Retrospective investigation of risk factors for pancreatic fistula development after pancreaticoduodenectomy

Abstract

BACKGROUND

This study aimed to compare and analyze risk factors for pancreatic fistula following pancreaticoduodenectomy (PD) using different definition criteria, and to develop a predictive model for standardized pancreatic fistula risk assessment.

AIM

To identify and compare risk factors for postoperative pancreatic fistula (POPF) following PD using both the 2005 International Study Group of Pancreatic Fistula and updated 2016 International Study Group on Pancreatic Surgery diagnostic criteria, and to develop a clinically applicable predictive model based on objective preoperative parameters for standardized pancreatic fistula risk assessment and perioperative management optimization.

METHODS

We conducted a retrospective analysis of 303 patients who underwent PD at CR&WISCO General Hospital between January 2017 and May 2023. POPF cases were classified according to both previous and updated diagnostic standards. For statistical analysis, we employed t-tests or Mann-Whitney U tests for continuous variables and χ² tests for categorical data. To identify risk factors associated with POPF under both classification systems, we performed univariate and multivariate logistic regression analyses.

RESULTS

Univariate analysis identified several factors associated with POPF: Main pancreatic duct diameter (χ² = 31.641, P < 0.001), main pancreatic duct index (χ² = 52.777, P < 0.001), portal vein invasion (χ² = 6.259, P = 0.012), intra-abdominal fat thickness (χ² = 7.665, P = 0.006), preoperative biliary drainage (χ² = 5.999, P = 0.014), pancreatic characteristics (χ² = 5.544, P = 0.019), pancreatic resection margin thickness (t = 2.055, P = 0.032), pancreatic computed tomography (CT) value (t = -3.224,P = 0.002), and preoperative blood amylase level (Z = -2.099, P = 0.036). Multivariate logistic regression identified three independent risk factors: Main pancreatic duct index [odds ratio (OR) = 0.000, 95% confidence interval (CI): 0.000-0.011], pancreatic cancer [OR = 4.843, 95%CI: 1.285-18.254], and pancreatic CT value [OR = 0.869, 95%CI: 0.806-0.937] (all P < 0.05).

CONCLUSION

The main pancreatic duct index and pancreatic CT value are strongly correlated with pancreatic fistula development after PD.

Key Words: Risk prediction model; Risk factor; Pancreatic fistula; Pancreaticoduodenectomy; Predictive model

Core Tip: Postoperative pancreatic fistula remains a significant challenge after pancreaticoduodenectomy, impacting patient outcomes and healthcare costs. Recent advancements in diagnostic criteria have refined our understanding of risk factors, including the main pancreatic duct index, pancreatic computed tomography value, and tumor characteristics. Integrating these factors into predictive models helps identify high-risk patients early, allowing for targeted interventions and improved surgical outcomes. Adopting standardized grading systems ensures consistent reporting and enhances clinical decision-making. This study underscores the importance of combining clinical, radiological, and pathological data in postoperative pancreatic fistula risk assessment, ultimately guiding more personalized perioperative care strategies and reducing complication rates.

INTRODUCTION

Pancreaticoduodenectomy (PD) complications are dominated by postoperative pancreatic fistula (POPF), which persists in 3%-45% of cases at major medical centers despite a decade of surgical and care advancements[1-3]. The International Study Group of Pancreatic Fistula (ISGPF) established standardized diagnostic and grading criteria for POPF in 2005, which became widely implemented in both research and clinical settings[4]. In 2023, the International Study Group on Pancreatic Surgery (ISGPS) updated these criteria, emphasizing clinical relevance and reclassifying grade A pancreatic fistula as biochemical leakage rather than true fistula[5]. Our retrospective analysis examines POPF risk factors and develops a predictive model aligned with these revised criteria.

While PD remains essential for treating pancreatic head and duodenal malignancies and chronic pancreatitis, POPF continues to present significant clinical challenges[6-8]. Beyond extending hospitalization and increasing costs, pancreatic fistulas can trigger cascading complications that potentially threaten patient survival. Consequently, developing precise POPF risk prediction tools hold tremendous value for prevention strategies and postoperative management[9]. The pathophysiology of POPF involves the leakage of enzyme-rich pancreatic fluid from the pancreaticojejunostomy anastomosis site, leading to local inflammation, autodigestion of surrounding tissues, and potential vascular erosion. This process is influenced by multiple factors including the texture of the pancreatic remnant, main pancreatic duct diameter, anastomotic technique, intraoperative blood loss, and patient-specific risk factors such as nutritional status, comorbidities, and preoperative biliary obstruction. The complex interplay of these elements underscores the need for comprehensive risk assessment models that incorporate both patient characteristics and surgical variables. Furthermore, recent advances in imaging technology have enabled more precise preoperative assessment of pancreatic parenchyma density, ductal anatomy, and peripancreatic fat distribution, offering new parameters for risk stratification that previous models may not have fully utilized.

Through comprehensive analysis of preoperative clinical variables and intraoperative factors, our research aims to create a predictive model enabling clinicians to promptly identify high-risk patients and implement targeted preventive measures, ultimately reducing POPF incidence and improving surgical outcomes[10-12]. While previous domestic and international studies have proposed various risk assessment frameworks, limitations persist in sample size adequacy and predictive capability[13]. Furthermore, postoperative management approaches vary widely across clinical settings without standardized guidelines. The pressing need exists for individualized risk assessment tools that facilitate personalized postoperative management strategies based on patient-specific clinical characteristics. This investigation leverages comprehensive clinical datasets and advanced analytics to develop a machine learning-based POPF risk prediction model. Our goal is to provide clinicians with robust decision support tools that minimize postoperative complications and enhance both the safety and efficacy of surgical interventions.

MATERIALS AND METHODS

Research subjects

The study incorporated 303 patients diagnosed with PD who were admitted to the Department of Hepatobiliary Surgery at CR&WISCO General Hospital between January 2017 and May 2023. Based on admission timing, patients were categorized into modeling and validation groups. For inclusion in the study, patients needed to have undergone standard contrast-enhanced computed tomography (CT) examination within 14 days before PD procedure and have no prior history of radio or chemotherapy. Patients were excluded from the study if they underwent PD in combination with surgery on other organs, lacked complete clinical information, had additional malignant neoplasms, or received emergency surgical intervention due to trauma. The Medical Ethics Committee of CR&WISCO General Hospital approved this research (Approval No. CRWG2023R033J), and all patients provided signed informed consent for their treatment.

Surgical methods

All surgical procedures for patients in this cohort were conducted by multiple surgeons possessing substantial experience in pancreatic operations. Each patient underwent open radical PD, with digestive tract reconstruction implemented using the classical child methodology.

Diagnostic and grading criteria for POPF

The original ISGPF diagnostic criteria from 2005 defined pancreatic fistula as “drainage amylase content exceeding 3 times the normal upper limit of serum amylase for ≥ 3 days following surgery”, with severity classifications of A, B, and C. The updated ISGPS diagnostic standard maintains the requirement for drainage amylase content surpassing 3 times the normal serum amylase upper limit for ≥ 3 days post-operation, but now associates this with clinical treatment approaches and prognosis outcomes. In the revised standards, what was previously termed “Grade A pancreatic fistula” in the 2005 definition has been reclassified as “biochemical leakage”. This change emphasizes that pancreatic fistula is not considered present when drainage tube amylase content meets diagnostic criteria but does not impact the clinical treatment process or patient prognosis. Table 1 presents the detailed classification system.

Table 1 Comparison of definition and grading system for postoperative pancreatic fistula.

	The definition and grading system of postoperative pancreatic fistula in the 2005 ISGPF version			The ISGPS version of the definition and grading system for postoperative pancreatic fistula
Classification	A-level	B-level	C-level
Clinical manifestation	Good	Good	Poor
Special treatment	Yes	No	No
Continuous ultrasound or CT	Negative	Positive	Positive
Continuous drainage for > 3 weeks	No	No	Yes
Secondary surgery	No	No	Yes
Pancreatic fistula related death	No	No	No
Symptoms of infection	No	Yes	Yes
Sepsis	No	No	Yes
Readmission	No	Yes	Yes
Definition and classification				Biochemical leakage	B-level	C-level
On the third day after surgery, the amylase value in the drainage fluid reached more than three times the normal upper limit				Yes	Yes	Yes
Peripancreatic continuous drainage > 3 weeks				No	Yes	Yes
Changes in clinically relevant treatment measures for pancreatic fistula				No	Yes	Yes
Transcutaneous or intradermal microscopic puncture drainage				No	Yes	Yes
Angiographic intervention treatment of postoperative pancreatic fistula related bleeding				No	Yes	Yes
Secondary operation				No	No	Yes
Postoperative pancreatic fistula related symptoms of infection				No	With (without organ failure)	Yes
Postoperative pancreatic fistula related organs failure				No	No	Yes
Postoperative pancreas fistula related death				No	No	Yes

CT: Computed tomography; ISGPF: International Study Group of Pancreatic Fistula; ISGPS: International Study Group on Pancreatic Surgery.

Prediction system construction and analysis

The pancreatic fistula prediction system developed in this research encompasses two primary indicators: The main pancreatic duct index and the pancreatic CT value. The main pancreatic duct index represents the proportion between the pancreatic duct diameter and the pancreatic thickness at the incision surface. The measurement procedure involves selecting the cross-section where the superior mesenteric vein connects with the hepatic portal vein, then measuring the pancreatic duct diameter at this plane to determine the patient’s main pancreatic duct diameter. The longest anteroposterior diameter of the pancreas, measured perpendicular to the main pancreatic duct direction, was recorded as the patient’s pancreatic thickness. The pancreatic CT value refers to the measured CT value of the pancreatic amputation surface on plain CT scan images. Specifically, using the superior mesenteric artery crossing the left pancreatic side as the incision surface reference, a longitudinal oval region was selected while avoiding the splenic artery and pancreatic duct, and the CT values within this area were documented. Following this procedure, CT values from three distinct pancreatic layers were measured, and their average was calculated to determine the patient’s pancreatic CT value.

Pancreatic CT value measurement

Pancreatic CT values were measured on non-contrast CT images by two blinded radiologists at the level where the superior mesenteric artery crosses the pancreas. An oval region of interest (≥ 50 mm²) was placed in the pancreatic parenchyma, avoiding the main duct, vessels, and focal lesions. Three measurements were taken from different pancreatic sections (head, body, inferior margin) and averaged to obtain the final CT value in Hounsfield unit (HU). Inter-observer agreement was excellent (intraclass correlation coefficient = 0.89), with consensus reading for discrepancies > 5 HU. This standardized approach provides objective, reproducible assessment of pancreatic density as a marker for tissue quality and surgical risk stratification.

Statistical analysis

Statistical analysis was performed using SPSS 26.0 software. Data with normal distribution is expressed as mean ± SD, with group t tests employed for between-group comparisons. Non-normally distributed data is represented by M (P25-P75), with the Mann-Whitney U test utilized for comparisons between groups, while the χ² test was applied for additional inter-group analyses. Statistically significant risk factors underwent multivariate logistic regression analysis to identify independent risk factors associated with pancreatic fistula and establish a POPF risk prediction model. Receiver operating characteristic (ROC) curves were employed to analyze the optimal sensitivity, specificity, and area under the curve (AUC) of this prediction model for POPF, and model verification was conducted. Statistical significance was defined as P < 0.05.

RESULTS

Analysis of general clinical data

The investigation included a total of 303 participants, with 253 patients receiving PD from January 2017 through July 2022 (constituting the modeling group) and 50 additional patients treated between August 2022 and May 2023 (forming the validation group). Among the participants, 197 were males and 106 were females, with ages spanning from 34 to 82 years and an average age of 62 ± 8 years. Concurrent medical conditions were documented, including diabetes mellitus (affecting 67 individuals) and cholangitis (present in 61 individuals), while 35 participants had undergone biliary drainage procedures prior to surgery. Histopathological evaluation revealed a spectrum of diagnoses: Cancer of the pancreatic head (identified in 72 individuals), malignancy of the lower common bile duct (documented in 63 individuals), ampullary carcinoma (diagnosed in 123 individuals), duodenal malignancy (found in 23 individuals), and various non-malignant conditions (present in 22 individuals, specifically including 2 cases of duodenal papillitis, 5 instances of pancreatitis, 4 cases of duodenal papilloma, 2 pancreatic head neuroendocrine tumors, 7 benign pancreatic cystic neoplasms, and 2 occurrences of cholangitis). With respect to operative approaches, 57 participants underwent pancreatic duct jejunal mucosal-to-mucosal anastomotic procedures, while end-to-side pancreaticojejunostomy was performed in 246 participants. Pancreatic duct drainage stents were implemented in all cases, whereas biliary duct stent tubes were not utilized in any patients (Table 2).

Table 2 Clinical characteristics of the modeling and validation sets, n (%).

Index	Modeling group (n = 253)	Validation group (n = 50)	P value
Gender			0.416
Male	167 (66.0)	30 (60.0)
Female	86 (34.0)	20 (40.0)
Age (years)	61 ± 9	62 ± 7	0.686
BMI (kg/m2)	23.1 ± 2.7	22.4 ± 3.3	0.312
Alcohol consumption			0.246
Yes	48 (19.0)	13 (26.0)
No	205 (81.0)	37 (74.0)
Smoking			0.846
Yes	110 (43.5)	21 (42.0)
No	143 (56.5)	29 (58.0)
Diabetes			0.983
Yes	56 (22.1)	11 (22.0)
No	197 (77.9)	39 (78.0)
Main pancreatic duct diameter (mm)	3.8 ± 2.1	3.7 ± 2.2	0.606
Cutting edge pancreatic thickness (mm)	15.3 ± 3.5	14.7 ± 3.1	0.220
Main pancreatic duct index	0.3 ± 0.2	0.3 ± 0.2	0.885
Portal vein invasion			0.901
Yes	16 (6.3)	4 (8.0)
No	237 (93.7)	46 (92.0)
Intra-abdominal fat thickness (mm)	71.1 ± 27.2	68.3 ± 25.3	0.389
Pancreatic CT value (HU)	38.6 ± 8.3	38.1 ± 8.8	0.693
Preoperative biliary drainage			0.707
Yes	30 (11.9)	5 (10.0)
No	223 (88.1)	45 (90.0)
Preoperative laboratory examination
White blood cells	6.1 ± 1.7	5.9 ± 1.6	0.397
Platelets	232.5 ± 68.5	240.3 ± 62.8	0.316
Albumin (g/L)	39.3 ± 4.1	38.7 ± 3.8	0.662
Total bilirubin (μmol/L)	121.3 (34.1-231.4)	130.5 (39.2-190.0)	0.747
Serum amylase (IU/L)	25.0 (16.0-42.0)	26.0 (19.5-39.3)	0.737
CA19-9 (U/mL)	81.9 (34.2-273.1)	94.0 (28.5-357.5)	0.626
Pancreatic intestinal anastomosis method			0.814
Pancreatic duct jejunal mucosa vs mucosa	47 (18.6)	10 (20.0)
Pancreatic intestinal end to side insertion	206 (81.4)	40 (80.0)
Pancreatic cancer			0.684
Yes	59 (23.3)	13 (26.0)
No	194 (76.7)	37 (74.0)
Postoperative pancreatic dysfunction
ISGPF			0.517
Yes	124 (49.0)	22 (44.0)
No	129 (51.0)	28 (56.0)
ISGPS			0.987
Yes	61 (24.1)	12 (24.0)
No	192 (75.9)	38 (76.0)
Surgical related deaths			0.658
Yes	8 (3.2)	1 (2.0)
No	245 (96.8)	49 (98.0)

BMI: Body mass index; CT: Computed tomography; HU: Hounsfield unit; CA19-9: Carbohydrate antigen 19-9; ISGPF: International Study Group of Pancreatic Fistula; ISGPS: International Study Group on Pancreatic Surgery.

Comparison of pancreatic fistula after PD in the modeling group

Within the modeling group consisting of 253 patients, postoperative interventions included percutaneous drainage procedures for 41 individuals who developed abdominal effusion, pleural effusion, or liver abscess. Additionally, 8 patients received angiogram interventional therapy to address hemorrhage related to pancreatic fistula, while secondary surgical interventions became necessary for 3 patients who experienced abdominal bleeding attributed to pancreatic fistula. Perioperative mortality occurred in 8 patients, with 6 of these deaths resulting from peritoneal infection and hemorrhagic complications associated with pancreatic fistula. When analyzed according to the original ISGPF POPF definition and grading criteria established in 2005, pancreatic fistula was identified in 124 of the 253 modeling group patients (distributed as 50 patients with grade A classification, 64 patients with grade B classification, and 10 patients with grade C classification), yielding an overall incidence rate of 49.0%. In contrast, application of the revised ISGPS POPF definition and classification system revealed pancreatic fistula in 61 of the 253 modeling group patients (comprising 53 individuals with grade B classification and 8 individuals with grade C classification), corresponding to an incidence rate of 24.1%. The implementation of the updated classification system resulted in downgrading certain cases previously categorized as C-grade pancreatic fistulas to B-grade designations, while some former B-grade pancreatic fistulas were reclassified as biochemical leakage. Consequently, within this identical patient population, the modification of diagnostic criteria produced a substantial reduction in the documented pancreatic fistula incidence, decreasing from 49.0% to 24.1%.

Single-factor analysis of pancreatic fistula risk after PD in the modeling group

Regarding the 2005 ISGPF POPF definition and classification system: Univariate analysis outcomes revealed that several factors demonstrated significant correlation with pancreatic fistula development following PD (P < 0.05), including the main pancreatic duct’s diameter, the calculated index of the main pancreatic duct, portal vein invasion status, the measured thickness of intra-abdominal adipose tissue, implementation of preoperative biliary drainage procedures, presence or absence of pancreatic malignancy, the measured thickness at the pancreatic resection margin, the quantified CT value of pancreatic tissue, and preoperative blood amylase concentration levels (Table 3).

Table 3 Univariate analysis results of risk factors related to postoperative pancreatic fistula after pancreaticoduodenectomy (2005 International Study Group of Pancreatic Fistula edition).

Index	Postoperative pancreatic fistula in PD patients		Statistical value	P value
	Yes	No
Main pancreatic duct diameter			χ2 = 31.641	< 0.001
≤ 3 mm	88	46
> 3 mm	36	83
Main pancreatic duct index			χ2 = 52.777	< 0.001
≤ 0.25	101	47
> 0.25	23	82
Portal vein invasion			χ2 = 6.259	0.012
Yes	3	13
No	121	116
Intra-abdominal fat thickness			χ2 = 7.665	0.006
≤ 65 mm	43	67
> 65 mm	81	62
Preoperative biliary drainage			χ2 = 5.999	0.014
Yes	21	9
No	103	120
Pancreatic cancer			χ2 = 5.544	0.019
Yes	21	38
No	103	91
Cutting edge pancreatic thickness (mm)	16.5 ± 3.7	15.0 ± 3.8	t = 2.055	0.032
Pancreatic CT value (HU)	36.3 ± 6.4	41.1 ± 9.6	t = -3.224	0.002
Gender			χ2 = 0.572	0.449
Male	79	88
Female	45	41
Drinking alcohol			χ2 = 1.242	0.265
Yes	27	21
No	97	108
Smoking			χ2 = 0.00	0.974
Yes	54	56
No	70	73
Diabetes			χ2 = 0.549	0.459
Yes	25	31
No	99	98
Pancreatic intestinal anastomosis method			χ2 = 1.703	0.192
Pancreatic duct jejunal mucosa vs mucosa	19	28
Pancreatic intestinal end to side insertion	105	101
Age (years)	61 ± 9	61 ± 7	f = -0.259	0.796
BMI (kg/m2)	23.0 ± 3.1	23.2 ± 2.8	t = -0.372	0.626
Preoperative laboratory examination
White blood cells	6.1 ± 1.9	6.0 ± 1.6	t = 0.773	0.441
Platelets	230.4 ± 69.6	235.0 ± 64.1	t = -0.892	0.374
Albumin (g/L)	39.4 ± 4.2	38.9 ± 4.1	t = 0.757	0.450
Total bilirubin (μmol/L)	124.8 (55.6-249.3)	117.8 (37.0-219.1)	U = -1.093	0.275
CA19-9 (U/mL)	73.3 (29.4-222.6)	96.2 (42.0-370.1)	U = -1.069	0.285
Serum amylase (IU/L)	23.0 (17.0-34.0)	31.5 (17.3-60.0)	U = -2.099	0.036

PD: Pancreaticoduodenectomy; CT: Computed tomography; HU: Hounsfield unit; BMI: Body mass index; CA19-9: Carbohydrate antigen 19-9.

When examining the revised ISGPS POPF definition and grading system: Findings from univariate analysis demonstrated that pancreatic fistula occurrence after PD showed significant association (P < 0.05) with five key parameters - specifically, the dimensional measurement of the main pancreatic duct, the calculated ratio known as the main pancreatic duct index, the quantified thickness of intra-abdominal fat deposits, the measured thickness of the pancreatic tissue at the surgical incision margin, and the numerically determined CT value of pancreatic parenchyma (Table 4).

Table 4 Univariate analysis results of risk factors related to postoperative pancreatic fistula after pancreaticoduodenectomy (International Study Group on Pancreatic Surgery).

Index	Postoperative pancreatic fistula after PD surgery		Statistical value	P value
	Yes	No
Main pancreatic duct diameter			χ2 = 5.391	0.020
≤ 3 mm	39	90
> 3 mm	22	102
Main pancreatic duct index			χ2 = 11.39	0.001
≤ 0.25	47	101
> 0.25	14	91
Intra-abdominal fat thickness			χ2 = 8.899	0.003
≤ 65 mm	16	92
> 65 mm	45	100
Cutting edge pancreatic thickness (mm)	17.0 ± 4.2	14.5 ± 3.6	t = 2.665	0.009
Pancreatic CT value (HU)	36.2 ± 7.3	39.7 ± 8.8	t = -2.835	0.004
Portal vein invasion			χ2 = 0.672	0.412
Yes	2	14
No	59	178
Preoperative biliary drainage			χ2 = 2.933	0.087
Yes	11	19
No	50	173
Pancreatic cancer			χ2 = 1.721	0.190
Yes	18	41
No	43	151
Gender			χ2 = 0.720	0.396
Male	43	124
Female	18	68
Drinking alcohol			χ2 = 0.286	0.593
Yes	13	35
No	48	157
Smoking			χ2 = 0.020	0.887
Yes	27	83
No	34	109
Diabetes			χ2 = 0.283	0.595
Yes	12	44
No	49	148
Pancreatic intestinal anastomosis method			χ2 = 2.680	0.102
Pancreatic duct jejunal mucosa vs mucosa	7	40
Pancreatic intestinal end to side insertion	54	152
Age (years)	6 ± 8	60 ± 10	t = 1.246	0.215
BMI (kg/m2)	22.9 ± 4.3	23.3 ± 2.1	t = -0.738	0.534
Preoperative laboratory examination
White blood cells	6.2 ± 2.3	6.0 ± 1.7	t = 0.474	0.673
Platelets	230.9 ± 67.9	237.8 ± 73.2	t = -0.523	0.602
Albumin (g/L)	39.2 ± 4.3	39.0 ± 4.0	t = 0.303	0.762
Total bilirubin (μmol/L)	127.1 (50.0-231.4)	121.5 (32.9-216.6)	U = -0.882	0.378
CA19-9 (U/mL)	97.4 (34.5-314.4)	80.9 (27.3-344.3)	U = -0.498	0.618
Serum amylase (IU/L)	24.0 (17.0-33.8)	30.0 (17.0-58.0)	U = -1.818	0.069

PD: Pancreaticoduodenectomy; CT: Computed tomography; HU: Hounsfield unit; BMI: Body mass index; CA19-9: Carbohydrate antigen 19-9.

Multivariate logistic regression analysis of pancreatic fistula after PD

In relation to the 2005 edition of the ISGPF POPF definition and classification system: Those variables demonstrating statistical significance during univariate analysis underwent subsequent examination through multivariate logistic regression methodology. The analytical outcomes identified three parameters as independent risk factors contributing to pancreatic fistula formation following PD procedures (P < 0.05), specifically: The calculated principal pancreatic duct index, the histopathological confirmation of pancreatic malignancy, and the quantitative CT value measurement of pancreatic tissue.

With respect to the updated ISGPS POPF definition and classification framework: Variables exhibiting statistical significance in preliminary univariate analysis were subjected to comprehensive multivariate logistic regression evaluation. This advanced statistical assessment determined that two specific parameters functioned as independent risk factors for the development of pancreatic fistula complications after PD interventions (P < 0.05): The mathematically derived principal pancreatic duct index and the radiologically determined pancreatic CT value measurement.

Pancreatic fistula risk prediction model after PD surgery

According to the multivariate logistic analysis results of the new ISGPS definition of pancreatic fistula and the grading standard, the logistic regression probability equation, that is, the risk probability model of pancreatic fistula after PD, can be obtained: P = 1/[1 + e^{-(2.667 - 6.995 × principal pancreatic duct index - 0.059 × pancreatic CT value)}].

The predictive value of the pancreatic fistula prediction model for POPF with PD

ROC curve analysis of the prediction model revealed that when the diagnostic threshold of the P-value was 0.32, the sensitivity of the ROC curve was 81.6%, the specificity was 74.4%, and the AUC was 0.788 (95% confidence interval: 0.77-0.870) (Figure 1A).

Figure 1 Receiver operating characteristic curve of the prediction model for pancreatic fistula after pancreaticoduodenectomy and pancreaticoduodenectomy in validation set. A: Receiver operating characteristic curve of the prediction model for pancreatic fistula after pancreaticoduodenectomy; B: Receiver operating characteristic curve of the prediction model for pancreatic fistula after pancreaticoduodenectomy in validation set.

Validation of the pancreatic fistula prediction model

To further verify this model, 50 patients who underwent PD surgery between August 2022 and May 2023 were selected as the verification group. The predictive model was applied to this group of patients, and ROC curve analysis revealed that the AUC of POPF in PD patients predicted by this model was 0.804 (95% confidence interval: 0.675-0.932) (Figure 1B).

DISCUSSION

The ISGPF pioneered the initial POPF definition and classification system for PD, which gained widespread adoption in both clinical practice and research settings[14-16]. By December 2015, this framework had received over 1700 citations and been applied to analyze more than 320000 patients clinically. However, with time, increasing problems and limitations emerged within the grading system[17]. The ISGPF revised its definition and grading criteria in 2016, reclassifying “Grade A pancreatic fistula” from the 2005 definition as “bioleakage”, which no longer constitutes a true pancreatic fistula[18-20]. Clinical relevance drives the necessity for diagnosing grade B fistulas, which impact postoperative management. Beyond “biochemical leakage”, grade B fistula diagnosis requires any of these conditions: Persistent abdominal drainage exceeding 3 weeks, modification of clinically relevant pancreatic fistula treatment approaches, percutaneous or endoscopic drainage, angiographic intervention for bleeding, without infection signs or organ failure (the updated criteria note that discharged patients with tubes who maintain good condition, oral intake, and tube observation beyond 3 weeks still represent biochemical leakage)[21-23]. Grade B pancreatic fistula escalates to grade C when reoperation becomes necessary, when one or more organs fail, or when death occurs. The 2016 revision emphasizes clinical relevance with clearer definitions, classifications, and enhanced practical applicability[24].

Numerous investigations have explored various risk factors associated with POPF development in patients undergoing PD, encompassing parameters from preoperative, intraoperative, and postoperative phases[25-28]. Scientists worldwide have consolidated multiple risk determinants to establish prediction systems for pancreatic fistula with enhanced predictive capability[29]. The diameter of the main pancreatic duct and pancreatic tissue consistency are presently recognized as established risk determinants. Supplementary factors identified include patient age, gender, main pancreatic duct index, body mass index (BMI), intra-abdominal adipose tissue thickness, histopathological diagnosis, volume of intraoperative blood loss, preoperative amylase concentration in blood, inflammatory markers following surgery, invasion of the portal vein, along with additional variables documented throughout scientific literature[30-32]. This current research endeavor screened potential risk determinants according to both previous and updated definitional criteria, producing outcomes that align with findings from earlier research efforts[33-35]. The index of the main pancreatic duct and the CT value of pancreatic tissue emerged as independent factors associated with risk under both grading methodologies. The main pancreatic duct index - representing the proportional relationship between ductal diameter and pancreatic thickness[36-38]. The present investigation substituted pancreatic texture evaluation with pancreatic CT value measurement based on two fundamental considerations: (1) Pancreatic texture assessment lacks standardized evaluative criteria, predominantly relying on subjective tactile assessment during surgical procedures, which resists precise quantification; and (2) Pancreatic textural characteristics demonstrate correlation with measured CT values. Elevated pancreatic CT values indicate increased tissue density, more pronounced fibrotic changes, and consequently reduced risk of fistula formation. Among the 253 patients comprising the modeling cohort, cases developing pancreatic fistula demonstrated CT values measuring 36.3 ± 6.4 HU (according to 2005 edition criteria) and 36.2 ± 7.3 HU (according to 2016 edition criteria), whereas cases without fistula development exhibited measurements of 41.1 ± 9.6 HU (according to 2005 edition criteria) and 39.7 ± 8.8 HU (according to 2016 edition criteria), revealing statistically meaningful differences across both classification systems (P = 0.002 and P = 0.004, respectively)[36-38]. In this research investigation, conventional soft vs hard pancreatic texture classification was supplanted by pancreatic CT value measurement based on dual rationales: (1) The absence of universally acknowledged standards for pancreatic texture assessment, which traditionally relies on intraoperative tactile evaluation by the operating surgeon, introducing significant subjective variation that defies reliable quantification; and (2) The established correlation between pancreatic textural properties and pancreatic CT value measurements. One may reasonably conclude that increased pancreatic CT values correspond to greater pancreatic tissue density, more advanced pancreatic fibrosis, and consequently diminished risk of pancreatic fistula development. Within the modeling group comprising 253 patients in this investigation, patients developing pancreatic fistula demonstrated pancreatic CT values of 36.3 ± 6.4 HU (according to 2005 edition criteria) and 36.2 ± 7.3 HU (according to 2016 edition criteria). Conversely, patients without pancreatic fistula exhibited pancreatic CT values of 41.1 ± 9.6 HU (according to 2005 edition criteria) and 39.7 ± 8.8 HU (according to 2016 edition criteria). Statistically significant distinctions were observed between pancreatic CT values in patients with vs without pancreatic fistula according to different edition grading methodologies (P = 0.002 and P = 0.004). Our findings align with histopathological studies demonstrating that pancreatic fibrosis and fat infiltration significantly influence POPF risk. Strong correlation between CT density values and histological fibrosis scores (r = 0.72, P < 0.001). Compared to existing models, our AUC values of 0.788-0.804 are comparable to the widely-used Fistula Risk Score (FRS) which reports AUCs of 0.763-0.821 in validation studies. Based on our ROC analysis, we recommend the following risk-stratified management approach: Patients with predicted POPF probability > 32% should be considered high-risk, warranting prophylactic measures including somatostatin analog administration, extended drainage tube placement, and consideration of modified anastomotic techniques (duct-to-mucosa rather than end-to-side).

Throughout the previous decade, researchers domestically and internationally have formulated systems for predicting post-PD pancreatic fistula by integrating multiple fistula-related risk determinants. Trudeau et al[39] developed the FRS prediction scoring methodology (ranging from 0-10 points) incorporating pancreatic textural assessment, main pancreatic duct dimensional measurement, pathological findings, and intraoperative blood transfusion volume, stratifying patients into negligible risk (0 points), low risk (1-2 points), moderate risk (3-6 points), and high risk (7-10 points) categories. There is a study incorporated pancreatic textural assessment, main pancreatic duct dimensional measurement, and BMI, calculating fistula occurrence probability through a specific mathematical formula - representing a simplified, user-oriented alternative to the comprehensive FRS methodology[39]. First postoperative day intraperitoneal amylase concentration exceeding 2615 U/L as independent risk determinants, establishing a model with 0.888 predictive value. A scoring methodology utilizing exclusively first postoperative day measurements including abdominal drainage amylase ≥ 1000 IU/L and serum C-reactive protein ≥ 90 mg/L, achieving comparable predictive accuracy[40-43]. The parameters comprising this model derive entirely from preoperative CT imaging assessment, offering straightforward, objective, quantitative clinical application with robust reproducibility[41-43]. This predictive model provides meaningful clinical utility for preoperative risk assessment regarding pancreatic fistula development. Nevertheless, as this represents a single-center retrospective investigation, subsequent research initiatives should examine relationships between preoperative CT indicators and pancreatic fistula occurrence through large-sample prospective studies to further validate this prediction methodology.

Limitations and future directions

This single-center retrospective study has limited generalizability and potential selection bias. Our prediction model relies only on preoperative CT parameters, which may not capture important intraoperative variables affecting POPF development. The substitution of pancreatic CT values for subjective gland texture assessment requires further validation across different imaging protocols. We are planning a multicenter prospective validation study to confirm our model’s generalizability across different institutions. Future research will integrate dynamic intraoperative factors including real-time gland texture assessment and blood flow measurements through standardized protocols. We also plan to explore machine learning approaches to incorporate additional variables such as serum biomarkers (carbohydrate antigen 19-9, albumin), operative time, and surgeon experience for personalized risk prediction. These advances may enable more precise POPF prediction and tailored perioperative management strategies to improve surgical outcomes.

CONCLUSION

The main pancreatic duct index and pancreatic CT value are significant independent risk factors for pancreatic fistula development following PD. Our prediction model based on these parameters demonstrates strong predictive capability with AUCs of 0.788 and 0.804 in the modeling and validation groups, respectively.