P value of .05 was considered significant. All statistical computations were performed using IBM SPSS, version 25 (IBM Corp).

A total of 1102 patients underwent pancreatic surgery for primary pancreatic malignant tumors between 2007 and 2017 at the Massachusetts General Hospital. Of these patients, 342 individuals (31.0%) met the exclusion criteria and were removed from this study. Seven patients (<1%) were excluded owing to a completion pancreatectomy after a prior pancreatic head resection or distal pancreatectomy. Among the 753 patients who were available for the analysis with confirmed PDAC at pathologic examination, 364 were men (48.3%); median (interquartile range [IQR]) age was 68 (61-75) years. A total of 346 patients (45.9%) received NAT and 407 patients (54.1%) underwent upfront resection without any additional therapy prior to surgery. The median follow-up was 16 (IQR, 8-30) months for the entire cohort. Patient baseline characteristics and pathologic findings are summarized in Table 1. Patients with NAT compared with the upfront resection cohort were younger (66 [IQR, 59-72] vs 70 [IQR, 62-77] years; P < .001), had a lower mean (SD) BMI (calculated as weight in kilograms divided by height in meters squared) (25.8 [5.1] vs 26.9 [6.4]; P = .01) and had lower preoperative CA19-9 levels (median [IQR], 40 [ 13-139] vs 97 [23-403] U/mL; P < .001). Furthermore, patients receiving NAT were more frequently associated with favorable posttreatment pathologic characteristics, including smaller tumor size (mean [SD], 26.0 [15.3] vs 32.7 [14.4] mm; P < .001) or lower pT category (pT<3, 126 [36.4%] vs 32 [7.8%]; P < .001), negative lymph node status (191 [55.2%] vs 102 [25.2%]; P < .001), and higher R0 rates (257 [74.3%] vs 239 [58.7%]; P < .001). In contrast, patients who received NAT showed a higher frequency of preoperative biliary stenting (225 [65%] vs 234 [57.5%]; P = .04), longer operative times (mean [SD], 382 [111] vs 331 [118] minutes; P < .001), higher estimated intraoperative blood loss (median [IQR], 600 [400-1000] vs 500 ml [300-900] mL; P < .001), as well as a higher rate of vascular resections (71 [20.5%] vs 45 [11.1%]; P < .001). Among the NAT subgroup, most patients received FOLFIRINOX in combination with radiotherapy (185 [53.5%]); 125 patients (36.1%) received short-course chemoradiotherapy with concurrent capecitabine or fluorouracil, and the rest received gemcitabine-based treatment (25 [7.2%]) or other regimens (11 [3.2%]).

No significant differences in the overall complication, major complication (Clavien-Dindo class ≥3), and 30-day and 90-day mortality rates were detected between the upfront resection and NAT groups (Table 2). Biochemical leakage was significantly reduced in patients with NAT compared with the upfront resection group (28 [8.1%] vs 78 [19.2%]; P < .001) as well as the rate of CR-POPF in the NAT group compared with the upfront resection group (13 [3.8%] vs 56 [13.8]; P < .001). Stratified by the type of surgery, patients undergoing distal pancreatectomy in the upfront resection group had a biochemical leakage rate of 26.1%, and in those undergoing pancreatoduodenectomy, the rate was 17.2% (P = .10). The comparative rates in the NAT cohort were 23.0% and 4.9% (P < .001). Regarding CR-POPF, the frequencies were 15.9% after distal pancreatectomy and 13.2% after pancreatoduodenectomy in the upfront resection group and 3.3% and 3.9% after distal pancreatectomy and pancreatoduodenectomy, respectively, in the NAT group.

In the NAT cohort, we detected a lower prevalence of soft pancreatic texture compared with the upfront resection cohort (26 [7.5%] vs 53 [13.0%]; P = .005). A lower rate of CR-POPF in the FOLFIRINOX cohort was observed in comparison with short-course chemoradiotherapy (2.7% vs 5.6%), although this finding was not statistically significant (eFigure in the Supplement). The intra-abdominal drains were removed earlier (median [IQR], 4 [2-5] vs 5 [4-7] days; P < .001), and the median (IQR) postoperative hospital stay was significantly shorter in the NAT group compared with the upfront resection cohort (6 [6-8] vs 7 [6-9]; days P = .001).

Factors associated with the occurrence of CR-POPF were analyzed for the NAT and the upfront resection groups (Table 3). In the cohort of upfront resection, patients with high BMI (mean [SD], 28.7 [ 5.6] vs 26.6 [6.5]; P = .03), long duration of surgery (mean [SD], 401 [154] vs 320 [107] minutes; P < .001), small duct size (median [IQR], 3 [2-4] vs 4 [3-5] mm; P < .001), and high intraoperative estimated blood loss (median [IQR], 800 [ 500-1500] vs 500 [300-800] mL; P < .001) were more likely to develop a CR-POPF. None of these variables was associated with a higher incidence of CR-POPF in the NAT group, while the presence of soft pancreatic tissue significantly increased the risk of CR-POPF (38.5% vs 6.3%; P < .001).

The occurrence of CR-POPF was associated with OS in both groups, using Kaplan-Meier curves (Figure). In the upfront resection group, no association of CR-POPF with OS was observed, with a median OS of 25 months in the nonfistula group compared with 26 months in patients with CR-POPF (P = .66). By contrast, patients in the NAT group experienced a significantly worse OS when a CR-POPF occurred, with a median OS of 34 months in patients without fistula compared with 17 months in patients with a CR-POPF (P = .002). The unfavorable association of CR-POPF with OS was independent of the occurrence of severe complications (HR, 2.08; 95% CI, 1.07-4.04; P = .03).

With use of a Cox proportional hazards regression model (Table 4), CR-POPF was shown to be a possible independent prognostic factor for OS (HR, 2.80; 95% CI, 1.44-5.45; P < .002), together with other well-established prognostic parameters including tumor size OS (HR, 1.02; 95% CI, 1.01-1.03; P < .001), lymph node involvement (for N0 vs N1: HR, 1.61; 95% CI, 1.07-2.42; P = .02 and N2 vs N0: HR, 2.05; 95% CI, 1.19-3.55; P = .01), R1 status (HR, 1.88; 95% CI, 1.28-2.76; P = .001), administration of adjuvant treatment (HR, 0.62; 95% CI, 0.42-0.93; P = .02), and high CA19-9 levels (HR, 2.74 U/mL; 95% CI, 1.70-4.42 U/mL; P < .001).

In the past decade, neoadjuvant treatment has increased for borderline resectable and locally advanced PDAC, yet its associations with postoperative outcomes remain largely unexplored. Although some studies have evaluated different NAT regimens and postoperative morbidity, definite conclusions cannot be drawn, as most observations come from small studies, including heterogeneous NAT regimens and without comparison with upfront resection groups.²⁴ Larger cohort analyses have shown similar morbidity rates between NAT patients and primary resections,^3,6,25 while previous data reported a significantly lower complication rate with FOLFIRINOX than upfront resections.⁵ The present study comparing 346 patients receiving NAT with a contemporaneous cohort of 407 patients with upfront resection of PDAC is consistent with prior reports. We observed an overall morbidity rate of 52% for the NAT group vs 56% in the upfront resection group, with a rate of severe complications of 14% and 17%, respectively. In addition, the length of hospital stay was generally 1 day shorter in patients receiving NAT, despite higher rates of intraoperative bleeding and longer operative time than patients with upfront resections. Moreover, a significantly reduced rate of pulmonary complications, postoperative collections, and procedure-related complications was detected in the NAT group.

Neoadjuvant therapy has been generally associated with a reduction in POPF occurrence, although reported POPF rates broadly range between 0% and 15%.^26-31 Reasons for this variability may reside in the heterogeneity of neoadjuvant chemotherapy and/or radiotherapy schemes and the lack of a standardized CR-POPF definition across the studies. In our cohort, the occurrence of CR-POPF was 3.6 times lower in patients who received NAT compared with upfront resection, with rates of 3.8% vs 13.8%, respectively. These findings could seem contradictory, as patients who receive NAT experience longer operations, increased blood loss, and a higher rate of vascular resections, all of which have been associated in the past with an increased risk of CR-POPF.³² However, in the NAT group, we also observed a lower prevalence of soft pancreatic parenchyma, which historically is one of the major factors associated with CR-POPF. It has been described that NAT drives pathologic changes in the pancreatic gland, resulting in increased fibrosis and atrophy, which affects not only the neoplastic tissue, but also the remnant pancreas.^33,34 Arguably, this protective factor successfully overruled the increased risk associated with the intraoperative POPF factors. We also found a different CR-POPF rate according to the type of NAT; specifically, the FOLFIRINOX regimen was associated with the lowest risk of POPF (2.7%) and the short-course chemoradiotherapy had the highest risk (5.6%), suggesting that duration and type of NAT have different effects on the pancreas.

Several fistula risk measures have been used in the past, with the one most often applied described by Callery et al,¹⁶ although validation studies failed to confirm high blood loss as a factor associated with the occurrence of CR-POPF.^35,36 An alternative fistula risk score was published in 2019, which includes BMI, pancreatic texture, and pancreatic duct size.³⁷ In our cohort of patients who underwent upfront resection, we validated some of these well-described factors associated with CR-POPF, such as high BMI, small pancreatic duct, high estimated intraoperative blood loss, and duration of surgery.³⁸ However, we also found that these CR-POPF factors were no longer applicable after NAT, and that only soft pancreatic texture was independently associated with the occurrence of fistula. It can be argued that a firm pancreas may act as a surrogate marker for the response to NAT and a loss of that marker translates into a CR-POPF. Whether soft pancreatic texture is intrinsically associated with the gland or reflects a nonresponsive disease, and consequently a patient condition leading to higher risk of complication, is unclear.

To our knowledge, this is the first study specifically addressing whether CR-POPF is associated with patient long-term outcome after NAT. We found that the OS rate was halved in patients who experienced a CR-POPF compared with patients with no fistula. This outcome was also shown in multivariate analysis, where CR-POPF was an independent prognostic variable of worse survival, with a HR of 2.80, together with other well-established predictors of survival, such as tumor size, positive nodal status, positive resection margins, and high CA19-9 levels.

A recent study from our institution including patients with previous NAT reported that major complication occurring after pancreatoduodenectomy is significantly associated with reductions in long term survival in comparison with survival in patients without major complications.³⁹ Yet, we observed an exclusive association between CR-POPF and survival, which was independent of the type and severity of other complications. We believe this finding may be representative of a diminished response to NAT, thereby resulting in a softer gland in conjunction with an expectedly lower survival rate. It has been described that an effective response to NAT translates into fibrotic remodeling of the pancreas, and that this response is associated with improved survival.⁴ It is also possible that an impaired immunologic status following NAT results in an even worse long-term outcome once CR-POPF occurs. Release of proinflammatory mediators has been described with large tumor burdens,⁴⁰ and a recent study in esophagogastric cancer showed that patients with a high inflammatory score who were receiving NAT experienced a shorter overall and disease-free survival than those with a normal inflammatory status.⁴¹ In addition, during NAT, patients often experience hematologic abnormalities, such as neutropenia, with consequences for long-term survival that are poorly understood.⁴²

The present study has several limitations. First, the retrospective design cannot exclude confounders, such as the occurrence of any complications during NAT, the need for suspension of treatment, or the use of specific medications, such as corticosteroids or growth factors, which could have affected CR-POPF risk or OS. Moreover, we detected a low rate of CR-POPF, and therefore, no subgroup comparisons according to the severity of fistulas or the type of NAT were possible. In addition, no subgroup analysis of patients who received only neoadjuvant chemotherapy was performed, as the small sample size according to the fistula rate would have biased any statistical considerations. Also, despite the independent association between the fistula and OS, the indications for additional adjuvant treatment in our NAT group were not uniform.

Neoadjuvant therapy appears to provide significant protection from the development of CR-POPF, which is reduced by more than 60%. In addition, standard factors associated with CR-POPF appear to be no longer applicable following NAT. However, when CR-POPF occurs, it may severely impair long-term survival, independent of pathologic findings, receipt of adjuvant treatment, and CA19-9 levels. Therefore, patients with CR-POPF should undergo closer follow-up and perhaps be considered for additional chemotherapy. Future prospective validation studies are needed to confirm these findings.

Accepted for Publication: May 17, 2019.

Corresponding Author: Carlos Fernández-del Castillo, MD, Department of Surgery, Massachusetts General Hospital and Harvard Medical School, 15 Parkman St, Boston, MA 02114-3117 (cfernandez@mgh.harvard.edu).

Published Online: August 14, 2019. doi:10.1001/jamasurg.2019.2272

Author Contributions: Drs Hank and Fernández-del Castillo had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Hank, Sandini, Lillemoe, Fernández-del Castillo.

Acquisition, analysis, or interpretation of data: Hank, Ferrone, Rodrigues, Weniger, Qadan, Warshaw.

Drafting of the manuscript: Hank, Fernández-del Castillo.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Hank, Sandini, Weniger.

Administrative, technical, or material support: Lillemoe.

Supervision: Qadan, Lillemoe, Fernández-del Castillo.

Conflict of Interest Disclosures: Dr Hank reported grants from German Cancer Aid during the conduct of the study. Dr Weniger reported grants from Deutsche Forschungsgmeinschaft during the conduct of the study. Dr Qadan reported personal fees from Olympus outside the submitted work. Dr Warshaw reported being a partner in a preoperative process consulting group, OR DX+RX, Solutions for Surgical Safety. No other disclosures were reported.

Funding/Support: Dr Hank is funded by a Mildred-Scheel-Postdoctoral Fellowship from German Cancer Aid. Dr Weniger received funding from the German Research Foundation (grant 401299842).

Role of the Funder/Sponsor: The supporting organizations had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.