Predictors of Post-Operative Pancreatic Fistula Formation in Pancreatic Neuroendocrine Tumors: a National Surgical Quality Improvement Program Analysis

Abstract

Background:

Post-operative pancreatic fistula (POPF) is a serious complication following pancreas surgery. We aimed to establish factors associated with POPF specifically in patients with pancreatic neuroendocrine tumors (PNET).

Methods:

The 2014 – 2018 American College of Surgeons National Surgical Quality Improvement Program database was querried for patients undergoing resection for PNET. The impact of patient, tumor, and operative factors on POPF formation was evaluated.

Results:

3,532 patient underwent resections for PNET. The POPF rate was significantly higher in patients with PNET (24.8%) versus non-PNET (16.4%) (p<0.0001). Male sex (OR 1.45, 95% CI 1.11 – 1.89), enucleation (OR 3.14, 95% CI 1.10 – 8.98), pancreaticoduodenectomy (OR 1.51, 95% CI 1.13 – 2.03), small duct size < 3 mm (OR 3.24, 95% CI 1.62 – 6.48), and soft gland texture (OR 1.81, 95% CI 1.18 – 2.77) were independently associated with POPF in PNET patients on multivariable analysis.

Conclusions:

POPF is more common in patients undergoing resection for PNET and is dictated primarily by surgical approach and gland characteristics.

Introduction

Post-operative pancreatic fistula (POPF) is estimated to occur in up to 34% of patients and is one of the greatest sources of morbidity and mortality associated with pancreatic surgery(1–3). POPF develops due to leakage of pancreatic exocrine enzymes after resection (such as enucleation bed, anastomosis, or staple line) and can progress to deadly complications such as arterial pseudoaneurysms, hemorrhage, intraabdominal collections, infection, and sepsis; prolonged hospitalization and morbidity rates of 30–50% have been reported even in high volume centers(4–6). Considerable effort, therefore, has been dedicated towards identifying risk factors for POPF and optimal preventive and management strategies. For instance, small pancreatic duct size and soft pancreatic gland texture have been established as risk factors for POPF, while administration of the somatostatin analog pasireotide has been shown in several studies to reduce POPF rates and infectious complications by greater than 50%(7, 8). The mainstay of management for pancreatic fistula remains percutaneous drainage, and if necessary, reoperation (9).

Pancreatic neuroendocrine tumors (PNETs) are a rare subset of pancreatic tumors comprising 1–2% of pancreatic neoplasms and approximately 7% of all neuroendocrine tumors(10). Resection remains a cornerstone of treatment for PNETs. For instance, guidelines published by the North American Neuroendocrine Tumor Society (NANETS) in 2020 and the European Neuroendocrine Tumor Society (ENETS) in 2016 recommended resection for all functioning sporadic PNETs regardless of tumor size and for resection of non-functioning PNETs greater than or equal to 2 cm or for symptomatic patients, those with high-grade tumors, or based on patient preference(11). Additionally, the 2020 NANETS guidelines highlight that lesions greater than 1 cm and less than 2 cm – which are being discovered at increased rates given advances in imaging techniques – can be reasonably managed by observation; however, 8% of PNETs <2 cm that were resected developed recurrences or metastases(12). Altogether, these guidelines leave room for a wide array of provider practices across the United States and establish resection as one of the most common treatments for PNETs.

Given the central role of surgery in PNET management, it is important to understand the rate of POPF and the risk factors associated with POPF specifically in patients with PNET. In the present study, we aimed to evaluate the characteristics of PNET-associated POPF using a large, national database: the American College of Surgeons National Surgical Quality Improvement Project (NSQIP). Given the limited data in the current literature regarding POPF specifically for PNETs, we first compared the rate of POPF in patients with PNET to that of patients with non-PNET to establish a baseline rate of POPF for this histology. Next, we focused on factors associated with POPF formation, specifically in patients with PNETs to uncover any disease-specific risk factors.

Materials and Methods

Study Population

The ACS NSQIP Participant User Files and Pancreatectomy Procedure Targeted Participant User Files for the years 2014 through 2018 were used to identify patients undergoing pancreas-directed procedures. Patients were established to have PNET if the “Malignant Histology Subtype” variable was coded as “Neuroendocrine-functioning” or “Neuroendocrine-nonfunctioning,” while codes “Ampullary carcinoma,” “Cystadenocarcinoma,” “Distal cholangiocarcinoma,” “Duodenal carcinoma,” “IPMN-invaisive,” “Pancreatic adenocarcinoma,” and “Other Type” were defined as “non-PNET”. Type of surgical intervention was coded according to Current Procedural Terminology (CPT) code as follows: “Enucleation” for code 48120; “Total Pancreatectomy” for codes 48146 and 48155; “Ampullectomy” for code 48148 “Distal pancreatectomy” for codes 48140 and 48145; “Pancreaticoduodenectomy” for codes 48150, 48152, 48153, and 48154; and “Unspecified” for code 48999. Patients with code 48999 “Unspecified” were excluded from univariate and multivariate analysis. Patients were categorized as “No fistula”, “Grade A leak”, or “Grade B/C leak” as previously described (13, 14). Specifically, patients were classified as: “No fistula” if coded as “No” or “No evidence of Biochemical Leak or POPF” in NSQIP PUF; “Grade A leak” if coded as “Biochemical leak only,” “Yes-clinical diagnosis, drain continued > 7 days,” “Yes-clinical diagnosis, spontaneous drainage,” and “Yes-persistent drainage, drain continued > 7 days”; and “Grade B/C Leak” if coded as “Yes clinical diagnosis, NPO-TPN,” “Yes-clinical diagnosis, percutaneous drainage performed,” “Yes, clinical diagnosis, reoperation performed,” “Yes-persistent drainage, NPO-TPN,” “Yes-persistent drainage, percutaneous drainage performed,” and “Yes-persistent drainage, reoperation performed(13, 14).” Additionally, demographic data (age, gender, ethnicity), tumor T, N, and M stage, and pancreas-specific factors (duct size, gland texture, vascular reconstruction) were collected as defined by NSQIP PUF. Given the use of de-identified patient data, this study was exempt from IRB review.

Statistical Analysis

Data extracted from NSQIP PUF as above was analyzed using the Statistical Package for the Social Sciences (SPSS) version 26.0 (IBM, New York). Chi-squared analysis and two-tailed student t-tests were used to determine statistical significance for categorical and continuous variables, respectively. Binary and multinomial logistic regression models were employed for univariable and multivariable assessments of factors associated with POPF. The impact of tumor characteristics (functional status, tumor (T), nodal (N), and metastasis (M) stage) and operative technical variables (type of surgery, pancreatic duct size, pancreatic gland texture, type of pancreatic reconstruction, and vascular reconstruction) on POPF formation were evaluated. A two-sided, p-value of less than 0.05 was considered statistically significant for all analysis. Multivariate analysis hazard ratios were calculated along with 95% confidence intervals (CI). A 95% CI that crosses 0 was considered to be not significant.

Results

Baseline Patient Characteristics

Of the 22,890 patients identified in NSQIP 2014 – 2018 to have undergone pancreatic surgeries, 19,358 (84.6%) underwent surgeries for non-PNET pathology and 3,532 (15.4%) patients underwent resection for PNET. Patient demographics in the PNET and non-PNET cohort are described in Table 1. Compared to patients with non-PNETs, patients with PNETs were significantly younger with 65.9% of patients with PNET ≤65 years of age compared to 43.7% of patients with non-PNET histology (p<0.001, mean age for PNET 59 years vs. 64 years for non-PNET) had higher rates of soft pancreatic gland texture (36% vs 28%, p<0.001), pancreatic duct size smaller than 3 mm. (25% vs 20%, p<0.001), and distal pancreatic resection (53.5% vs 20.1%, p <0.001). The majority of patients with PNET undergoing resection had non-functioning tumors (76.1%). The overall rate of POPF was significantly higher in the PNET cohort at 24.8% compared to 16.4% in the non-PNET cohort (p <0.001). This increased rate of POPF was observed for both Grade A, and Grade B/C POPF (PNET vs non-PNET Grade A: 14.8% vs. 9.6%, p <0.001, and Grade B/C: 10% vs. 6.8%, p <0.001, respectively) (Figure 1). Interestingly, the rate of POPF was similar amongst all years from 2014 to 2018 with no significant changes to leak rate over time (Supplemental Figure 1). Next, we analyzed the entire study cohort (NET and non-NET) to assess the associations between age and pancreatic characteristics (duct size and gland texture) and the rate of POPF. There were statistically significant associations between younger age (<65 years) and leak rate, smaller duct size, and softer gland texture. However, for PNETs specifically, younger age was not associated with a higher rate of POPF and softer gland texture, while there was an association between younger age and smaller duct size (Supplemental Tables 1 and 2). We also analyzed the leak rate by procedure, and found that for distal pancreatectomies, total pancreatectomies, and pancreaticoduodenectomies, the leak rates were significantly higher for PNET than for non-PNET (Supplemental Table 3).

Table 1:

Patient characteristics by pancreatic pathology

C	Clinical features	PNET n (%)	Non-PNET n (%)	P-Value
Age	≤65 years	2,326 (65.9%)	8,460 (43.7%)	<0.001
	>65 years	1,206 (34.1%)	10,898 (56.3%)
Total Resections		3,532 (15.4%)	19,358 (84.6%)	-
Gender	Female	1,666 (47.2%)	9,118 (47.1%)	0.94
	Male	1,866 (52.8%)	10,240 (52.9%)
	Yes	211 (6.0%)	900 (4.7%)	<0.001
Hispanic	No	2,949 (83.5%)	16,382 (84.6%)
	Unspecified	372 (10.5%)	2,076 (10.7%)
	No fistula	2,657 (75.2%)	16,186 (83.6%)	<0.001
POPF Grade	A	521 (14.8%)	1,857 (9.6%)
	B/C	354 (10%)	1,315 (6.8%)
Tumor	Functioning	822 (23.3%)	-	-
Functional	Non-Functioning	2,710 (76.7%)	-	-
Status	Hard	303 (8.6%)	5,832 (30.1%)	<0.001
Pancreatic	Intermediate	173 (4.9%)	1,576 (8.1%)
Gland Texture	Soft	1,278 (36.2%)	5,494 (28.4%)
	Unknown	1,778 (50.3%)	6,456 (33.4%)
	<3 mm	865 (24.5%)	3,880 (20%)	<0.001
Main Pancreatic	>6 mm	133 (3.8%)	2,190 (11.3%)
Duct Size	3–6 mm	442 (12.5%)	6,654 (34.4%)
	Unknown	2,092 (59.2%)	6,634 (34.3%)
	Enucleation	227 (6.4%)	48 (0.25%)	<0.001
	Distal Pancreatectomy	1,890 (53.5%)	3,891 (20.1%)
Procedure	Total Pancreatectomy	85 (2.4%)	464 (2.4%)
	Ampullectomy	11 (0.3%)	32 (0.17%)
	Pancreaticoduodenectomy	1,123 (31.8%)	14,676 (75.8%)
	Unspecified	196 (5.6%)	247 (1.28%)
	Yes	275 (7.8%)	3,593 (18.6%)	<0.001
Vascular	No	3,230 (91.4%)	15,564 (80.4%)
Resection	Unknown	27 (0.7%)	201 (1.0%)
	T0 or Tis	13 (0.4%)	247 (1.3%)	<0.001
	T1	1,103 (31.2%)	2,056 (10.6%)
Tumor (T) Stage	T2	1,166 (33.0%)	3,944 (20.4%)
	T3	1,023 (29.0%)	10,947 (56.6%)
	T4	61 (1.7%)	1,077 (5.5%)
	Unknown	166 (4.7%)	1,087 (5.6%)
	NO	1,995 (56.5%)	7,326 (37.8%)	<0.001
Node (N) Stage	N1	1,034 (29.3%)	10,660 (55.1%)
	Unknown	503 (14.2%)	1,372 (7.1%)
Metastases (M)	MO/Mx	2,310 (65.4%)	13,368 (69.1%)	<0.001
Stage	Ml	274 (7.8%)	469 (2.4%)
	Unknown	948 (26.8%)	5,521 (28.5%)

Figure 1. POPF Rates for PNET vs. Non-PNET Cohort:

A comparison of POPF Rate by severity and pancreatic pathology(**** represents p < 0.001). Patients with PNET histology had significantly higher rates of Grade A and Grade B/C POPF compared to patients with Non-PNET histology.

Univariate Analysis

The results from univariate analysis of patients who developed POPF compared to those who did not develop POPF within the PNET group are summarized in (Table 2). Distal pancreatectomy, small duct size < 3 mm, soft pancreatic gland texture, and male sex were all statistically significant factors associated with POPF formation in NET patients. Other patient-specific factors such as age and race, technical factors such as vascular reconstruction, and tumor-specific factors such as T stage, nodal spread, or functioning tumor, did not have a statistically significant association with POPF formation.

Table 2:

Univariate and Multivariate Analysis of PNET Characteristics

Variable		P-Value (Univariate)	P-Value (Multivariate)	OR (95% CI)
	Distal Pancreatectomy	Reference	-	-
Procedure	Enucleation	0.075	0.033	3.139 (1.097 – 8.983)
	Total Pancreatecomy	0.076	0.999	-
	Ampullectomy	0.999	1.000	-
	Pancreaticodeuodenectomy	0.011	0.006	1.511 (1.126 – 2.027)
	>6 mm	Reference	-	-
Duct Size	<3 mm	<0.001	0.001	3.240 (1.619 – 6.483)
	3–6 mm	0.005	0.006	2.675 (1.318 – 5.431)
Pancreatic	Hard	Reference	-	-
Gland Texture	Intermediate	0.962	0.667	0.864 (0.443–1.683)
	Soft	<0.001	0.007	1.805 (1.176 – 2.771)
Sex	Female	Reference	-	-
	Male	0.003	0.006	1.450 (1.11 – 1.893)
	T0/Tis	Reference	-	-
	T1	0.999	-	-
Tumor (T) Stage	T2	0.999	-	-
	T3	0.999	-	-
	T4	0.999	-	-
Node (N) Stage	N0	Reference	-	-
	N1	0.390	-	-
Metastasis (M)	M0	Reference	-	-
Stage	M1	0.296	-	-
	White	0.232	-	-
	Black	0.078	-	-
Race	Asian	0.474	-	-
	Other	0.372	-	-
Age	<65 Years	Reference	-	-
	>65 Years	0.985	-	-
Functioning	Non-Functioning	Reference	-	-
	Functioning	0.828	-	-
Vascular	No	Reference	-	-
Reconstruction	Yes	0.988	-	-

Multivariate Analysis

We next performed multivariable analysis within the PNET cohort. For our multivariable model, we did include tumor stage, functional status, vascular reconstruction, nodal disease and metastasis in our model as these were clinically relevant (Table 2). Small duct < 3 mm or 3 – 6 mm, soft gland texture, enucleation, pancreaticoduodenectomy, and male sex were significantly associated with development of POPF with odds ratio greater than 1 (Figure 2).

Figure 2. Factors Associated with POPF Formation:

A Forrest Plot of Odds Ratio for Gender, Pancreatic Gland Texture, Pancreatic Duct Size, and Surgery demonstrates that male sex, soft gland texture, pancreatic duct size <3 mm and 3–6 mm, and pancreaticoduodenectomy are all significantly associated with increased rate of POPF. PD represents pancreaticoduodenectomy and DP represents distal pancreatectomy

Discussion

Surgical resection is a cornerstone of treatment for pancreatic lesions, and PNETs are no exception. As such, accurate assessment and appreciation of the complications of surgery is critical for both providers and patients alike. POPF is a debilitating complication for pancreatic surgery, and in this study, we present a large, national level assessment of this complication specifically as it pertains to PNET. Reviewing the records of 3,563 patients who underwent surgery for PNET across a 5-year span, we find that patients with PNET are more likely to experience POPF of all grades compared to non-PNET pathologies. Our data suggests that this is due to the higher proportion of PNET patients having smaller main pancreatic duct and soft gland texture. Duct size <3mm and soft pancreatic gland texture are independent predictors irrespective of patient age for leaks in PNETs. The differences in POPF rate between PNET and non-NET pathology cannot be explained by age differences in the patient populations. Tumor-specific factors, such as the size of tumor, presence of nodal or distant metastasis, and functional status did not correlate with POPF rates in PNETs.

Our results are not surprising and are altogether consistent with previous single-institution series evaluating this topic. For instance, Atema and colleagues, on evaluating the post-resection outcomes of 88 patients with PNET and 744 patients with non-PNET pathologies at a tertiary medical center in the Netherlands determined that PNETs had a higher POPF rate than non-NET (23% vs. 17%). In their population, pancreatic duct diameter < 3mm and the type of resection (central pancreatectomy) being associated with POPF formation; however, this assessment was made for the entire study population and not specifically for the patients with PNET(15). More recently, Partelli and colleagues evaluated the outcomes after pancreaticoduodenectomy for 179 patients with PNET across three European centers and noted that this population had softer gland texture, smaller pancreatic duct size, and higher rates of POPF (34% vs 19%) compared with patients undergoing this procedure for PDAC(16). We now provide data from a national cohort confirming that PNETs are indeed associated with higher rates of POPF. Most importantly, our larger sample size enabled uni- and multivariable analysis specifically for PNETs (not possible or performed in previous case series) which confirmed that gland texture and small ductal size are independent risk factors for POPF in PNET. Indeed, in this regard PNETs are similar to PDAC where previous reports have also confirmed small pancreatic duct size < 3 mm and soft gland textures are independent risk factors for POPF (17, 18). Compared to PDACs (and other non-NET pathologies), however, the proportion of PNET patients with these high-risk features is significantly higher. This could be explained, in part, by the fact that most PDAC tumors are associated with dense fibrotic changes (leading to harder gland texture) and arise from pancreatic ductal cells (thereby causing ductal obstruction and dilation), while PNETs, arising from the pancreatic parenchyma itself, does not lead to these changes in pancreas.

Another interesting finding from our study was that the overall rates of POPF did not change significantly between the years 2014 and 2018, despite the seminal publication by Allen and colleagues in 2014 demonstrating the significant benefit of pasireotide in reducing leak rate by greater than 50%(7). We were unable to study the usage of pasireotide directly in this study due to a lack of coding for pasireotide administration in the NSQIP database. One reason for this lack of change could be that the interval window was too close to the introduction of an intervention to have an impact on practice patterns. Further examination is needed at future time points to see if pasireotide has an impact on reducing POPF in real-world clinical practice. Another possible explanation is regarding the cost of pasireotide itself and difficulty obtaining the drug in hospitals. When the cost of pasireotide is included in overall cost it increased the cost of hospital stay from $42,159 to $77,202(19). However, when looking at the cost of reducing the associated morbidity of POPF including readmission, reoperation, prolonged hospital stays, and parenteral nutrition saving $390 per patient even given the expense of pasireotide(20). Additionally, other studies have shown that pasireotide did not reduce the incidence or severity of POPF after pancreaticoduodenectomy in a single center study(21). POPF rates did not decrease between 2014 through 2018, which indicates there are still many improvements needed in reducing the rate of POPF.

There are a few limitations to our study pertaining to the use of the ACS NSQIP database that should be mentioned. One is that we are unable to reliably establish secondary diagnoses from the NSQIP database. Between 15–20% of patients with PNET have an inherited genetic syndrome such as multiple endocrine neoplasia (MEN) type 1, von Hippel-Lindau syndrome (VHL), neurofibromatosis type 1(NF), and tuberous sclerosis complex (TSC)(22). These syndromic patients often undergo multiple resections for different tumors throughout their lives and therefore have special considerations for surgical management compared to patients with sporadic PNET. Patients with VHL are more likely to develop PNETs at a younger age with a mean age of diagnosis in the 30’s as compared to older age of diagnosis for pancreatic adenocarcinoma, however, these patients also had a lower incidence of metastasis compared to sporadic non-functioning PNETs(23). Syndromic patients likely comprise a different subgroup of patients with PNET that will behave differently their non-syndromic counterparts. Further analysis on this subgroup is needed. Another limitation is inherent to using a deidentified, national database such as NSQIP, which requires both clinicians and data entry teams to make assumptions for data entry into the database. Additionally, NSQIP does not provide granular details regarding potential leak reducing interventions such as pasireotide administration or the use of intraoperative ultrasound to identify the pancreatic duct or oversew or seam sealing techniques in surgery. Nevertheless, utilizing NSQIP does enable us to examine the POPF rates on a national level as well as to evaluate the risk factors for POPF formation in patients with PNET specifically. Since PNET is a relatively rare diagnosis, institutional or even multi-institutional collaboration data would not provide a broad evaluation into the risk factors for POPF in this particular group.

Conclusion

This study helps to establish risk factors for development of POPF in patients undergoing pancreatic resection for PNET. We demonstrate that PNETs have higher rates of POPF compared to non-PNET pathologies, and describe small pancreatic duct and soft gland texture as specific risk factors for POPF in PNET. Based on the findings of this study, there is no difference in risk factors for functioning versus non-functioning tumors. Patients undergoing resection for PNET have a significantly higher rate of POPF and, in particular, clinically relevant grade B and C POPF compared to other histologies. Resection for PNET remains a mainstay of treatment, but surgeons should consider the high morbidity associated with this particular pathology in their post-operative care. Patients with soft pancreatic glands and small pancreatic ducts are more likely to experience POPF and patients with PNET histology are more likely to fit those descriptions than patients undergoing pancreatic resection for other pathologies.

Highlights.

• Post-operative pancreatic fistula is higher in pancreatic neuroendocrine tumors

• Male sex, pancreaticoduodenectomy, small duct, and soft gland have higher leak rate

• Tumor stage, nodes, metastasis, race, and age were not associated with leak rate

• Functioning and non-functioning tumors had similar post-operative leak rates

• Post-operative pancreatic fistula rates did not change between 2014 and 2018