Predicting pancreatic fistula post-splenectomy in cirrhosis with splenomegaly: Risk factors and nomogram validation

Abstract

BACKGROUND

Clinically relevant postoperative pancreatic fistula (CR-POPF) is a critical complication of splenectomy, particularly in patients with cirrhosis, splenomegaly, and hypersplenism. However, there is a lack of comprehensive risk assessment models.

AIM

To identify independent risk factors for CR-POPF after splenectomy in this cohort and to develop a clinically useful predictive nomogram.

METHODS

This retrospective study included 186 patients who underwent splenectomy for cirrhotic splenomegaly with hypersplenism at our institution, between January 2019 and December 2024. PF was diagnosed and graded according to the 2016 International Study Group on Pancreatic Fistulas criteria. Univariate and multivariate least absolute shrinkage and selection operator-penalized logistic regression analyses were performed to identify independent risk factors. A nomogram was constructed based on these factors, and its performance was evaluated using the area under the receiver operating characteristic curve, calibration curves, and decision curve analysis.

RESULTS

Of the 186 patients, 21 developed biochemical leak, and 6 developed CR-POPF, including 4 grade B and 2 grade C. Multivariate least absolute shrinkage and selection operator-penalized logistic analysis revealed that body mass index (BMI) [odds ratio = 3.578, 95% confidence interval (CI): 1.156-11.080; P = 0.027] and splenic thickness (odds ratio = 1.250; 95%CI: 1.014-1.542; P = 0.037) were independent risk factors for CR-POPF. The nomogram integrating these factors demonstrated excellent discriminative ability (area under the receiver operating characteristic curve = 0.816, 95%CI: 0.500-0.993), good calibration (Hosmer-Lemeshow statistic = 0.027, degrees of freedom = 3, P = 0.999 > 0.05), and favorable clinical utility (positive net benefit across a threshold probability range of approximately 0% to 99% in the decision curve analysis). PF grade analysis showed significant differences in BMI and pancreatic texture (P < 0.05).

CONCLUSION

This study identified BMI and splenic thickness as independent risk factors for CR-POPF after splenectomy in patients with cirrhosis with splenomegaly. The developed nomogram provides a reliable tool for individualized risk prediction, facilitating targeted perioperative management to reduce CR-POPF incidence and improve patient outcomes.

Key Words: Pancreatic fistula; Splenectomy; Splenomegaly; Risk factors; Nomogram

Core Tip: This retrospective study included 186 cirrhotic patients with splenomegaly who underwent splenectomy. Through multivariate least absolute shrinkage and selection operator-penalized logistic regression, body mass index and splenic thickness were confirmed as independent risk factors for clinically relevant postoperative pancreatic fistula. The nomogram built on these factors achieved an area under the receiver operating characteristic curve of 0.816, providing reliable individualized perioperative risk prediction to cut clinically relevant postoperative pancreatic fistula occurrence and boost patient recovery.

INTRODUCTION

Pancreatic fistula (PF) is a significant and complex complication with diverse etiologies. Pancreatic surgeries such as pancreaticoduodenectomy and distal pancreatectomy carry a high risk of PF, with reported incidence rates ranging from 3% to 45%, as shown in multiple studies[1]. This wide variation can be ascribed to the differences in surgical techniques, patient populations, and diagnostic criteria. Traumatic injuries, particularly pancreatic contusions, lacerations, or crush injuries, can also lead to PF because of the often underestimated extent of tissue damage and difficulties in managing pancreatic duct injuries[2]. In addition, acute pancreatitis, especially severe cases, is frequently associated with PF, where local ischemia and the digestive action of pancreatic enzymes can result in pancreatic tissue and duct necrosis, and approximately 20% of severe acute pancreatitis survivors develop PF[3]. The overall mortality rate related to PF is approximately 1%, mainly due to subsequent severe infections and multi-organ failure.

PF has emerged as a common complication of splenectomy, primarily due to iatrogenic injury to the pancreatic tail during surgical manipulation. As most cases of PF induced by splenectomy present with mild symptoms and achieve a favorable prognosis after adequate drainage and other supportive interventions, this complication has long been insufficiently emphasized in clinical practice. Notably, the pathogenesis of PF differs substantially between pancreatic surgery and splenectomies. In pancreatic surgery, PF typically arises from pancreatic-intestinal anastomotic leaks or defects on the pancreatic surface. In contrast, splenectomy-related PF is predominantly caused by iatrogenic injury to the pancreatic tail parenchyma during splenic mobilization or pedicle dissection. This mechanistic distinction indicates that the risk factors for pancreatic surgery-associated PF are irrelevant to splenectomy cohorts. Given that this study focused on patients with cirrhotic splenomegaly, it emphasizes cirrhosis-related factors, which are necessary for addressing the unmet clinical needs of this specific patient population. Massive splenomegaly, especially when the spleen is significantly enlarged, increases the complexity of the surgical procedure and the likelihood of accidental pancreatic injury. The main manifestations of PF after splenectomy are consistent with those in other scenarios, including persistent abdominal pain, fever, and elevated amylase levels in peritoneal drainage fluid. If not treated or managed properly, it can progress to serious complications, such as intra-abdominal abscess formation, sepsis, and even life-threatening hemorrhage, which significantly prolong hospital stays and escalate healthcare costs[4]. The existing literature has identified some risk factors for PF after splenectomy, such as the type of splenectomy and use of energy-based devices[5]. However, these studies often have limitations in terms of the sample size, patient heterogeneity, and inconsistent definitions of risk factors. Currently, there is a lack of comprehensive and unified nomogram-based analyses of risk factors for PF.

The effective management of PF depends on accurate monitoring and classification. According to the diagnostic criteria revised by the International Study Group on Pancreatic Fistulas (ISGPS) in 2016, the amylase level in the drainage fluid three days after surgery exceeded three times the normal serum value, which is in line with the diagnosis of PF[6,7]. The ISGPS classifies PF into three grades: Grade A (biochemical leak) is characterized by an increase in amylase without clinical symptoms; grade B is characterized by continuous drainage for more than 3 weeks accompanied by signs of infection; and grade C is characterized by organ failure or even requires reoperation. The incidence of these severe adverse outcomes further increases with the presence of grade B or C fistulas, as defined by the ISGPS. Therefore, investigators have focused on improving predictive models to assess the risk of developing clinically relevant postoperative pancreatic fistulas (CR-POPF). The harmful effects of CR-POPF can lead to severe complications in various systems. The leakage of pancreatic enzymes into the abdominal cavity can cause corrosive bleeding. The literature shows that approximately 12.7% of patients with grade B/C fistulas will have secondary erosion of abdominal blood vessels, and rupture of the splenic artery or short gastric artery can lead to fatal massive hemorrhage[8]. Pancreatic juice spillage can also disrupt the peritoneal barrier, and approximately 34.5% of cases progress to diffuse peritonitis, which in turn can induce sepsis and multiple organ failure, with a mortality rate of 8.3% to 15.6%. Among patients with CR-POPF complicated by liver cirrhosis and portal hypertension, 28.6% have an increased risk of bleeding due to coagulation disorders, and the incidence of infectious complications is 4.7 times higher than that in patients without CR-POPF[8]. These data underscore the clinical threat of CR-POPF, which not only directly impairs organ function but also triggers systemic inflammatory response syndrome and consumptive coagulopathy, thereby initiating a life-threatening pathological cascade[9].

Given the significant morbidity and potential mortality associated with CR-POPF, understanding its risk factors is crucial[10]. Although several risk factors have been identified in previous studies, comprehensive and integrated analyses that synthesize all these factors are still lacking. In addition, no nomogram prediction model has been specifically developed for CR-POPF after splenectomy. In this study, we conducted a thorough analysis of CR-POPF risk factors using a retrospective review of 186 cases collected over a 6-year period. We aimed to develop a more accurate predictive nomogram model by systematically evaluating both the patient- and surgery-related variables. This nomogram can help surgeons identify high-risk patients pre- or intra-operatively, enabling more proactive and personalized management strategies to prevent or better manage CR-POPF after splenectomy, thus contributing to clinical prevention efforts.

MATERIALS AND METHODS

General information

This retrospective study consecutively enrolled patients who underwent splenectomy for cirrhotic splenomegaly with hypersplenism at the First Affiliated Hospital of the Anhui University of Chinese Medicine between January 2019 and December 2024. In total, 197 patients were initially identified during the study period. After applying the strict inclusion and exclusion criteria, 186 patients were finally included in the analysis (Figure 1). Clinical data, including demographic characteristics, laboratory results, surgical details, postoperative outcomes, and complications, were collected from the electronic medical records and surgical databases. All the data were reviewed and verified by two independent researchers to ensure accuracy.

Figure 1 The patient selection flow diagram. CR-POPF: Clinically relevant postoperative pancreatic fistula; LASSO: Least absolute shrinkage and selection operator; ROC: Receiver operating characteristic; DCA: Decision curve analysis.

Inclusion and exclusion criteria

Inclusion criteria were defined as follows: (1) Definite cirrhosis diagnosis per the European Association for the Study of the Liver Clinical Practice Guidelines[11], confirmed by: Clinical evidence of portal hypertension (splenomegaly, jaundice, ascites, varices, or encephalopathy); biochemical markers: Albumin < 35 g/L, bilirubin > 1.2 mg/dL; imaging: Liver stiffness ≥ 17.5 kPa by FibroTouch or nodular hepatic morphology on contrast-enhanced computed tomography (CT)/magnetic resonance imaging; (2) Presence of splenomegaly with hypersplenism: Ultrasound-proven splenic length > 13 cm or splenic thickness > 4 cm; cytopenia (platelets < 100 × 10⁹/L, white blood cells < 3 × 10⁹/L); clinical signs of portal hypertension, such as splenomegaly, jaundice, ascites, variceal bleeding history; (3) Underwent splenectomy; (4) Postoperative monitoring of drainage fluid amylase levels on days 2-3 after surgery; and (5) All patients, particularly adolescents aged ≤ 18 years, underwent hematology consultation and bone marrow aspiration to exclude hematological causes of splenomegaly before enrollment.

Exclusion criteria included: (1) Incomplete postoperative monitoring of drainage fluid amylase; (2) Incomplete clinical data (missing more than two key parameters); (3) Concurrent pancreatic surgery during splenectomy; (4) Severe comorbidities that affect postoperative recovery; (5) Refusal to provide informed consent; and (6) Early hospital discharge.

Ethics statement

Prior to commencing this study, the objectives and specific details were comprehensively explained to the participants, and informed consent was obtained from each participant or their respective representatives. This study was conducted in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki, revised in Fortaleza, Brazil, in October 2013). The study protocol was approved by the Institutional Review Board of the First Affiliated Hospital of Anhui University of Chinese Medicine, No. 2024AH-30. Patient data were de-identified and stored in a password-protected database compliant with the Health Insurance Portability and Accountability Act using Research Electronic Data Capture for secure data management (the raw data were shown in the Supplementary material).

Surgical procedures

All surgeries were performed by a single surgical team consisting of six experienced surgeons to ensure consistency in the operative techniques. Preoperative evaluations within 1 week before surgery included: (1) General lifestyle factors and medical conditions include smoking, alcohol consumption, hypertension, coronary artery disease, diabetes, dyslipidemia, body mass index (BMI), and American Society of Anesthesiologists Physical Status Classification (ASA classification); (2) Liver function assessment using the Child-Pugh classification system (class A, B, or C); (3) Routine hematological indices, including complete blood count (white blood cells, red blood cells, hemoglobin, and platelets), liver function tests (alanine aminotransferase, aspartate aminotransferase, total bilirubin, and albumin, and coagulation profiles [prothrombin time (PT), international normalized ratio, activated partial thromboplastin time, and D-dimer]; (4) Abdominal ultrasonography to measure splenic length, splenic thickness, portal vein diameter, and splenic vein diameter; and (5) Pancreatic CT value measured by abdominal CT.

Surgical procedures included splenectomy alone or in combination with periesophagogastric devascularisation. The postoperative diagnosis of PF was based on the 2016 ISGPS criteria: Grade A (biochemical leak), drainage fluid amylase > 3 times the upper limit of normal serum amylase without clinical symptoms and requiring no specific intervention. Grade B: Drainage fluid amylase > 3 times the upper limit of normal serum amylase, accompanied by prolonged drainage (> 3 weeks), clinical signs of infection (fever, leukocytosis), or need for imaging-guided intervention (percutaneous drainage). Grade C: Severe PF requiring reoperation, organ failure, or life-threatening complications (massive hemorrhage or sepsis). CR-POPF was defined as grade B or C.

Surgical variables were recorded as follows

First-order splenic pedicle vessel division: The primary splenic pedicle was dissected, ligated, and sutured after dissociation (Figure 2).

Figure 2 Different splenic pedicle division methods.

Second-order splenic pedicle vessel division: The serosa tissue in front of the secondary splenic pedicle space was separated and dissected after the bifurcation of the splenic pedicle was clinically determined; the secondary splenic pedicle or tertiary splenic pedicle space could be clearly visualized. Every secondary splenic pedicle vessel or tertiary splenic pedicle vessel was dissected and ligated under direct vision (Figure 2).

Stapler division: This indicates whether a surgical stapler is used to cut and seal the splenic pedicle vessels during the procedure (Figure 2).

Pancreatic texture: The pancreatic texture was evaluated intra-operatively by the surgeon via palpation, categorized as “soft”, “firm”, or “hard”. The hard texture was attributed to underlying chronic pancreatitis or pancreatic fibrosis. A soft texture is a normal physiological texture or a manifestation of pancreatic steatosis, and “firm” refers to a moderate consistency between soft and hard (Figure 2).

Blood loss: Calculated as the sum of the blood collected by the suction devices and the estimated amount of blood absorbed by the surgical gauze (1 g gauze = 1 mL blood) (Figure 2).

Statistical analysis

All statistical analyses were performed using the R software (version 4.5.1, R Foundation for Statistical Computing, Vienna, Austria).

Missing data: Missing values with < 5% missingness were handled using multiple imputations with ten imputed datasets via the mice package. Cases with > 5% of missing data were excluded from the analysis.

Outlier detection: Outliers in continuous variables were identified using the interquartile range (IQR) method (values > 1.5 × IQR above the 75^th percentile or below the 25^th percentile) and verified by visual inspection of boxplots. Outliers were retained after the confirmation of data accuracy.

Normality assessment: The normality of continuous variables was tested using the Shapiro-Wilk test. Normally distributed data are presented as mean ± SD, and non-normally distributed data are presented as median IQR.

Group comparisons: Categorical variables were compared using the χ² test (via χ² test in the base package) or Fisher’s exact test (via Fisher’s test in the base package), as appropriate. Continuous variables were compared using the independent samples t-test (via the t-test in the base package) for normally distributed data or the Mann-Whitney U test (via the Wilcoxon test in the base package) for non-normal data.

Risk factor analysis: Univariate and multivariate least absolute shrinkage and selection operator (LASSO)-penalized logistic regression analyses were performed using the glm function (base statistics package) and the glmnet function (glmnet package). Variables with a P-value < 0.05 in univariate analysis were included in the multivariate model. Additionally, the Bootstrap internal validation method was used to reduce optimistic bias of the reported area under the receiver operating characteristic (ROC) curve (AUC). Correlation analysis heatmaps were generated using R software (pheatmap package) to analyze key variables from the univariate analysis, and variables to be included in the multivariate regression were screened based on correlation heatmaps and collinearity.

Nomogram construction: A nomogram was developed based on the multivariate LASSO-penalized logistic regression model using the nomogram function in the rms package. The nomogram was evaluated using the AUC, calibration curves, and decision curve analysis (DCA) using pROC, dplyr, and ggplot2 packages. Model performance was evaluated using the concordance index and calibration curves using the rms package. Statistical significance was defined as a two-tailed P-value of < 0.05.

RESULTS

General information

Among the 186 patients included in the study, the mean age was 35.48 ± 14.25 years, with a range from 11 years to 78 years. The study included 106 male patients and 80 female patients. In terms of liver function, 145 patients were classified as Child-Pugh class A, 37 as class B, and four as class C. The mean splenic length was 188.08 ± 26.93 mm, the mean splenic thickness was 63.95 ± 11.21 mm, the mean portal vein diameter was 13.56 ± 2.17 mm, and the mean splenic vein diameter was 13.40 ± 2.53 mm.

Comparison between the CR-POPF group and the non-CR-POPF group

This study identified 6 patients with CR-POPF and 180 without CR-POPF. In terms of demographic characteristics, there were no significant differences in age, sex, alcohol consumption, smoking, hypertension, coronary artery disease, dyslipidemia, diabetes, ASA classification, and Child-Pugh distribution between the two groups (P > 0.05). The CR-POPF group exhibited a significantly higher BMI than the non-CR-POPF group [26.9 (24.33, 28.53) kg/m² vs 22.00 (20.30, 23.20) kg/m², P < 0.05]. PT in the CR-POPF group was significantly lower than that in the non-CR-POPF group [11.75 (10.93, 12.20) seconds vs 13.50 (12.50, 14.88) seconds, P < 0.05]. Statistically significant differences were observed in splenic vein diameter [15 (15, 16.25) mm vs 13.00 (11.00, 15.00) mm] and splenic thickness (75.67 ± 8.55 mm vs 63.56 ± 11.10 mm) between the two groups (P < 0.05). The use of stapler divisions differed between the two groups, with a higher percentage of staplers used in the CR-POPF group (P < 0.05). The pancreatic texture also differed between the two groups (P < 0.05). In addition, the hospital stay was significantly longer in the CR-POPF group [14 (11.75, 33.25) days] than in the non-CR-POPF group [14.00 (12.00, 17.00) days, P < 0.05] (Table 1).

Table 1 Comparison of demographic and clinical characteristics between patients with and without pancreatic fistula after splenectomy, mean ± SD.

Variables	CR-POPF group (n = 6)	Non-CR-POPF group (n = 180)	P value
Age (years)	33.17 ± 21.78	35.56 ± 14.01	0.687
Gender (male/female)			0.701
Male	4	102
Female	2	78
Alcohol consumption			0.401
Yes	1	14
No	5	166
Smoking			0.185
Yes	2	23
No	4	157
Hypertension			0.552
Yes	1	22
No	5	158
Coronary artery disease			0.310
Yes	1	10
No	5	170
Dyslipidemia			0.518
Yes	1	20
No	5	160
Diabetes			0.249
Yes	2	28
No	4	152
BMI (kg/m2)	26.90 (24.33, 28.53)	22.00 (20.30, 23.20)	0
ASA classification			0.150
I	3	131
II	2	39
III	1	10
WBC (× 109/L)	2.29 (1.27, 3.64)	2.35 (1.72, 3.02)	0.884
RBC (× 1012/L)	3.67 ± 0.65	3.87 ± 0.59	0.470
HGB (g/L)	102.50 ± 10.62	109.31 ± 18.80	0.379
PLT (× 109/L)	43.50 (33.00, 65.50)	41.50 (32.00, 53.75)	0.522
ALT (U/L)	26.40 (17.75, 45.50)	25.05 (19, 33.95)	0.920
AST (U/L)	26.50 (17.50, 42.50)	28.50 (21, 36.75)	0.934
TB (μmol/L)	14.63 (8.88, 17.85)	17.90 (12.88, 27.23)	0.123
ALB (g/L)	35.95 (30.25, 40.43)	36.20 (32.90, 39.60)	0.710
PT (seconds)	11.75 (10.93, 12.20)	13.50 (12.50, 14.88)	0.020
INR	1.09 (1.02, 1.18)	1.23 (1.11, 1.36)	0.067
APTT (seconds)	30.45 (27.73, 32.63)	34.75 (31.70, 38.28)	0.078
D-Dimer (mg/L FEU)	0.31 (0.19, 0.57)	0.24 (0.19, 0.59)	0.611
Child-Pugh classification			0.910
A	5	140
B	1	36
C	0	4
Ascite			0.206
No ascites	3	132
Mild ascites	2	24
Moderate to severe ascites	1	24
Splenic length (mm)	203.17 ± 7.41	187.58 ± 27.20	0.164
Splenic thickness (mm)	75.67 ± 8.55	63.56 ± 11.10	0.009
Portal vein diameter (mm)	14.00 (13.00, 15.50)	13.00 (12.00, 15.00)	0.380
Splenic vein diameter (mm)	15.00 (15.00, 16.25)	13.00 (11.00, 15.00)	0.038
Main pancreatic duct diameter (mm)	2.25 (1.85, 4.28)	2.70 (2.20, 3.10)	0.959
Pancreatic CT value (HU)	34.00 (30.25, 41.25)	35.00 (29.25, 39.00)	0.709
Surgical method			0.917
Splenectomy	5	147
Devascularization	1	33
Splenic pedicle vessel division method			0.092
First-order splenic pedicle vessel division	5	79
Second-order splenic pedicle vessel division	1	101
Stapler division			0.018
Yes	5	58
No	1	122
Pancreatic texture			0.048
Soft	4	87
Firm	1	97
Hard	1	2
Drainage fluid amylase	20762.50 (10877.25, 42335.75)	216.00 (100.00, 441.25)	0
Hospital stay (days)	14.00 (11.75, 33.25)	14.00 (12.00, 17.00)	0.041
Blood loss (mL)	395.00 (290.00, 450.00)	230.00 (200.00, 350.00)	0.144

CR-POPF: Clinically relevant postoperative pancreatic fistula; BMI: Body mass index; ASA: Acetylsalicylic acid; WBC: White blood cell; RBC: Red blood cell; HGB: Hemoglobin; PCT: Plateletcrit; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TB: Total bilirubin; ALB: Albumin; PT: Prothrombin time; APTT: Activated partial thromboplastin time; INR: International normalized ratio; FEU: Fibrinogen equivalent unit; CT: Computed tomography; HU: Hounsfield unit.

Of the 21 patients were classified as having biochemical leak, four as grade B, and two as grade C, according to the 2016 ISGPS criteria. Comparisons of clinical characteristics among the three grades revealed no significant differences in age, sex, alcohol consumption, smoking, hypertension, coronary artery disease, dyslipidemia, diabetes, ASA classification, or most laboratory indices, including white blood cells, hemoglobin, liver function parameters (alanine aminotransferase, aspartate aminotransferase, total bilirubin, and albumin), and coagulation indices, such as PT and international normalized ratio. No significant differences were observed in splenic size (length, thickness), portal vein diameter, splenic vein diameter, main pancreatic duct diameter, pancreatic CT value, Child-Pugh classification, ascites, surgical method, splenic pedicle vessel division method, stapler usage, hospital stay, or blood loss among the three grades (all P > 0.05). Notably, pancreatic texture showed a significant difference among the groups (P = 0.008). BMI also differed significantly, with higher BMI in grades B [26.9 (24.95, 28.78) kg/m²] and C [26.05 (23.8, 26.05) kg/m²] compared to biochemical leak [22.6 (21.15, 23.20) kg/m², P < 0.001] (Table 2).

Table 2 Comparison of general data among different pancreatic fistula grades after surgery.

Variables	Biochemical leak (n = 21)	Grade B (n = 4)	Grade C (n = 2)	P value
Age (years, mean ± SD)	34.19 ± 12.76	29.25 ± 20.85	41 ± 29.70	0.664
Gender (male/female)				0.648
Male	9	3	1
Female	12	1	1
Alcohol consumption				0.545
Yes	2	1	0
No	19	3	2
Smoking				0.813
Yes	9	1	1
No	12	3	1
Hypertension				0.338
Yes	7	0	1
No	14	4	1
Coronary artery disease				0.545
Yes	2	1	0
No	19	3	2
Dyslipidemia				0.752
Yes	4	1	0
No	17	3	2
Diabetes				0.542
Yes	4	1	1
No	17	4	1
BMI (kg/m2)	22.60 (21.15, 23.20)	26.90 (24.95, 28.78)	26.05 (23.80, 26.05)	0
ASA				0.677
I	14	2	1
II	5	1	1
III	2	1	0
WBC (× 109/L)	2.67 (2.05, 3.94)	1.67 (1.24, 2.41)	4.04 (3.24, 4.04)	0.190
RBC (× 1012/L)	3.89 ± 0.59	3.33 ± 0.15	4.42 ± 0.70	0.082
HGB (g/L)	112.05 ± 19.60	98.25 ± 5.91	111 ± 15.56	0.397
PLT (× 109/L)	51.00 (32.00, 55.50)	49.50 (34.00, 72.50)	41.50 (33.00, 41.50)	0.760
ALT (U/L)	24.00 (16.80, 38.00)	26.40 (19.75, 53.45)	28.50 (17.00, 28.50)	0.962
AST (U/L)	28.00 (21.00, 35.50)	31.00 (19.25, 49.50)	23.00 (16.00, 23.00)	0.570
TB (μmol/L)	16.20 (13.05, 31.17)	11.60 (7.03, 15.25)	18.70 (17.00, 18.70)	0.392
ALB (g/L)	38.90 (33.85, 40.75)	31.95 (29.15, 37.68)	41.45 (38.80, 41.45)	0.259
PT (seconds)	13.30 (12.00, 13.75)	11.80 (10.48, 12.60)	11.55 (11.20, 11.55)	0.155
INR	1.16 (1.05, 1.26)	1.14 (0.98, 1.21)	1.06 (1.04, 1.06)	0.423
APTT (seconds)	35.20 (31.80, 37.80)	30.05 (26.78, 33.40)	30.60 (29.00, 30.60)	0.061
D-Dimer (mg/L FEU)	0.23 (0.19, 0.56)	0.24 (0.19, 0.79)	0.38 (0.32, 0.38)	0.872
Child-Pugh classification				0.752
A	17	3	2
B	4	1	0
C		0	0
Ascite				0.227
No ascites	17	2	1
Mild ascites	2	1	1
Moderate to severe ascites	2	1	0
Splenic length (mm)	188.62 ± 19.74	207 ± 5.72	195.5 ± 0.71	0.192
Splenic thickness (mm)	67.71 ± 11.89	76 ± 2.94	75 ± 18.38	0.344
Portal vein diameter (mm)	15.00 (13.00, 16.00)	14.00 (13.25, 14.75)	15.00 (13.00, 15.00)	0.773
Splenic vein diameter (mm)	15.00 (14.00, 16.00)	15.00 (15.00, 16.50)	15.50 (15.00, 15.50)	0.828
Main pancreatic duct diameter (mm)	2.80 (2.05, 3.05)	2.25 (1.93, 3.70)	3.25 (1.70, 3.25)	0.742
Pancreatic CT value (HU)	32.00 (27.50, 35.00)	32.50 (28.75, 37.75)	41.00 (34.00, 41.00)	0.093
Surgical method				0.162
Splenectomy	20	4	1
Devascularization	1	0	1
Splenic pedicle vessel division method				0.733
First-order splenic pedicle vessel division	16	3	2
Second-order splenic pedicle vessel division	5	1	0
Stapler division				0.497
Yes	12	3	2
No	9	1	0
Pancreatic texture				0.008
Soft	16	3	0
Firm	5	1	0
Hard	0	0	2
Drainage fluid amylase	1485.00 (1294.50, 1961.00)	12197.50 (10377.75, 24509.75)	54671.50 (30000.00, 54671.50)	0
Hospital stay (days)	14.00 (12.00, 16.50)	22.50 (11.25, 33.75)	14.00 (14.00, 14.00)	0.079
Blood loss (mL)	340.00 (300.00, 372.50)	380.00 (230.00, 447.50)	400.00 (350.00, 400.00)	0.460

BMI: Body mass index; ASA: Acetylsalicylic acid; WBC: White blood cell; RBC: Red blood cell; HGB: Hemoglobin; PCT: Plateletcrit; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TB: Total bilirubin; ALB: Albumin; PT: Prothrombin time; APTT: Activated partial thromboplastin time; INR: International normalized ratio; FEU: Fibrinogen equivalent unit; CT: Computed tomography; HU: Hounsfield unit.

LASSO-penalized logistic regression analysis results

A univariate logistic regression analysis was performed to identify several potential risk factors for PF. Variables including BMI, PT, splenic thickness, splenic vein diameter, stapler division, and pancreatic texture showed significant differences in univariate analysis (P < 0.05). The correlation heatmap revealed significant collinearity among splenic thickness, splenic vein diameter, and pancreatic texture (Figure 3). Notably, splenic thickness showed a more significant difference (P = 0.009) between the two groups than the splenic vein diameter (P = 0.038) and pancreatic texture (P = 0.048). Furthermore, the total number of CR-POPF events (grades B/C) reported in the manuscript was substantially lower than the number of predictors that were initially planned for the multivariate model. To mitigate predictor redundancy, splenic thickness was used to replace splenic vein diameter and pancreatic texture and was finally included in the multivariate LASSO-penalized logistic regression. The multivariate LASSO-penalized logistic regression model identified that BMI [odds ratio = 3.578, 95% confidence interval (CI): 1.156-11.080; P = 0.027] and splenic thickness (odds ratio = 1.250; 95%CI: 1.014-1.542; P = 0.037) were independent risk factors for CR-POPF (Table 3).

Figure 3 The correlation heatmap of influencing factors. BMI: Body mass index; PT: Prothrombin time. ^aP < 0.05, ^bP < 0.01, ^cP < 0.001.

Table 3 Multivariate least absolute shrinkage and selection operator-penalized logistic regression analysis of risk factors for pancreatic fistula after splenectomy.

Variables	B	SE	Wald	df	Significance	Exp (B)	Lower 95%CI	Upper 95%CI
BMI	1.275	0.577	4.887	1	0.027	3.578	1.156	11.080
PT	-2.531	1.482	2.916	1	0.088	0.080	0.004	1.453
Splenic thickness	0.223	0.107	4.352	1	0.037	1.250	1.014	1.542
Stapler division	1.155	1.493	0.549	1	0.439	3.175	0.170	59.227

B: Unstandardized beta coefficient; SE: Standard error; CI: Confidence interval; Exp: Exponentiated beta coefficient; BMI: Body mass index; PT: Prothrombin time.

Nomogram development and validation

A nomogram was constructed to visually quantify the risk of CR-POPF based on LASSO-penalized logistic regression, with the optimal LASSO penalty coefficient determined as 0.0086 via 10-fold cross-validation, integrating the four independent risk factors identified: Stapler division, splenic thickness, PT, and BMI (Figure 4). The concordance index of the nomogram was 0.816 with a 95%CI: 0.500-0.993, further validating its robust predictive performance. To evaluate the discriminative efficacy of the nomogram in distinguishing between patients with and without CR-POPF, we plotted an ROC curve. Given that the nomogram was constructed based on LASSO-penalized logistic regression, 1000 bootstrap internal validations were performed to calibrate the AUC and reduce optimism bias. The AUC was 0.816 (consistent with the model performance), indicating excellent discriminative power, with a corresponding maximum Youden’s index of 0.903 (Figure 5).

Figure 4 A nomogram of pancreatic fistula based on the multivariate least absolute shrinkage and selection operator-penalized logistic regression model. ¹An independent risk factor for clinically relevant postoperative pancreatic fistula by multivariate least absolute shrinkage and selection operator-penalized logistic regression (odds ratio = 1.250; 95% confidence interval: 1.014-1.542; P = 0.037). ²An independent risk factor for clinically relevant postoperative pancreatic fistula by multivariate least absolute shrinkage and selection operator-penalized logistic regression (odds ratio = 3.578, 95% confidence interval: 1.156-11.080, P = 0.027). PT: Prothrombin time; BMI: Body mass index.

Figure 5 Receiver operating characteristic curve for the predictive nomogram of pancreatic fistula after splenectomy. ROC: Receiver operating characteristic; LASSO: Least absolute shrinkage and selection operator; AUC: Area under the receiver operating characteristic curve; CI: Confidence interval.

The calibration of the nomogram was evaluated using calibration curves with 1000 bootstrap resamples to correct for potential overfitting. The calibration curve showed good agreement between the predicted probabilities of CR-POPF and actual occurrence probabilities. The Hosmer-Lemeshow test indicated that the model had excellent calibration (Hosmer-Lemeshow statistic = 0.027, degrees of freedom = 3, P = 0.999 > 0.05), demonstrating no significant discrepancy between the predicted probabilities and the actual outcomes (Figure 6). DCA was performed to assess the clinical utility of the nomogram. The DCA curve showed that when the threshold probability of CR-POPF ranged from approximately 0% to 99%, the nomogram provided a positive net benefit that was superior to both the “treat all” and “treat none” strategies (Figure 7). This indicates that the nomogram can effectively assist clinicians in identifying high-risk patients who may benefit from targeted preventive interventions, thereby optimizing clinical decision-making.

Figure 6 Calibration curve of the nomogram for predicting pancreatic fistula. CR-POPF: Clinically relevant postoperative pancreatic fistula.

Figure 7 Decision curve analysis of the nomogram for pancreatic fistula prediction. CR-POPF: Clinically relevant postoperative pancreatic fistula; LASSO: Least absolute shrinkage and selection operator.

DISCUSSION

CR-POPF (grade B/C) remains a significant complication following splenectomy, contributing to prolonged hospital stays, increased healthcare costs, and, in severe cases, life-threatening outcomes such as sepsis and multi-organ failure[12]. Our study aimed to identify independent risk factors for CR-POPF after splenectomy in patients with cirrhotic splenomegaly and hypersplenism, and to develop a clinically useful prediction tool. These findings provide novel insights into the pathogenesis of CR-POPF and offer practical guidance for its risk stratification and prevention. Univariate analysis indicated that BMI, PT, splenic thickness, splenic vein diameter, stapler division, and pancreatic texture were the risk factors for CR-POPF. Multivariate LASSO-penalized logistic regression analysis identified two independent risk factors for CR-POPF: BMI and splenic thickness. These factors collectively reflect the interplay between the anatomical, surgical, and tissue-specific characteristics that influence CR-POPF development. Splenic vein diameter was found to be a risk factor, with larger diameters associated with a higher CR-POPF risk. This may be attributed to the anatomical proximity of the splenic vein to the pancreatic tail. An enlarged splenic vein can complicate surgical dissection, increasing the likelihood of accidental pancreatic parenchymal or duct injury during splenic pedicle manipulation. Previous studies have highlighted that portal hypertension-related venous dilation alters the local anatomical structure, making pancreatic tissue more vulnerable to iatrogenic damage, which aligns with our findings[13,14]. Stapler use for splenic pedicle division was associated with a significantly higher risk of CR-POPF, indicating that stapler use may increase the risk of CR-POPF. This increased risk can be attributed to the potential for direct injury to the pancreatic tail during the stapler placement. Although staplers are designed to deliver precise, rapid, and uniform tissue transections, their use in the area around the splenic pedicle adjacent to the pancreatic tail carries the inherent risk of accidentally enclosing or compressing pancreatic tail tissue within the stapling line. Direct mechanical trauma to the pancreatic parenchyma or ductal structures may disrupt their integrity, thereby increasing the likelihood of leakage of pancreatic juice. In contrast, non-use of a stapler may pose a lower risk of direct mechanical injury to the pancreatic tail owing to its more gradual and controlled approach to vessel division[15]. This finding underscores the importance of cautious stapler use in regions adjacent to the pancreas, as devices designed for precision can cause iatrogenic damage when operating in anatomically complex areas[16]. Pancreatic texture has been identified as a risk factor for CR-POPF. A hard texture, often linked to underlying chronic pancreatitis or pancreatic fibrosis, renders pancreatic tissue less resilient to surgical trauma. A fibrotic pancreas may have altered ductal integrity, and even minor injury during surgery can disrupt ductal continuity, increasing the risk of pancreatic juice leakage. Conversely, a soft texture, often associated with pancreatic edema, acute inflammation, or fatty infiltration, may indicate a fragile parenchyma with reduced structural stability. Such tissues are prone to tearing or accidental damage during surgical manipulation because their delicate consistency makes precise dissection and hemostasis challenging. Even subtle trauma can compromise the integrity of the pancreatic ducts, leading to the leakage of pancreatic secretions. This finding underscores the importance of intra-operative assessment of pancreatic texture as a quick and actionable indicator of CR-POPF risk[17,18].

Furthermore, univariate analysis revealed several other variables associated with CR-POPF development, providing supplementary insights into the complex interplay between patient and surgical factors influencing CR-POPF risk. The PT [11.75 (10.93, 12.20) seconds vs 13.50 (12.50, 14.88) seconds, P = 0.020] was lower in the CR-POPF group. These findings may reflect altered coagulation dynamics in patients with cirrhosis and portal hypertension, potentially affecting pancreatic tissue perfusion and healing. Splenic thickness was greater in the CR-POPF group (75.67 ± 8.55 mm vs 63.56 ± 11.10 mm, P = 0.009), indicating that more severe splenomegaly might distort the peripancreatic anatomy, increasing the risk of iatrogenic pancreatic injury during dissection. BMI was significantly higher in the CR-POPF group [26.9 (24.33, 28.53) kg/m² vs 22.00 (20.30, 23.20) kg/m², P = 0.000], which could be attributed to increased technical difficulty in surgical exposure due to adipose tissue, potentially elevating trauma risk to adjacent pancreatic structures. The association of these variables with CR-POPF in the univariate and multivariate analyses highlights the multifaceted nature of CR-POPF pathogenesis, emphasizing the need for a comprehensive perioperative assessment of both patient characteristics and surgical variables.

The nomogram developed from these four risk factors, including BMI, PT, splenic thickness, and stapler division, provides a practical tool for individualized CR-POPF risk prediction. Its excellent discriminative ability (AUC = 0.816, 95%CI: 0.500-0.993), high calibration accuracy (Hosmer-Lemeshow statistic = 0.027, degrees of freedom = 3, P = 0.999 > 0.05), and positive net benefit across a broad range of threshold probabilities via DCA validated its clinical value. By integrating easily measurable variables (BMI and PT) and intra-operative observations (splenic thickness and stapler division), the nomogram allows surgeons to assess CR-POPF risk preoperatively or intra-operatively, enabling timely preventive measures[19]. High-risk patients could benefit from enhanced intraoperative care, including meticulous dissection near the pancreatic tail and cautious use of staplers, or postoperative monitoring with more frequent amylase testing in drainage fluid to detect early CR-POPF and initiate interventions, such as administration of pancreatic secretion inhibitors or targeted drainage.

The nomogram developed based on multivariate LASSO-penalized logistic regression results provides a valuable tool for predicting the risk of CR-POPF. Similar nomogram-based prediction models have been successfully used in other surgical fields to identify high-risk patients early[20]. Using our nomogram, clinicians could estimate the individual risk of CR-POPF in each patient before and during surgery. This early prediction is crucial as it allows for the timely implementation of preventive measures. Early monitoring of amylase in drainage fluid, as emphasized in international guidelines, can help detect CR-POPF at an early stage. Additionally, the early administration of appropriate medications, such as pancreatic secretion inhibitors, can potentially reduce the severity of CR-POPF. Previous studies have shown that timely intervention based on accurate risk prediction can significantly improve patient outcomes.

Our findings expand on prior research by identifying BMI and splenic thickness as novel risk factors for CR-POPF after splenectomy, which are not widely reported in the literature[21]. This differs from previous studies that merely emphasized “obesity” or “splenic volume” as qualitative risk factors[22,23]. Notably, BMI cannot be simply equated with obesity, which may be attributed to the specific pathophysiological characteristics of cirrhotic patients, such as latent edema and abnormal intra-abdominal fat distribution caused by portal hypertension; thus, BMI is a more sensitive quantitative indicator than the qualitative description of “obesity” in this population[24]. Splenic thickness and splenic volume are correlated, but not interchangeable. Previous studies have mostly focused on splenic volume; however, its clinical application is limited by complex measurements and high medical costs. In contrast, splenic thickness can be quickly obtained using preoperative ultrasound. While previous studies highlighted obesity, liver function, or surgical technique as potential risk factors, our multivariate analysis did not confirm Child-Pugh class as an independent predictor, likely because the specific cohort of cirrhotic patients exhibits portal hypertension-related anatomical changes such as splenic vein dilation, which may play a more direct role in CR-POPF pathogenesis than liver function alone[24]. Additionally, our focus on the stapler division provides actionable insights for intra-operative decision making, complementing prior work that emphasizes patient-related factors.

Currently, the treatment of CR-POPF mainly involves continuous drainage, antibiotic therapy, and the use of pancreatic secretion inhibitors[25]. In our experience, continuous drainage remains the cornerstone of CR-POPF treatment[26,27]. For most patients with biochemical leakage, no special treatment is required, and amylase levels in the drainage fluid return to normal within 7-10 days of splenectomy. For CR-POPF (grades B/C), while the effect of somatostatin remains controversial, our cases showed that after 2-3 weeks of drainage, the amylase level in the drainage fluid decreased significantly or even returned to normal. We believe that, in most cases, the importance of continuous drainage outweighs the use of trypsin inhibitors.

However, this study had several limitations. First, as this was a single-center retrospective study, the findings may be limited by institutional-specific surgical techniques and patient demographics, which reduce generalizability. Multicenter prospective studies with larger sample sizes are required to validate our nomogram. Second, the total number of CR-POPF events in this study was small, while the number of predictors was relatively large (events per variable < 10), which may lead to unstable model coefficients. Although penalized logistic regression was used to mitigate overfitting caused by excessive predictors, bootstrap internal validation was performed to calibrate that AUC optimism bias, residual bias, or instability in the results may still exist. Third, the definition of “soft, firm, and hard” pancreatic texture was based on intra-operative palpation (a standard clinical practice). Although the pancreatic CT values were integrated for supplementary assessment, there was no accurate correspondence between the two. Therefore, the development of histopathological or imaging-derived markers for pancreatic fibrosis may further optimize the risk stratification. Despite these limitations, the study identified BMI and splenic thickness as independent risk factors for CR-POPF in patients with cirrhotic splenomegaly and provided a practical nomogram, which has important clinical guiding value.

CONCLUSION

In summary, this study identified BMI and splenic thickness as independent risk factors of CR-POPF after splenectomy in patients with cirrhosis. The nomogram effectively predicted the individual CR-POPF risk, offering a valuable tool for personalized perioperative management. By implementing targeted preventive strategies based on risk assessments, clinicians can reduce the incidence of CR-POPF and improve patient outcomes.