Nomogram for prediction of six-month mortality following endovascular treatment of delayed post-pancreatectomy hemorrhage

Abstract

BACKGROUND

Delayed post-pancreatectomy hemorrhage (PPH) is life-threatening, and endovascular interventions show promise. This retrospective study aimed to evaluate endovascular treatment outcomes for delayed PPH and identify mortality risk factors.

AIM

To conduct a single-center retrospective study of 88 patients with delayed PPH to systematically evaluate the clinical efficacy of endovascular treatment, identify independent risk factors for six-month mortality, and propose and validate a predictive model for individualized management of high-risk patients.

METHODS

This retrospective analysis included 88 patients with delayed PPH treated by endovascular intervention. Patients were stratified into survival (n = 64) and mortality (n = 24) groups. Clinical and procedural variables were assessed using univariate and multivariate logistic regression. Significant predictors were incorporated into a prognostic nomogram. Model performance was assessed through discrimination (area under the receiver operating characteristic curve), calibration, and decision curve analysis.

RESULTS

Technical and clinical success rates were 92.0% and 60.2%, respectively. The overall six-month mortality rate was 27.3% (24/88). Independent predictors of mortality included advanced age, prolonged operative time, shorter hospital stay, intra-abdominal infection, coagulation dysfunction, common hepatic artery bleeding, and failure to achieve clinical success. The nomogram demonstrated excellent discrimination (area under the receiver operating characteristic curve = 0.943), with good calibration and favorable net benefit on decision curve analysis.

CONCLUSION

We proposed and validated a predictive nomogram for six-month mortality following endovascular treatment for delayed PPH. The model facilitates individualized risk stratification and may guide clinical decision-making. Early identification of high-risk patients - particularly older individuals or those with infection or coagulopathy - and prompt, personalized intervention may improve outcomes in this high-risk population.

Key Words: Delayed post-pancreatectomy hemorrhage; Endovascular intervention; Pancreaticoduodenectomy; Mortality; Clinical outcome; Nomogram

Core Tip: This retrospective study is the first to establish and validate a nomogram for predicting six-month mortality in patients with delayed post-pancreatectomy hemorrhage treated with endovascular intervention. By retrospectively analyzing 88 cases, several independent prognostic factors were identified, including advanced age, prolonged operative time, shorter in-hospital days, intra-abdominal infection, coagulation abnormalities, and clinical failure. Notably, bleeding from the common hepatic artery was confirmed as an independent predictor of mortality. The proposed nomogram demonstrated good discrimination and clinical applicability, offering a practical tool for individualized risk stratification and management in high-risk post-pancreatectomy hemorrhage patients.

INTRODUCTION

Pancreaticoduodenectomy (PD) is a complex and technically demanding procedure primarily indicated for malignancies of the pancreatic head, periampullary region, and distal bile duct[1]. Despite advances in surgical techniques and postoperative care, PD remains associated with substantial morbidity and mortality rates[2,3]. Among various complications, post-pancreatectomy hemorrhage (PPH) is one of the most devastating, with an incidence of 2%-12% and a mortality rate approaching 50%[4-8]. PPH is associated with increased early postoperative mortality and severe complications[9]. According to the International Study Group on Pancreatic Surgery (ISGPS) definition, PPH is classified by timing (early ≤ 24 hours vs delayed > 24 hours), location (intraluminal vs extraluminal), and severity (grades A-C)[10]. Although this classification remains the clinical standard, recent studies have highlighted its limitations in discriminating clinical severity. For instance, grade A hemorrhage is generally not associated with significantly different outcomes compared with patients without PPH, whereas delayed grade C hemorrhage accompanied by clinically relevant postoperative pancreatic fistula (POPF) is linked to markedly increased mortality and prolonged hospital stay[11-13]. Consequently, refinements incorporating factors such as clinically relevant POPF or sentinel bleeding have been proposed to enhance risk stratification and clinical applicability. Delayed PPH typically results from vessel erosion due to anastomotic leakage, intra-abdominal infection, intraoperative vascular injury or postoperative pseudoaneurysm formation[14,15].

With the widespread adoption of endovascular techniques, vascular intervention has become the first-line therapy for postoperative PPH after PD[16-19]. Endovascular treatment is preferred whenever feasible; however, urgent laparotomy remains indicated in cases of anastomotic breakdown, refractory pancreatic fistula, or hemodynamic instability[20]. Although endovascular hemostasis rates exceed 90%[5,8,21,22], clinical outcomes exhibit a biphasic distribution: Some patients achieve definitive cure, whereas others progress to multi-organ failure and death[23]. This variability underscores the intricate pathophysiology driving PPH-related mortality and highlights the need to identify the principal determinants of outcome.

Previous studies have identified patient-related risk factors for PPH, including advanced age, comorbidities, and poor nutritional status, as well as factors such as prolonged operative time[24], excessive intraoperative blood loss[4,6,14], and vascular manipulation[2,25,26]. However, most studies have focused on overall bleeding incidence rather than outcome-specific mortality[8,9,27]. Delayed PPH, which is generally more lethal than early hemorrhage, remains poorly characterized regarding bleeding site and predictors of death[19,28-30]. Many studies are further limited by small sample sizes, heterogeneous patient populations, or lack of integration between clinical and procedural variables[11-13]. To address these gaps, we conducted a single-center retrospective study of 88 patients with delayed PPH to: (1) Systematically evaluate the clinical efficacy of endovascular treatment in delayed PPH; (2) Identify independent risk factors for six-month mortality; and (3) Propose and validate a predictive model to inform individualized management of high-risk patients.

MATERIALS AND METHODS

Ethics approval

This study was approved by the Institutional Review Board of Beijing Chao-Yang Hospital, Capital Medical University (Approval No. 2024-6-24-2). Given the retrospective nature of the analysis, the requirement for informed consent was waived.

Study design and patient selection

We retrospectively reviewed 88 patients who developed delayed PPH following PD between 2014 and 2024 and subsequently underwent endovascular intervention. Sample size was estimated based on previously reported mortality of delayed PPH (approximately 27%) using the formula: n = [Z² · p (1 - p)]/d².

This calculation yielded a minimum required sample size of 76 patients. Our group included 88 patients, thereby meeting this requirement. Patients were grouped solely based on 6-month post-PPH survival: Survival group (n = 64) and mortality group (n = 24). No matching on baseline characteristics or covariates was performed; the survival-to-death ratio reflects the observed outcomes.

Data were retrieved from electronic medical records and included demographics, preoperative comorbidities, operative details, characteristics of the PPH event, and clinical outcomes. We also collected information on length of hospital stay, incidence of POPF, coagulopathy, intra-abdominal infection, and other perioperative complications. Based on post-PPH survival, patients were allocated to the survival group (n = 64) or the mortality group (n = 24).

Inclusion criteria: (1) Delayed PPH as defined by ISGPS guidelines (> 24 hours post-PD); and (2) Complete perioperative and intervention records.

Exclusion criteria: (1) Early PPH (≤ 24 hours post-PD); (2) Hemorrhage unrelated to PD (e.g., postoperative trauma or non-pancreatic vascular lesions); and (3) A flowchart illustrating patient selection and group allocation is shown in Figure 1.

Figure 1 Flow diagram of patient selection and treatment allocation for delayed post-pancreatectomy hemorrhage. PPH: Post-pancreatectomy hemorrhage; PD: Pancreaticoduodenectomy; DSA: Digital subtraction angiography.

Definitions and grading

All blood test results were obtained at the time of bleeding onset. PPH was classified per the 2007 ISGPS consensus by timing (early vs delayed), location, and severity. Early PPH (≤ 24 hours post-PD). Delayed PPH (> 24 hours post-PD).

Bleeding was further categorized as intraluminal (hematemesis or melena) or extraluminal (hemorrhagic drain output or unexplained hemodynamic instability). Severity was graded as: Mild: Hemoglobin drop or responsive to ≤ 3 units packed red blood cells; Severe: Hemoglobin drop, hemodynamic instability, and requirement for substantial transfusion. Overall severity and timing were combined into grade A (early mild), grade B (early severe or late mild), and grade C (late severe)[10].

Digital subtraction angiography findings

Positive: Contrast extravasation or pseudoaneurysm (direct), arterial spasm or irregular vessel contour (indirect). POPF was defined by the 2016 ISGPS update as drain fluid amylase > 3 × upper limit of normal, graded as biochemical leak, grade B, or grade C. Bile leakage was defined as bilious drain output persisting beyond postoperative day 5. Intra-abdominal infection required persistent fever, leukocytosis, purulent drainage, and radiologic evidence of fluid collection. Coagulopathy was defined as activated partial thromboplastin time (APTT) > 40 seconds, platelet count < 80 × 10⁹/L, or international normalized ratio (INR) > 1.5 at the time of PPH occurrence.

Surgical and interventional procedures: All PDs (classic Whipple or pylorus-preserving pancreaticoduodenectomy) were performed by experienced hepato-pancreato-biliary surgeons. Vascular resection and reconstruction were undertaken when indicated, with duct-to-mucosa pancreaticojejunostomy and routine internal stenting.

PPH management followed institutional protocol: Conservative measures (transfusion, hemostatic agents, somatostatin analogues) for hemodynamically stable or low-grade hemorrhage. Endoscopic, angiographic, or surgical hemostasis for grade B/C PPH (hemoglobin decline > 30 g/L in 24 hours with hypovolemic shock or failed conservative therapy). Urgent digital subtraction angiography (DSA) localized bleeding via selective catheterization of the celiac trunk and superior mesenteric artery. Active bleeding was treated with transcatheter arterial embolization using coils, gelatin sponge, or n-butyl cyanoacrylate; covered stent-grafts were deployed when vessel patency was critical (e.g., common hepatic artery; Figure 2). Empiric embolization was considered if DSA was negative but clinical signs persisted. Repeat DSA or surgery was reserved for clinical deterioration or recurrent hemorrhage with hemodynamic instability.

Figure 2 Endovascular management of post-pancreaticoduodenectomy arterial pseudoaneurysms. A and B: A 61-year-old male patient who underwent radical pancreaticoduodenectomy for bile duct cancer presented with acute hemorrhage from the gastroduodenal artery stump on postoperative day 27. Conventional hepatic angiography demonstrated a large pseudoaneurysm at the gastroduodenal artery stump with active contrast extravasation, associated with a drop in hemoglobin concentration (A); targeted embolization of the common hepatic artery achieved successful hemostasis (B); C and D: A 62-year-old male patient underwent radical pancreaticoduodenectomy combined with segmental resection of the transverse colon for a duodenal malignancy. Postoperatively, he developed intra-abdominal infection and presented with gastrointestinal bleeding on postoperative day 14, receiving both interventional treatment and laparotomy for debridement and drainage. The patient ultimately died from respiratory failure on postoperative day 77. Digital subtraction angiography revealed a pseudoaneurysm of the common hepatic artery (C); after placement of a covered stent in the common hepatic artery, follow-up angiography confirmed complete exclusion of the pseudoaneurysm while preserving distal hepatic arterial flow (D).

Outcome measures and follow-up: Patients were followed from the date of PD until death or six months after surgery. Primary outcome: Mortality or survival within six months after endovascular management of delayed PPH. Secondary outcomes: Technical and clinical success rates of intervention, and rebleeding incidence. Technical success: Immediate angiographic occlusion of the bleeding vessel without residual hemorrhage. Clinical success: Hemodynamic stability, rising hemoglobin, and absence of rebleeding within 30 days (complete vs partial success). Rebleeding: Recurrence requiring ≥ 2 units packed red blood cells or repeat endoscopic, endovascular, or surgical intervention within 30 days.

Statistical analysis

All analyses were conducted using R (version 4.3.2, R Core Team, Vienna, Austria). Continuous variables were assessed for normality using the Shapiro-Wilk test. Variables with approximately normal distribution are presented as mean ± SD, while non-normally distributed variables are presented as median (interquartile range). Categorical variables are expressed as counts and percentages [n (%)]. Group comparisons were performed using the Student’s t-test for normally distributed continuous variables and the Mann-Whitney U test for non-normally distributed continuous variables. Categorical variables were compared using the χ² test or Fisher’s exact test, as appropriate.

Variables with a P < 0.1 in the univariate analysis were subjected to collinearity diagnostics and subsequently entered into multivariate logistic regression to identify independent predictors of mortality. The threshold of P < 0.1 was used only for screening variables into the multivariate model, following the purposeful selection strategy to avoid prematurely excluding potentially important predictors[31]. Statistical significance in the final model was defined as P < 0.05. Multicollinearity was assessed using the variance inflation factor (VIF), with a VIF > 5 indicating potential multicollinearity[32,33]. In our study, all variables included in the multivariate model had VIF < 5 in Table 1, indicating no significant multicollinearity. Backward elimination was performed based on the Akaike information criterion. Variables retained to minimize the Akaike information criterion with P values between 0.05 and 0.1 were reported as contributing to the optimal model fit.

Table 1 Variance inflation factor values for variables in multivariate analysis.

Variable	VIF
Age	1.38
In-hospital days	1.28
Operating duration	1.35
Intra-abdominal infection	1.22
APTT	2.71
INR	2.53
Coagulopathy	1.87
Clinical success	1.35
CHA bleeding	1.05

Variance inflation factor values < 5 indicate no significant multicollinearity among the variables included in the multivariate regression model. VIF: Variance inflation factor; APTT: Activated partial thromboplastin time; INR: International normalized ratio; CHA: Celiac hepatic artery.

RESULTS

Clinical characteristics and outcomes

A total of 88 patients were analyzed, including 64 in the PPH survival group and 24 in the PPH mortality group (Table 2). Survivors were younger (62.6 ± 9.9 years vs 67.6 ± 9.0 years; P = 0.028) and had longer median hospital stays (50 days vs 39 days; P = 0.01). Other baseline demographics, comorbidities, pathology types, and preoperative laboratory values, including total and direct bilirubin, did not differ significantly between groups.

Table 2 Patient demographics and perioperative characteristics stratified by survival outcome, mean ± SD/n (%)/median (interquartile range).

Characteristic	Total (n = 88)	PPH survival group (n = 64)	PPH mortality group (n = 24)	P value
Age, years	64.0 ± 9.9	62.6 ± 9.9	67.6 ± 9	0.028
Gender				0.218
Male	56 (63.6)	38 (59.4)	18 (75)
Female	32 (36.4)	26 (40.6)	6 (25)
Hypertension	43 (48.9)	35 (54.7)	8 (33.3)	0.095
Diabetes	11 (12.5)	5 (7.8)	6 (25)	0.063
In-hospital days	45 (36.5-63.25)	50 (38-64.25)	39 (24.25-51.5)	0.01
Pathology diagnosis				0.223
PDCA	23 (26.1)	16 (25.0)	7 (29.2)
CBD cancer	34 (38.6)	21 (32.8)	13 (54.2)
Ampullary cancer	5 (5.7)	4 (6.2)	1 (4.2)
Duodenal cancer	13 (14.8)	10 (15.6)	3 (12.5)
Other malignancies	7 (8.0)	7 (10.9)	0 (0)
Other benign disease	6 (6.8)	6 (9.4)	0 (0)
Preoperative total bilirubin (mmol/L)	62.1 (19.8-127.2)	62.5 (19-131.1)	62.0 (20.7-119.5)	0.892
Preoperative direct bilirubin (mmol/L)	42.5 (14.2-101.1)	48.0 (13.6-108)	39.5 (17.1-72.2)	0.818
Operating duration (minutes)	655.0 (540.0-753.8)	610.0 (498.8-727.5)	697.5 (611.2-768.8)	0.065
Perioperative complications
POPF	58 (65.9)	38 (59.4)	20 (83.3)	0.044
Gastrointestinal fistula	5 (5.7)	4 (6.2)	1 (4.2)	1
Bile leakage	5 (5.7)	9 (14.1)	5 (20.8)	0.516
Intra-abdominal infection	43 (48.9)	27 (42.2)	16 (66.7)	0.056
Hemorrhage details
Bleeding latency, days	15 (11-23)	15 (10-23)	17.5 (13.2-19.5)	0.422
Hemoglobin concentration(g/L)	74.7 ± 15.7	75.2 ± 14.8	73.5 ± 18.3	0.688
PLT (109/L)	250.6 ± 101.9	262.2 ± 90.8	219.5 ± 123.7	0.133
APTT (s)	32.8 (27.6-41.8)	32.3 (26.5-39.3)	37.1 (30.2-49.8)	0.037
INR	1.2 (1.1-1.3)	1.1 (1.1-1.2)	1.3 (1.2-1.7)	< 0.001
Coagulopathy	31 (35.2)	17 (26.6)	14 (58.3)	0.011
Severity of PPH				0.088
Grade B	36 (40.9)	30 (46.9)	6 (25)
Grade C	52 (59.1)	34 (53.1)	18 (75)

PDCA: Pancreatic ductal adenocarcinoma; CBD: Common bile duct; POPF: Postoperative pancreatic fistula; PLT: Platelet count; APTT: Activated partial thromboplastin time; INR: International normalized ratio; PPH: Post-pancreatectomy hemorrhage.

Perioperative variables and complications

POPF occurred more frequently in non-survivors (83.3% vs 59.4%; P = 0.044). Median operative duration tended to be longer in non-survivors. Other perioperative complications, including intra-abdominal infection, bile leakage, and gastrointestinal fistula, showed non-significant trends and are summarized in Table 2.

Hemorrhage characteristics

Non-survivors had worse coagulation profiles, with longer APTT (37.1 seconds vs 32.3 seconds; P = 0.037), higher INR (1.3 vs 1.1; P < 0.001), and more frequent coagulopathy (58.3% vs 26.6%; P = 0.011). Grade C PPH was more common in non-survivors, whereas hemoglobin and platelet counts showed no significant differences.

Angiographic findings and embolization

Table 3 summarizes angiographic findings, embolization strategies, and clinical outcomes in our group. The most common bleeding site was the common hepatic artery. Contrast medium overflow on angiography (58.3% vs 26.6%; P = 0.011) and hemorrhagic shock as the onset symptom (58.3% vs 31.2%; P = 0.027) were significantly more frequent in non-survivors. Targeted embolization, predominantly using coils, was performed in most patients. Technical success was high (92%), and clinical success was significantly higher in survivors (71.9% vs 29.2%; P < 0.001).

Table 3 Angiographic findings, embolization strategies, and clinical outcomes in patients with delayed post-pancreatectomy hemorrhage, n (%).

Characteristic	Total	PPH survival group	PPH mortality group	P value
Bleeding site on angiogram	92 (100)	68 (73.9)	24 (26.1)
Negative	18 (20.5)	15 (23.4)	3 (12.5)	0.376
Common hepatic artery	23 (26.1)	13 (20.3)	10 (41.7)	0.057
Stump of gastroduodenal artery	16 (18.2)	11 (17.2)	5 (20.8)	0.759
SMA	8 (9.1)	7 (10.9)	1 (4.2)	0.438
Proper hepatic artery	13 (14.8)	10 (15.6)	3 (12.5)	1
Right hepatic artery	3 (3.4)	2 (3.1)	1 (4.2)	1
Left hepatic artery	4 (4.5)	3 (4.7)	1 (4.2)	1
Left gastric artery	5 (5.7)	5 (7.8)	0 (0)	0.317
Splenic artery	1 (1.1)	1 (1.6)	0 (0)	1
Pancreaticoduodenal artery	1 (1.1)	1 (1.6)	0 (0)	1
Embolism site on angiogram	92 (100)	68 (73.9)	24 (26.1)
Conservative treatment	15 (17)	13 (20.3)	2 (8.3)	0.222
Common hepatic artery	47 (53.4)	30 (46.9)	17 (70.8)	0.056
Stump of gastroduodenal artery	1 (1.1)	1 (1.6)	0 (0)	1
SMA	7 (8)	7 (10.9)	0 (0)	0.183
Proper hepatic artery	7 (8)	5 (7.8)	2 (8.3)	1
Right hepatic artery	3 (3.4)	2 (3.1)	1 (4.2)	1
Left hepatic artery	3 (3.4)	2 (3.1)	1 (4.2)	1
Left gastric artery	7 (8)	6 (9.4)	1 (4.2)	0.669
Splenic artery	1 (1.1)	1 (1.6)	0 (0)	1
Pancreaticoduodenal artery	1 (1.1)	1 (1.6)	0 (0)	1
Angiographic sign	103 (100)	73 (70.9)	30 (29.1)
Negative	16 (18.2)	15 (23.4)	1 (4.2)	0.059
Contrast medium overflow	31 (35.2)	17 (26.6)	14 (58.3)	0.011
Pseudoaneurysm	28 (31.8)	20 (31.2)	8 (33.3)	1
Vessel stenosis	18 (20.5)	12 (18.8)	6 (25)	0.559
Stump of gastroduodenal artery	8 (9.1)	7 (10.9)	1 (4.2)	0.438
Suspicious of contrast agent staining	2 (2.3)	2 (3.1)	0 (0)	1
Onset symptom of hemorrhage	126	88	38
Pseudoaneurysm detected on postoperative follow-up CT	3 (3.4)	3 (4.7)	0 (0)	0.559
Melena or/and hematemesis	33 (37.5)	24 (37.5)	9 (37.5)	1
Sentinel bleeding	56 (63.6)	41 (64.1)	15 (62.5)	1
Hemorrhagic shock	34 (38.6)	20 (31.2)	14 (58.3)	0.027
Strategy of embolization	88	64	24	0.18
Targeted embolization	58 (65.9)	39 (60.9)	19 (79.2)
Empiric embolization	11 (12.5)	8 (12.5)	3 (12.5)
No embolization	19 (21.6)	17 (26.6)	2 (8.3)
Embolic material	74	51	23
Coils	60 (68.2)	40 (62.5)	20 (83.3)	0.075
Stent	8 (9.1)	7 (10.9)	1 (4.2)	0.438
Gelatin sponges	4 (4.5)	3 (4.7)	1 (4.2)	1
NBCA	2 (2.3)	1 (1.6)	1 (4.2)	0.473
Number of post-Whipple open abdominal surgeries				0.068
None	51 (58)	41 (64.1)	10 (41.7)
One time	32 (36.4)	21 (32.8)	11 (45.8)
Two times	5 (5.7)	2 (3.1)	3 (12.5)
Tech success	81 (92)	60 (93.8)	21 (87.5)	0.385
Clinical success	53 (60.2)	46 (71.9)	7 (29.2)	< 0.001
Re-bleeding	20 (22.7)	14 (21.9)	6 (25)	0.779

SMA: Superior mesenteric artery; CT: Computed tomography; NBCA: N-butyl cyanoacrylate; PPH: Post-pancreatectomy hemorrhage.

Univariate and multivariate analysis

Univariate analysis identified age, diabetes, in-hospital days, POPF, intra-abdominal infection, APTT, INR, coagulopathy, common hepatic artery bleeding, and hemorrhagic shock as predictors of six-month mortality (Table 4).

Table 4 Univariate and multivariate logistic regression analysis of mortality risk factors in patients with delayed post-pancreatectomy hemorrhage.

Characteristics	Univariate analysis		Multivariate analysis
	P value	OR (95%CI)	P value	OR 95%CI
Gender	0.1787	1.143 (0.942-1.388)
Age	0.0327	0.990 (0.980-0.999)	0.0018	1.014 (1.005-1.022)
Hypertension	0.0758	1.185 (0.985-1.425)
Diabetes mellitus	0.0300	0.732 (0.555-0.966)
In-hospital days	0.0301	1.003 (1.000-1.006)	< 0.001	0.996 (0.993-0.998)
Pathology diagnosis	0.5150	0.925 (0.732-1.169)
Preoperative total bilirubin	0.6898	1 (0.999-1.001)
Preoperative direct bilirubin	0.4815	1 (0.999-1.002)
Operating duration	0.0760	1 (0.999-1)	0.0044	1.001 (1-1.001)
POPF	0.0350	0.8 (0.667-0.982)
Gastrointestinal fistula	0.7108	1.1 (0.72-1.622)
Bile leakage	0.445	0.9 (0.7-1.169)
Intra-abdominal infection	0.0412	0.8 (0.685-0.99)	0.0053	1.253 (1.074-1.462)
Bleeding latency, days	0.5931	1 (0.997-1.005)
Hemoglobin	0.6565	1 (0.995-1.007)
PLT	0.0799	1 (1-1.002)
APTT	0.0379	1 (0.988-1)	0.0159	0.991 (0.984-0.998)
INR	0.0001	0.7 (0.565-0.819)	0.0113	1.374 (1.081-1.746)
Coagulopathy	0.0051	0.8 (0.629-0.916)	0.0732	1.204 (0.985-1.471)
Severity of PPH-grade B or C	0.0642	0.8 (0.693-1.008)
Asymptomatic hemorrhage	0.0372	1.3 (1.019-1.64)
Pseudoaneurysm detected on postoperative follow-up CT	0.2858	1.3 (0.792-2.22)
Melena or/and hematemesis	1.0000	1 (0.823-1.215)
Sentinel bleeding	0.8936	1 (0.833-1.232)
Hemorrhagic shock	0.0200	0.8 (0.661-0.961)
Strategy of embolization	0.0615	0.8 (0.636-1.008)
Negative on angiography	0.0372	1.3 (1.019-1.64)
Suspicious of contrast agent staining	0.3868	1.3 (0.705-2.479)
Vessel stenosis	0.5229	0.9 (0.734-1.17)
Pseudoaneurysm	0.8538	1 (0.802-1.201)
Contrast medium overflow	0.0051	0.8 (0.629-0.916)
Stump of gastroduodenal artery sign	0.3308	1.2 (0.849-1.629)
Conservative treatment	0.1873	1.2 (0.923-1.516)
Common hepatic artery	0.0454	0.8 (0.687-0.993)	0.0256	1.182 (1.023-1.364)
SMA	0.0933	1.3 (0.955-1.894)
Proper hepatic artery	0.9368	1 (0.696-1.396)
Right hepatic artery	0.8131	0.9 (0.559-1.577)
Left hepatic artery	0.8131	0.9 (0.559-1.577)
Left gastric artery	0.4271	1.2 (0.814-1.629)
Coils	0.0628	0.8 (0.678-1.008)
Stent	0.3308	1.2 (0.849-1.629)
Gelatin sponges	0.9180	1 (0.652-1.609)
NBCA	0.4711	0.8 (0.422-1.488)
Number of post-Whipple interventional embolization	0.6321	1.1 (0.819-1.391)
Number of post-Whipple open abdominal surgeries	0.1404	0.9 (0.71-1.048)
Tech success	0.3402	1.2 (0.838-1.674)
Clinical success	0.0002	1.4 (1.193-1.7)	0.0167	0.813 (0.688-0.96)
Re-bleeding	0.7587	1 (0.771-1.208)

POPF: Postoperative pancreatic fistula; PLT: Platelet count; APTT: Activated partial thromboplastin time; INR: International normalized ratio; PPH: Post-pancreatectomy hemorrhage; CT: Computed tomography; SMA: Superior mesenteric artery; NBCA: N-butyl cyanoacrylate; OR: Odds ratio; CI: Confidence interval.

Multivariate logistic regression confirmed several independent predictors. Advanced age increased mortality risk, reflecting reduced physiological reserve and higher susceptibility to postoperative complications. Longer hospital stays and prolonged operative duration indicated greater surgical complexity, increased intraoperative blood loss, and anesthesia burden, contributing to poorer outcomes. Intra-abdominal infection significantly elevated mortality risk, highlighting the detrimental effects of septic complications on hemodynamic stability and vascular integrity. Coagulation abnormalities, including elevated APTT and INR, indicated impaired hemostasis and higher bleeding risk; prolonged APTT may reflect anticoagulation or postoperative management, whereas elevated INR often signals impaired hepatic function. Coagulopathy showed a trend toward increased risk, suggesting potential clinical relevance.

Bleeding from the common hepatic artery carried particularly high risk due to challenges in endovascular management and limited hepatic tolerance to arterial occlusion, potentially leading to ischemia or liver dysfunction. Failure to achieve clinical success after embolization strongly predicted mortality, emphasizing the importance of effective hemostasis and timely intervention.

Although coagulopathy did not reach conventional statistical significance [odds ratio = 1.204; 95% confidence interval (CI): 0.985-1.471; P = 0.0732], its trend toward increased risk, consistent with univariate analysis (P = 0.0051), warrants further investigation in larger groups. Overall, these findings highlight key patient- and procedure-related factors that can guide risk stratification, individualized management, and early intervention in high-risk delayed PPH patients.

Nomogram model of delayed PPH

A nomogram was constructed to predict the probability of six-month mortality following endovascular treatment for delayed PPH, based on the significant predictors identified above (Figure 3). Each predictor (such as age, operative time, hospital stay, intra-abdominal infection, presence of coagulopathy, celiac hepatic artery (CHA) bleeding, and clinical success) contributes a number of points according to its value. Summing these points yields a total score, which corresponds to the estimated six-month mortality risk.

Figure 3 Nomogram for predicting the probability of mortality within 6 months following endovascular treatment for delayed post-pancreatectomy hemorrhage. Locate the corresponding value on the horizontal axis of age, draw a vertical line upward on this horizontal axis to determine the points obtained for this patient, and repeat this process for the horizontal axis of other predictors. Then, sum the points of the nine parameters to calculate the total points. Find its position on the horizontal axis of total points and draw a vertical line downward to determine the risk of mortality within 6 months after endovascular management of delayed post-pancreatectomy hemorrhage. APTT: Activated partial thromboplastin time; INR: International normalized ratio.

Model performance and internal validation

As shown in Figure 4, the performance of the nomogram was evaluated in terms of discrimination, calibration, and clinical utility. The model demonstrated excellent discrimination with area under the receiver operating characteristic curve of 0.943 (95%CI: 0.781-0.958). Calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test, which produced a non-significant P value of 0.1003, suggesting good agreement between predicted and observed outcomes. In the calibration plot, both the apparent curve and the bias-corrected curve closely followed the ideal 45-degree line. Additionally, decision curve analysis demonstrated that the nomogram provided a favorable net benefit across a range of threshold probabilities, indicating promising clinical applicability.

Figure 4 Performance evaluation of the nomogram predicting six-month mortality after endovascular treatment for delayed post-pancreatectomy hemorrhage. A: Receiver operating characteristic curve showing an area under the receiver operating characteristic curve of 0.943 (95% confidence interval: 0.781-0.958), indicating excellent discrimination; B: Calibration plot comparing predicted vs observed probabilities. Both the apparent and bias-corrected curves align closely with the ideal line, suggesting good calibration; C: Decision curve analysis demonstrating the clinical utility of the model, with greater net benefit across a range of threshold probabilities compared to treat-all or treat-none strategies. ROC: Receiver operating characteristic; AUC: Area under curve.

DISCUSSION

Delayed PPH is an uncommon but highly lethal complication. In our series of 88 patients treated with endovascular intervention, we identified several independent predictors of six-month mortality by examining both patient-specific factors and details of the interventional approach. Our goal was to inform more tailored management strategies for this high-risk population.

In our group, endovascular treatment achieved a technical success rate of 92% with an overall mortality rate of 27%, consistent with published reports[20]. Importantly, patients who achieved clinical success - defined as no rebleeding or need for repeat intervention within 30 days - had significantly lower mortality (12.5%) compared to those with clinical failure (66.7%; P < 0.001), underscoring its prognostic value. This finding aligns with Wolk et al[8], who reported that the need for repeat intervention was associated with higher in-hospital mortality. In our study, the primary indications for re-intervention were recurrent bleeding and stent thrombosis. Paradoxically, the main causes of death were multi-organ failure, hepatic ischemia or failure, and infectious complications, rather than uncontrolled hemorrhage. These observations suggest that, beyond achieving angiographic hemostasis, proactive management of bleeding-related organ dysfunction is essential for improving outcomes. Feng et al[27] similarly identified hemorrhagic shock and rebleeding as key predictors of death. Although hemorrhagic shock at presentation differed between our groups, it did not emerge as an independent risk factor in multivariate analysis. Nevertheless, rapid recognition and treatment of shock remain critical, since untreated shock can precipitate the lethal triad of coagulopathy, acidosis, and hypothermia, leading to multi-organ failure[34]. Likewise, prompt control of arterial bleeding - ideally before a substantial hemoglobin drop or massive transfusion - is vital for securing clinical success[35].

Previous studies have identified malignant disease, grade C pancreatic fistula, intra-abdominal infection, and concurrent intra- or extraluminal bleeding as predictors for clinical failure[18]. Gaudon et al[36] reported a technical success rate of 66.5%, a clinical success rate of 85%, and a mortality rate of 14.3% in a group of 35 PPH patients. They found that anatomical variants of the celiac artery - such as median arcuate ligament syndrome or acute arterial angulation - were independent predictors of mortality, likely due to higher rates of technical failure, rebleeding, and complications. Vascular tortuosity or variations increase procedural complexity and the risk of incomplete embolization or arterial injury. To mitigate these risks, preoperative multidetector computed tomography angiography for detailed vascular mapping and a tailored endovascular approach (e.g., brachial access with covered stent-graft placement) have been recommended. This approach may contribute to higher clinical success rates and lower mortality. In hemodynamically stable patients presenting with delayed postpancreatectomy arterial bleeding, computed tomography angiography remains an essential diagnostic tool, achieving a sensitivity of 79%-92% and a specificity of 92%-95% for hemorrhage exceeding 0.3 mL/minute[37-39]. Upon localization of the bleeding site, urgent celiac-mesenteric DSA and endovascular hemostasis are advocated as the cornerstone of treatment[21].

In our group, the clinical success rate was relatively low (60.2%) and the mortality rate relatively high (27.3%). This may be partly attributable to the lack of routine pre-procedural vascular imaging and individualized intervention strategies. We encountered technical challenges in patients with complex vascular anatomy, which likely contributed to suboptimal outcomes. These findings underscore the importance of comprehensive vascular assessment prior to intervention to guide optimal access routes and embolization techniques. Implementing individualized strategies may improve both technical and clinical success rates and reduce mortality risk.

In this study, 18 patients presented with negative angiographic findings, 11 of these underwent empirical embolization, while the remainder were managed conservatively. Empirical embolization had no significant impact on mortality, consistent with prior reports[40]. Although empirical arterial embolization or stent placement can be an effective hemostatic option for patients with negative angiography[21,41], it is associated with increased risk of complications such as hepatic infarction or abscess, delayed bile duct stricture, and rebleeding[42]. The role of empirical intervention in angiographically negative cases remains controversial. Angiographically negative results may be due to low bleeding rates, intermittent hemorrhage, or shock-induced arterial vasospasm, all of which can reduce blood flow below the detection threshold of DSA[17]. In such instances, if the bleeding source is not identified on the initial DSA, repeating the examination 6-24 hours later is recommended in hemodynamically stable patients[43]. Some investigators advocate for transcatheter embolization of the suspected area regardless of subsequent angiographic results if rebleeding occurs[44].

Patient factors, such as advanced age (often defined as ≥ 70 years), intra-abdominal infection, and coagulation disturbances, emerged as key predictors of mortality, in agreement with previous studies[35,41]. Intra-abdominal abscesses can perpetuate vascular erosion, thereby increasing the risk of grade C hemorrhage, rebleeding, and death[45]. Similarly, hemorrhagic shock compounded by coagulopathy at the time of intervention is associated with a dismal prognosis, as persistent bleeding can precipitate disseminated intravascular coagulation. Even when local hemostasis is achieved, deranged coagulation may lead to multi-organ failure and mortality[30]. Conversely, if severe coagulopathy prevents effective hemorrhage control, patients may succumb to massive intraoperative blood loss and hemorrhagic shock[27]. Interestingly, our analysis found that prolonged APTT as a protective prognostic factor, while elevated INR was significantly associated with worse outcomes. This likely reflects the divergent pathophysiological implications of these coagulation markers. Prolonged APTT may indicate effective anticoagulation and vigilant postoperative management, contributing to improved prognosis. In contrast, elevated INR signifies impaired hepatic synthetic function and an increased bleeding risk, correlating with poorer outcomes. Therefore, individual coagulation parameters should be analyzed separately and preferably treated as continuous variables to more accurately assess their prognostic value.

Both operative duration and length of hospital stay emerged as significant predictors of mortality. Prolonged operative time has been identified as a risk factor for death in patients with aneurysmal hemorrhage[24]. Prolonged surgery, often associated with more complex cases, increased blood loss, and longer anesthesia times, are frequently linked to aggressive lymphadenectomy and vascular skeletonization - maneuvers that may cause intimal or adventitial injury. In the context of postpancreatectomy patients, longer operative times have been associated with higher rates of complications (such as surgical site infection, thromboembolism, pneumonia, and other morbidities) and increased mortality. Previous studies have demonstrated a linear relationship between operative time and mortality[25,26]. In contrast, a longer hospital stay appeared protective in our analysis, likely reflecting survivor bias: Critically ill patients who die early have shorter hospitalizations. As Dusch et al[46] reported, grade C PPH leads to longer intensive care unit stays but shorter overall hospitalization, indicating early mortality in the sickest patients. This skews the association between length of stay and mortality. Thus, an extended hospital stay does not by itself reduce mortality, but rather indicates that survivors benefit from additional time for recovery and comprehensive care.

Notably, our study is among the first to systematically quantify the prognostic implications of bleeding originating from the CHA in patients with delayed PPH. CHA involvement was observed in 26.1% of all cases and was significantly more common in the mortality group. Multivariate analysis confirmed CHA bleeding as an independent predictor of mortality (odds ratio = 1.182, 95%CI: 1.023-1.364, P = 0.0256). These findings highlight the unique vulnerability of the CHA in the context of post-pancreatic resection hemodynamic instability and infection-driven arterial erosion.

Although previous studies have frequently identified the CHA as a common source of bleeding, few have systematically examined its independent impact on clinical outcomes[5]. The underlying mechanism may be related to the limited tolerance of the liver to CHA embolization, where arterial occlusion compromises hepatic perfusion, potentially resulting in ischemia, infarction, or liver failure[28,47]. Covered stent-grafts theoretically mitigate this risk by achieving hemostasis while preserving arterial flow, thus optimizing outcomes; However, their application may be limited by anatomical or technical factors[30]. In our study, embolization was performed only when stent placement was not feasible. It is also important to consider that the liver typically receives approximately 30% of its blood supply from the hepatic artery and 70% from the portal vein. Consequently, adequate portal flow, variant arterial anatomy, or development of collateral circulation can partially compensate for arterial occlusion, thereby reducing morbidity and mortality[29,48]. However, in patients with preexisting liver conditions (such as cirrhosis) or prior hepatic surgery, the risk complications from CHA embolization are significantly increased[44].

Clinically, a comprehensive, multi-dimensional risk assessment is essential for managing PPH. The predictive nomogram developed based on the identified factors demonstrated excellent predictive performance. Decision curve analysis indicated that the nomogram possesses substantial clinical utility, offering effective support for individualized risk assessment in patients with delayed PPH. Importantly, the variables included in the model are affordable and readily accessible, facilitating broad application in clinical practice. Therefore, in cases of suspected or confirmed CHA bleeding, the choice of intervention should carefully balance the need for urgent hemostasis against the risk of hepatic hypoperfusion. Whenever feasible, super-selective embolization or covered stent placement should be prioritized, with close postoperative monitoring of liver function.

Delayed PPH is influenced not only by independent risk factors but also by complex interactions among them. In our study, advanced age, coagulation dysfunction, intra-abdominal infection, and CHA bleeding may synergistically exacerbate patient vulnerability. For example, intra-abdominal infection can aggravate vascular erosion and endothelial fragility, which is further amplified in patients with pre-existing coagulopathy, consistent with prior studies showing that systemic inflammation and coagulation abnormalities interact to worsen vascular outcomes[49,50]. Similarly, CHA bleeding may have a more severe impact in the context of systemic inflammation or impaired coagulation, increasing the risk of multi-organ dysfunction and mortality[51]. Prolonged operative time may further interact with these conditions by promoting hemodynamic instability, oxidative stress, and inflammatory cascades, compounding adverse outcomes[52]. These complex interplays suggest that simple additive risk models may underestimate true patient vulnerability, highlighting the need for mechanistic studies to clarify how these factors modulate one another, as recommended in previous literature[53]. A better understanding of these interactions could guide more precise, individualized perioperative management, such as early correction of coagulation deficits in infected patients or preemptive interventions at high-risk vascular sites like the CHA. Future studies should focus on elucidating these mechanisms to refine risk stratification and improve outcomes in patients with delayed PPH.

This study has several limitations. This single-center, retrospective study with a relatively small sample size may limit the generalizability of our findings, and the lack of long-term follow-up precludes evaluation of survival and quality of life beyond six months. Although we developed and internally validated a nomogram for six-month mortality, external validation in larger, multicenter groups is needed. Our model was based on traditional regression methods; future studies could explore advanced approaches, such as neural networks, to capture complex variable interactions and improve predictive performance, as recent evidence has demonstrated their superiority over conventional models[54]. Prospective, multicenter research is warranted to refine risk stratification and optimize management strategies for PPH.

CONCLUSION

This study systematically identified independent risk factors for six-month mortality in patients with PPH undergoing endovascular treatment, including advanced age, coagulation dysfunction, prolonged operative duration, shorter in-hospital days, intra-abdominal infection, common hepatic artery bleeding, and clinical failure. We developed and validated a nomogram for rapid risk stratification to assist clinicians in perioperative decision-making. This tool can guide clinicians in implementing intensified interventions for high-risk groups - particularly elderly patients or those with intra-abdominal infection - including correction of coagulation dysfunction, stringent infection control, and optimization of interventional strategies. Shifting the treatment focus from mere technical hemostasis to organ protection may help improve clinical outcomes in this high-risk population.

ACKNOWLEDGEMENTS

We sincerely thank the medical and nursing teams of the Department of Interventional Radiology and Department of Hepatobiliary Surgery at Beijing Chao-Yang Hospital for their invaluable support during data collection and clinical management. We also appreciate the assistance provided by the hospital’s medical records and statistics departments.