Prognostic analysis of acute pancreatitis due to different etiologies and establishment and validation of an early prediction model

Abstract

Background

Acute pancreatitis (AP), an acute inflammatory condition commonly encountered in patients admitted to intensive care unit, is mainly caused by gallstones, alcohol abuse, or hyperlipidemia. Traditional views focus on single-etiology AP; however, Chinese studies have shown that mixed-etiology AP (two or more etiologies) is also fairly common. Comparing the clinical characteristics of AP with different etiologies is crucial for guiding personalized treatment strategies and understanding disease severity and recurrence patterns. However, few studies have explored the prognostic differences between single-etiology and mixed-etiology AP; moreover, the predictors of disease severity in mixed-etiology AP cases remain unclear.

Methods

This retrospective cohort study compared severity and prognostic outcomes between single-etiology AP and mixed-etiology AP and established a predictive model for severe AP (SAP) in mixed-etiology AP patients. For this purpose, patients diagnosed with AP from January 2019 to December 2023 at the First Affiliated Hospital of Xi’an Jiaotong University were recruited.

Results

Of 630 AP patients, 18.3% (n = 115) had mixed-etiology AP. Compared to single-etiology AP, mixed-etiology AP exhibited significant associations with higher incidences of SAP, organ failure, and local/systemic complications (all P < 0.05). Among mixed-etiology subgroups, cholelithiasis combined with hypertriglyceridemia showed an association with the highest persistent organ failure rate (50.0%), while cholelithiasis combined with alcohol was linked with more local complications and higher hospitalization costs. By employing least absolute shrinkage and selection operator regression and multivariate logistic regression analyses, this study constructed a SAP prediction model for mixed-etiology AP based on albumin (odds ratio [OR]: 0.87, 95% confidence interval [CI]: 0.79–0.95), serum calcium (OR: 0.08, 95% CI: 0.01–0.45), and D-dimer (OR: 1.13, 95% CI: 1.02–1.26). The model showed an area under the curve (AUC) value of 0.857, thus outperforming APACHE II, SOFA, and BISAP scores (AUC = 0.745, 0.764, and 0.748, respectively); additionally, it exhibited good consistency and positive net clinical benefits, as validated by calibration procedures and decision curve analyses.

Conclusion

Mixed-etiology AP might pose a high risk of adverse clinical outcomes in AP patients. The three-marker prediction model (albumin, serum calcium, and D-dimer) established in the current single-center retrospective study shows high accuracy and potential clinical utility for predicting SAP in mixed-etiology AP, which may enable to early recognize high-risk patients. However, because of limitations related to study design, further prospective multicenter studies are required for validating the generalizability and clinical utility of the model and optimizing treatment strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12944-026-02890-w.

Introduction

Acute pancreatitis (AP), a prevalent inflammatory abdominal disorder, has an annual incidence of 4.9–73.4 per 100,000 individuals, a global annual incidence increase of 3.07%, and a mortality rate of ~ 1.16 per 100,000 person-years [1, 2]. Although the pathogenesis of AP remains unclear, it is considered that damaging factors activate trypsinogen in pancreatic acinar cells, triggering autodigestion and inflammation. In severe cases, this process progresses from local tissue damage to systemic inflammatory response syndrome (SIRS) and organ failure [3–6]. While mild cases improve within a week after conservative treatment, 20% of the cases advance to severe acute pancreatitis (SAP), with 20–40% mortality rate [7], posing a significant clinical burden.

AP has more complex etiologies than other acute abdominal disorders, and its etiological management has gained increasing attention globally. Because significant differences in pathogenesis, severity, prognosis, and treatment response exist across different etiologies [2, 8–12], it is crucial to identify the actual etiology of AP for prompt management, optimal utilization of healthcare resources, and reduction in recurrence incidence. Globally, gallstones are the prevalent causative agent, followed by alcohol consumption [1, 8, 13]. However, hypertriglyceridemia (HTG) has substituted alcohol consumption as the second major etiology of AP in China [14, 15]. Traditionally, AP was considered a single-etiology disorder; however, accumulating clinical evidence and improved diagnostic techniques have confirmed that “mixed-etiology AP” (two or more concurrent etiologies) is also prevalent [16, 17]. This indicates that the etiological complexity of AP has been underestimated. More importantly, while some prognostic studies indicate that hyperlipidemic AP is prone to severe outcomes [18–20] and that hyperlipidemia + alcohol mixed-etiology AP has a poor prognosis [16, 17, 21]), the prognostic differences between most single-etiology and mixed-etiology subtypes remain poorly explored. This critical gap highlights the need to systematically investigate the prognostic characteristics of both single-etiology and mixed-etiology AP for precise clinical management.

Currently, there is a lack of specific tools to predict SAP in mixed-etiology AP. Moreover, the commonly used severity scores (bedside index for severity in acute pancreatitis [BISAP], acute physiology and chronic health evaluation II [APACHE II], and sequential organ failure assessment [SOFA]) have inherent limitations that make them suboptimal for this subgroup. BISAP enables rapid 24-h assessment but has low sensitivity [22–24]. APACHE II has good accuracy, but it is very complex for routine use [25, 26]. The SOFA score, as a general organ function assessment tool for critical illnesses, has inherent limitations: it lacks disease-specific optimization, is prone to inconsistent assessment due to ambiguous standards for certain components, primarily reflects established organ dysfunction rather than early risk, and is affected by missing data and inter-rater variability [27–30]. Importantly, these scores were developed for general AP and not for mixed-etiology AP; consequently, they fail to account for the unique pathogenic complexity of mixed-etiology AP.

Combined biomarkers have shown significant value in predicting AP prognosis (such as triglyceride and glucose [TyG] index). Furthermore, numerous biomarkers (such as procalcitonin [PCT], albumin, C-reactive protein [CRP], serum calcium, and D-dimer [D-D]) have been utilized to predict AP severity and adverse outcomes [25, 31–38]. However, it remains unknown whether these biomarkers could adequately predict severe disease in mixed-etiology AP. Given that mixed-etiology AP may have a poor prognosis [16, 17, 21, 39], an applicable model for early severity prediction is required to facilitate clinical decision-making.

Therefore, a retrospective study was performed to: (1) compare severity and prognosis between single-etiology AP and mixed-etiology AP and (2) construct and validate an early prediction model for SAP in mixed-etiology AP for providing evidence to optimize the clinical management of AP.

Materials and methods

Patients

This study recruited 630 inpatients diagnosed to have AP from January 2019 to December 2023 at the First Affiliated Hospital of Xi’an Jiaotong University. The hospital’s ethics committee approved this study [institutional review board reference number: XJTUAF2025LSYY-654].

Inclusion criteria

AP patients meeting two of three diagnostic criteria were recruited: (1) persistent abdominal pain as the hallmark clinical symptom; (2) serum lipase and/or amylase levels greater than three-fold the upper limit of normal; and (3) abdominal CT or ultrasound imaging revealing pathognomonic signs of pancreatitis. Additional enrollment prerequisites were defined as follows: (1) availability of the full set of laboratory parameters as well as clinical records obtained within the first 48 h of hospital admission and (2) age at the time of enrollment ≥ 18 years.

Exclusion criteria

The following criteria were applied to exclude patients: (1) lactating or pregnant females (to avoid potential risk to mothers and offspring, in accordance with ethical requirements); (2) chronic pancreatitis (to differentiate from AP in terms of pathological mechanisms and progression); (3) AP combined with cancers or tumors (to exclude interference from tumor-induced inflammation, nutritional disorders, or organ dysfunction on AP assessment); (4) pancreatitis caused by other etiologies (such as medications, endoscopic retrograde cholangiopancreatography [ERCP], or idiopathic pancreatitis; for focusing on core etiologies); (5) inconsistency in conventional treatment (fasting, acid suppression, enzyme inhibition, anti-infection therapy, fluid resuscitation, etc.; to avoid the effect of treatment differences on AP progression and prognosis analysis); (6) incomplete laboratory or imaging data (to ensure accurate AP diagnosis, etiology classification, and severity assessment); and (7) patients lost to follow-up (to secure complete prognostic data for outcome statistics).

Classification basis of etiology

Diagnostic criteria for acute biliary pancreatitis (ABP): imaging examinations reveal gallbladder or biliary stones or common bile duct dilatation. Additionally, the patient must show abnormal results for a minimum of two laboratory parameters: alanine aminotransferase (ALT) > 100 U/L with ALT higher than aspartate aminotransferase (AST); serum bilirubin > 1.9 mg/dL; and alkaline phosphatase (ALP) > 195 U/L accompanied by gamma-glutamyl transferase (GGT) > 45 U/L. Diagnostic criteria for alcoholic AP: long-term alcohol consumption (50 g/day), heavy alcohol consumption before pancreatitis onset, or alcohol drinking history of > 5 years. Diagnostic criterion for HTG-AP: serum triglyceride (TG) level > 11.3 mmol/L (or between 5.65 and 11.3 mmol/L with chylous serum); additionally, cases with transient, stress-related TG elevation during the acute phase of AP were excluded for confirming that the identified HTG was the etiological factor of AP rather than a stress-induced secondary change. Diagnostic criteria for mixed-etiology AP: patients meeting two or more than two diagnostic criteria for ABP, alcoholic AP, or HTG-AP [2, 19].

Patients were assigned to four groups based on the diagnostic criteria: cholelithiasis (n = 371), alcohol (n = 106), HTG (n = 102), and AP due to other causes (n = 51). Mixed-etiology AP cases were further subclassified into cholelithiasis–HTG (n = 26), cholelithiasis–alcohol (n = 36), and HTG–alcohol (n = 53) subtypes. To develop the prediction model, patients with mixed-etiology AP were reclassified into two groups on the basis of the revised 2012 Atlanta Acute Pancreatitis Classification Criteria [6]: non-SAP (NSAP) (n = 62) and SAP (n = 53) (Fig. 1).

Fig. 1.

Study screening and enrollment

Data collection

We retrospectively screened the electronic medical records of the recruited patients and collected the following data: demographic features (age, gender, and body mass index [BMI]), etiology (cholelithiasis, alcohol, HTG, and other etiologies), comorbidities (coronary artery disease, hypertension, fatty liver, and diabetes mellitus), smoking history, drinking history, district, organ failure (respiratory, renal, or cardiovascular), systemic complications (sepsis, infected pancreatic necrosis [IPN], or SIRS), local complications (pancreatic pseudo cyst [PPC], acute necrotic collection [ANC], acute peripancreatic fluid collection [APFC], or walled-off necrosis [WON]), severity (NSAP or SAP), interventions (mechanical ventilation [MV], continuous renal replacement therapy [CRRT], or vasoactive drugs), cost of hospitalization, intensive care unit (ICU) admission, hospital length of stay (LOS), and 28-day mortality. The following clinical vital parameters and laboratory indices were measured within 48 h of hospital admission: heart rate (HR), respiratory rate (RR), systolic blood pressure (SBP), diastolic blood pressure (DBP), temperature (T), mean arterial pressure (MAP), red blood cell (RBC), white blood cell (WBC), hemoglobin (HB), hematocrit (HCT), erythrocyte volume distribution width (RDW), platelet (PLT), neutrophil (Neu), lymphocyte (Lym), AST, ALT, alkaline phosphatase (ALP), albumin (ALB), gamma-glutamyl transferase (GGT), total bilirubin (TBIL), serum creatinine (Cr), serum calcium (Ca²⁺), blood urea nitrogen (BUN), amylase (AMY), lipase (LPS), D-D, CRP, PCT, lactate, and TyG index. The APACHE II, BISAP, and SOFA scores were also documented.

Statistical analysis

R software (version 4.5.1) was used to perform statistical analyses, with P < 0.05 considered statistically significant. For missing data, each variable’s missing rate was assessed: variables (AMY (3.97%), LPS (5.87%), PCT (10.63%), and BMI (17.62%)) with < 20% missing data were supplemented through multiple imputation, while those with ≥ 20% missing data were excluded to avoid over-imputation bias and ensure the quality of statistical analysis. Continuous data are presented as mean ± standard deviation (x ± s) and compared between groups with t-tests when normality and variance homogeneity were satisfied; otherwise, data are reported as median (interquartile range, [IQR]) and compared between groups by using Kruskal–Wallis tests. Categorical data are presented as proportions, and group comparison was achieved using chi-square test or Fisher’s exact test. The false discovery rate originating from multiple comparisons was controlled with the Benjamini–Hochberg procedure. For predictive modeling, Least Absolute Shrinkage and Selection Operator (LASSO) regression was first conducted to screen potential predictors of mixed-etiology AP severity (to avoid overfitting by penalizing redundancy). Multicollinearity was then evaluated using variance inflation factor (VIF) (no severe multicollinearity indicated by VIF < 10). Selected variables were included in multivariate logistic regression analysis for determining independent predictors of severe mixed-etiology AP, and the results were used to construct a nomogram. The following approaches were used for evaluating the model: (1) receiver operating characteristic (ROC) curves (with area under the curve [AUC] value for discriminative ability); (2) Youden index (sensitivity + specificity − 1) for determining sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the optimal cut-off value; (3) calibration curves (for comparing predicted/observed outcomes) with Hosmer–Lemeshow tests (P > 0.05 indicates good calibration); and decision curve analysis (DCA) for clinical utility. The model was validated by internal validation through bootstrap resampling (1000 iterations) to calculate the AUC, sensitivity, and specificity of the model in resampled datasets and evaluate model stability. Because of the single-center retrospective cohort design, external validation was not conducted, and future multicenter prospective studies have been planned to determine the model’s generalization ability (Fig. 2).

Fig. 2.

Predictive model development and validation

Results

Baseline patient characteristics

As shown in Fig. 1, we recruited 630 AP patients (383 male patients (60.8%), median age: 46.0 years [IQR: 35.0–57.75], median BMI: 25.3 kg/m² [IQR: 22.5–27.8]). The most common comorbidity was fatty liver (320 patients; 50.8%). Cholelithiasis was the leading etiology (371 patients; 58.9%) followed by alcohol consumption (106 patients; 16.8%) and HTG (102 patients; 16.2%). Organ failure occurred in a subset of patients, with respiratory involvement being the most common, including 176 patients (27.9%) with transient respiratory failure and 213 patients (33.8%) with persistent respiratory failure. SIRS was detected in 360 patients (57.1%). According to severity classification, 411 (65.2%) and 219 (34.8%) patients had NSAP and SAP, respectively. The median LOS was 12.0 days (IQR: 8.0–18.0), and the median hospitalization cost was ¥42048.03 (IQR: 23528.78–76697.82). The 28-day mortality rate was 3.3% (21 patients) (Supplementary Table S1).

Comparison of interventions, complications, severity of AP, and clinical prognosis in single-etiology and mixed-etiology AP patients with different combinations of AP

Comparison of cholelithiasis and its related mixed-etiology AP

Patients were categorized into three etiological subgroups: cholelithiasis alone (n = 309), cholelithiasis + HTG (n = 26), and cholelithiasis + alcohol (n = 36). These groups exhibited significant differences in disease severity and local complications, although the small sample size of the two mixed-etiology subgroups (particularly cholelithiasis + HTG, n = 26) may have limited the statistical power of this evaluation. The cholelithiasis + HTG group had a significantly greater number of patients classified as having SAP (57.69%) compared to the cholelithiasis alone group (32.69%). The incidence of ANC/WON also differed significantly and was highest in the cholelithiasis + alcohol group (41.67%), intermediate in the cholelithiasis + HTG group (26.92%), and lowest in the cholelithiasis alone group (20.39%). The subgroups showed no significant differences in systemic complications, organ failure subtypes, intervention requirements, or 28-day mortality (P > 0.05) (Table 1); however, these results might have been influenced by insufficient statistical power caused by the small sample size in the mixed-etiology subgroups.

Table 1.

Comparison of cholelithiasis and its related mixed-etiology AP

Variables	Cholelithiasis (n = 309)	Cholelithiasis + HTG (n = 26)	Cholelithiasis + Alcohol (n = 36)	P	Adjusted P
Transient organ failure, n (%)
Respiratory	83 (26.86)	3 (11.54)	9 (25.00)	0.227
Renal	11 (3.56)	2 (7.69)	1 (2.78)	0.372
Cardiovascular	21 (6.80)	3 (11.54)	3 (8.33)	0.468
Persistent organ failure, n (%)
Respiratory	100 (32.36)	13 (50.00)	16 (44.44)	0.085
Renal	18 (5.83)	3 (11.54)	4 (11.11)	0.206
Cardiovascular	13 (4.21)	1 (3.85)	3 (8.33)	0.395
Systemic complications, n (%)
Sepsis	16 (5.18)	1 (3.85)	1 (2.78)	1.000
IPN	16 (5.18)	1 (3.85)	0 (0.00)	0.458
SIRS	156 (50.49)	19 (73.08)	18 (50.00)	0.083
Local complications, n (%)
APFC/PPC	101 (32.69)	9 (34.62)	18 (50.00)	0.118
ANC/WON	63 (20.39) a	7 (26.92) ab	15 (41.67) b	0.014	0.022
Severity, n (%)				0.019	0.055
NSAP	208 (67.31) a	11 (42.31) b	20 (55.56) ab
SAP	101 (32.69) a	15 (57.69) b	16 (44.44) ab
Interventions, n (%)
Vasoactive drugs	34 (11.00)	4 (15.38)	6 (16.67)	0.400
MV	54 (17.48)	6 (23.08)	8 (22.22)	0.636
CRRT	23 (7.44)	2 (7.69)	3 (8.33)	0.925
Hospitalization cost, (IQR), ¥	48611.59 (32185.57,80617.75)	56644.89 (25142.10,84745.90)	47780.24 (31962.94,101482.58)	0.796
ICU admission, n (%)	66 (21.36)	9 (34.62)	13 (36.11)	0.057
LOS, (IQR), days	13.00 (10.00,19.00)	15.00 (8.00,21.50)	13.00 (9.75,20.00)	0.792
28-day mortality, n (%)	14 (4.53)	0 (0.00)	0 (0.00)	0.446

The bold values mean the P <0.05. Adjusted P stands for the P after Benjamini-Hochberg correction. A statistically significant difference was observed between two results with different labels (a, b). If the two results have the same label, the differences were not statistically significant

Abbreviations: IQR Interquartile range, HTG Hypertriglyceridemia, IPN Infected pancreatic necrosis, SIRS Systemic inflammatory response syndrome, APFC Acute peripancreatic fluid collection, PPC Pancreatic pseudo cyst, ANC Acute necrotic collection, WON Walled-off necrosis, NSAP Non-severe acute pancreatitis, SAP Severe acute pancreatitis, MV Mechanical ventilation, CRRT Continuous renal replacement therapy, ICU Intensive care unit, LOS Length of stay

Comparison of HTG and its related mixed-etiology AP

Patients were assigned to three groups: HTG alone (n = 102), cholelithiasis + HTG (n = 26), and HTG + alcohol (n = 53). While the majority of differences were nonsignificant (P > 0.05), the mixed-etiology groups—particularly cholelithiasis + HTG—exhibited numerical tendencies toward more severe outcomes but without statistical significance; this result might be because the mixed-etiology subgroup had a small sample size. This group presented the highest rates of SAP (57.69%) and mechanical ventilation use (23.08%), along with a significantly greater median hospital LOS (15.0 days, P = 0.034, adjusted P = 0.030). Patients with HTG alone generally had better outcomes across most measures. These results indicate a potential association between mixed-etiology AP and greater disease severity and resource utilization (Table 2); however, this hypothesis requires validation through studies using a larger sample size.

Table 2.

Comparison of HTG and its related mixed-etiology AP

Variables	HTG (n = 102)	Cholelithiasis + HTG (n = 26)	HTG + Alcohol (n = 53)	P	Adjusted P
Transient organ failure, n (%)
Respiratory	28 (27.45)	3 (11.54)	16 (30.19)	0.181
Renal	4 (3.92)	2 (7.69)	5 (9.43)	0.343
Cardiovascular	6 (5.88)	3 (11.54)	1 (1.89)	0.180
Persistent organ failure, n (%)
Respiratory	35 (34.31)	13 (50.00)	21 (39.62)	0.327
Renal	8 (7.84)	3 (11.54)	2 (3.77)	0.433
Cardiovascular	6 (5.88)	1 (3.85)	4 (7.55)	0.910
Systemic complications, n (%)
Sepsis	4 (3.92)	1 (3.85)	0 (0.00)	0.345
IPN	3 (2.94)	1 (3.85)	1 (1.89)	1.000
SIRS	68 (66.67)	19 (73.08)	34 (64.15)	0.730
Local complications, n (%)
APFC/PPC	25 (24.51)	9 (34.62)	22 (41.51)	0.086
ANC/WON	18 (17.65)	7 (26.92)	14 (26.42)	0.349
Severity, n (%)				0.139
NSAP	65 (63.73)	11 (42.31)	31 (58.49)
SAP	37 (36.27)	15 (57.69)	22 (41.51)
Interventions, n (%)
Vasoactive drugs	12 (11.76)	4 (15.38)	5 (9.43)	0.738
MV	15 (14.71)	6 (23.08)	13 (24.53)	0.377
CRRT	9 (8.82)	2 (7.69)	7 (13.21)	0.631
Hospitalization cost, (IQR), ¥	29617.79 (16796.31,53251.04)	56644.89 (25142.10,84745.90)	40836.06 (15257.90,89169.75)	0.051
ICU admission, n (%)	29 (28.43)	9 (34.62)	21 (39.62)	0.360
LOS, (IQR), days	10.00 (7.00,14.00)a	15.00 (8.00,21.50) b	11.00 (7.00,17.00) ab	0.034	0.030
28-day mortality, n (%)	5 (4.90)	0 (0.00)	0 (0.00)	0.214

The bold values mean the P <0.05. Adjusted P stands for the P after Benjamini-Hochberg correction. A statistically significant difference was observed between two results with different labels (a, b). If the two results have the same label, the differences were not statistically significant

Abbreviations: IQR Interquartile range, HTG Hypertriglyceridemia, IPN Infected pancreatic necrosis, SIRS Systemic inflammatory response syndrome, APFC Acute peripancreatic fluid collection, PPC Pancreatic pseudo cyst, ANC Acute necrotic collection, WON Walled-off necrosis, NSAP Non-severe acute pancreatitis, SAP Severe acute pancreatitis, MV Mechanical ventilation, CRRT Continuous renal replacement therapy, ICU Intensive care unit, LOS Length of stay

Comparison of alcohol and its related mixed-etiology AP

The hospitalization cost differed significantly among the alcohol-associated subgroups (alcohol alone, n = 53; cholelithiasis + alcohol, n = 36; HTG + alcohol, n = 53; P = 0.038, adjusted P = 0.022). The cholelithiasis + alcohol group incurred the highest hospital expense (median ¥47780), followed by the HTG + alcohol group (¥40836) and the alcohol alone group (¥31418). Although other endpoints were statistically nonsignificant (P > 0.05), the cholelithiasis + alcohol group showed the highest numerical rates of SAP (44%), ANC/WON (42%), MV (22%), and ICU admission (36%); however, these numerical trends were considered preliminary given the relatively small sample size of this subgroup (Table 3).

Table 3.

Comparison of alcohol and its related mixed-etiology AP

Variables	Alcohol (n = 53)	Cholelithiasis + Alcohol (n = 36)	HTG + Alcohol (n = 53)	P	Adjusted P
Transient organ failure, n (%)
Respiratory	17 (32.08)	9 (25.00)	16 (30.19)	0.767
Renal	3 (5.66)	1 (2.78)	5 (9.43)	0.472
Cardiovascular	2 (3.77)	3 (8.33)	1 (1.89)	0.340
Persistent organ failure, n (%)
Respiratory	18 (33.96)	16 (44.44)	21 (39.62)	0.600
Renal	2 (3.77)	4 (11.11)	2 (3.77)	0.376
Cardiovascular	1 (1.89)	3 (8.33)	4 (7.55)	0.342
Systemic complications, n (%)
Sepsis	2 (3.77)	1 (2.78)	0 (0.00)	0.471
IPN	5 (9.43)	0 (0.00)	1 (1.89)	0.074
SIRS	33 (62.26)	18 (50.00)	34 (64.15)	0.370
Local complications, n (%)
APFC/PPC	20 (37.74)	18 (50.00)	22 (41.51)	0.512
ANC/WON	18 (33.96)	15 (41.67)	14 (26.42)	0.320
Severity, n (%)				0.566
NSAP	35 (66.04)	20 (55.56)	31 (58.49)
SAP	18 (33.96)	16 (44.44)	22 (41.51)
Interventions, n (%)
Vasoactive drugs	3 (5.66)	6 (16.67)	5 (9.43)	0.230
MV	9 (16.98)	8 (22.22)	13 (24.53)	0.734
CRRT	2 (3.77)	3 (8.33)	7 (13.21)	0.221
Hospitalization cost, (IQR), ¥	31418.37 (17227.65,62997.63) a	47780.24 (31962.94,101482.58) b	40836.06 (15257.90,89169.75) ab	0.038	0.022
ICU admission, n (%)	13 (24.53)	13 (36.11)	21 (39.62)	0.232
LOS, (IQR), days	12.00 (7.00,15.00)	13.00 (9.75,20.00)	11.00 (7.00,17.00)	0.185
28-day mortality, n (%)	1 (1.89)	0 (0.00)	0 (0.00)	1.000

The bold values mean the P <0.05. Adjusted P stands for the P after Benjamini-Hochberg correction. A statistically significant difference was observed between two results with different labels (a, b). If the two results have the same label, the differences were not statistically significant

Abbreviations:IQR Interquartile range, HTG Hypertriglyceridemia, IPN Infected pancreatic necrosis, SIRS Systemic inflammatory response syndrome, APFC Acute peripancreatic fluid collection, PPC Pancreatic pseudo cyst, ANC Acute necrotic collection, WON Walled-off necrosis, NSAP Non-severe acute pancreatitis, SAP Severe acute pancreatitis, MV mechanical ventilation, CRRT Continuous renal replacement therapy, ICU Intensive care unit, LOS Length of stay

Comparison of general characteristics, complications, disease severity, interventions, and clinical prognosis in mixed-etiology AP patients in the NSAP and SAP groups

Comparative analysis of the derivation cohorts (NSAP and SAP, n = 62 and 53, respectively) revealed comparable baseline characteristics between both groups (all P > 0.05). Compared to NSAP, SAP showed a relationship with higher incidences of persistent respiratory failure (0% vs. 94%, P < 0.001) and renal failure (0% vs. 17%, P = 0.002); more local complications (APFC/PPC: 29% vs. 58%, ANC/WON: 19% vs. 45%; both P ≤ 0.003); and greater need for vasoactive drugs (6% vs. 26%), MV (3% vs. 47%), and CRRT (3% vs. 19%; all P ≤ 0.006). Additionally, compared to NSAP patients, SAP patients had longer hospital LOS (11 vs. 16 days), higher hospital expense (¥30999 vs. ¥78827), and greater number of ICU admission (8% vs. 72%; all P < 0.001), which indicated a tendency toward a higher organ failure complication burden and increased resource utilization (Supplementary Table S2).

Clinical variables of the mixed-etiology NSAP and SAP groups

The NSAP and SAP groups showed no significant differences in the RR, DBP, SBP, MAP, T, WBC, HB, HCT, Neu, Lym, ALT, GGT, TBIL, AMY, LPS, CRP, lactate, or TyG index (all P > 0.05). However, the SAP group exhibited remarkably greater HR, RDW, AST, BUN, Cr, D-D, and PCT values and lower PLT, ALP, ALB, and Ca²⁺ values in comparison with the NSAP group (all P < 0.05). Moreover, the SAP groups showed significantly elevated clinical scores, including APACHE II, SOFA, and BISAP (all P < 0.001). These results demonstrate more remarkable association of SAP with abnormalities in hemodynamic, hematological, biochemical, and coagulation parameters as well as with greater organ dysfunction and disease severity scores (Table 4).

Table 4.

Clinical variables of the mixed-etiology NSAP and SAP groups

Variables	NSAP (n = 62)	SAP (n = 53)	P
HR, (frequency/minute, Mean ± SD)	95.35 ± 20.91	107.55 ± 22.65	0.003
RR, (frequency/minute, IQR)	20.00 (19.00, 20.75)	20.00 (20.00, 24.00)	0.052
DBP, (mmHg, IQR)	127.00 (120.00, 138.75)	133.00 (120.00, 140.00)	0.254
SBP, (mmHg, IQR)	88.00 (80.00, 97.75)	86.00 (73.00, 96.00)	0.163
MAP, (mmHg) (Mean ± SD)	114.26 ± 13.63	115.38 ± 19.23	0.722
T, (℃, IQR)	36.50 (36.20, 36.60)	36.50 (36.20, 36.70)	0.739
WBC, (×109/L, Mean ± SD)	11.89 ± 4.26	12.40 ± 6.07	0.597
HB, (g/L, Mean ± SD)	141.32 ± 29.99	131.28 ± 33.77	0.094
HBC, (×1012/L, Mean ± SD)	4.47 ± 0.90	4.19 ± 1.06	0.127
HCT, (%, Mean ± SD)	42.29 ± 8.76	39.60 ± 9.83	0.124
RDW, (%, IQR)	12.85 (12.50, 13.47)	13.10 (12.70, 13.70)	0.049
PLT, (×109/L, Mean ± SD)	207.69 ± 80.26	176.57 ± 82.74	0.043
Neu, (×109/L, IQR)	1.11 (0.72, 1.67)	1.01 (0.58, 1.25)	0.136
Lym, (×109/L, Mean ± SD)	9.34 ± 5.25	10.71 ± 5.80	0.186
AST, (U/L, IQR)	25.00 (18.00, 36.75)	42.00 (27.00, 61.00)	0.000
ALT, (U/L, IQR)	25.50 (17.25, 44.00)	25.00 (18.00, 39.00)	0.762
ALP, (U/L, IQR)	86.00 (64.50, 113.00)	68.00 (51.00, 99.00)	0.011
GGT, (U/L, IQR)	75.50 (40.50, 148.75)	48.00 (35.00, 106.00)	0.111
TBIL, (µmol/L, IQR)	21.65 (16.90, 42.12)	25.90 (18.30, 40.30)	0.481
ALB, (g/L, Mean ± SD)	36.24 ± 6.31	29.94 ± 5.07	0.000
BUN, (mmol/L, IQR)	4.81 (3.71, 6.28)	7.80 (4.31, 11.56)	0.000
Cr, (µmol/L, IQR)	60.00 (50.00, 67.50)	71.00 (50.00, 145.00)	0.034
Ca2+, (mmol/L, IQR)	2.12 (2.00, 2.25)	1.85 (1.63, 2.03)	0.000
AMY, (U/L, IQR)	172.50 (68.25, 447.00)	228.00 (82.00, 449.00)	0.461
LPS, (U/L, IQR)	691.50 (302.75, 1604.50)	940.00 (231.00, 1638.00)	0.949
D-D, (mg/L, IQR)	1.92 (0.92, 4.17)	5.46 (3.50, 10.35)	0.000
CRP, (mg/L, IQR)	110.70 (44.15, 243.55)	158.20 (72.40, 231.20)	0.187
PCT, (ng/L, IQR)	0.33 (0.14, 1.51)	1.75 (0.44, 5.10)	0.000
Lactate, (mmol/L, IQR)	1.20 (0.90, 1.90)	1.70 (1.10, 2.20)	0.060
TyG index, (IQR)	10.80 (8.88, 11.88)	10.43 (9.26, 11.40)	0.722

The bold values mean the P <0.05

Abbreviations:SAP Severe acute pancreatitis, NSAP Non severe acute pancreatitis, IQR Interquartile range, HR Heart rate, RR Respiratory rate, SBP Systolic diastolic blood pressure, DBP Diastolic blood pressure, MAP Mean arterial pressure, T Temperature, WBC White blood cell, HB  Hemoglobin, RBC Red blood cell, HCT Hematocrit, RDW Erythrocyte volume distribution width, PLT Platelet, Neu Neutrophils, Lym Lymphocyte, AST Aspartate transaminase, ALT Alanine aminotransferase, ALP Alkaline phosphatase, GGT Gamma-glutamyl transferase, TBIL Total bilirubin, ALB Albumin, BUN Blood urea nitrogen, Cr serum creatinine, Ca²⁺ Serum calcium, AMY Amylase, LPS Lipase, D-D D-dimer, CRP C-reactive protein, PCT Procalcitonin, TyG Triglyceride and glucose

Screening of variables with LASSO and multivariate logistic regression analyses for mixed-etiology SAP

During model development, LASSO regression identified eight predictive variables—HR, T, Lym, AST, ALB, BUN, Ca²⁺, and D-D. A moderately simple lambda value of 0.0449137 with minimal error was selected, together with a parsimonious model structure as a reference, to ensure satisfactory model fitting (Fig. 3). To assess potential multicollinearity among these eight predictors, VIF analysis was performed, yielding VIF values of 1.194, 1.077, 1.027, 1.040, 1.294, 1.292, 1.503, and 1.374 for HR, T, Lym, AST, ALB, BUN, Ca²⁺, and D-D, respectively. These eight variables showed VIF values of < 10, indicating no severe multicollinearity among them. The identified predictors were then added to univariate and multivariate logistic regression analyses. Finally, three independent predictors of mixed-etiology SAP, namely ALB (g/L), Ca²⁺ (mmol/L), and D-D (mg/L), were obtained (Fig. 4, Table 5). Among these predictors, ALB and Ca²⁺ were independent protective factors, while D-D was an independent risk factor.

Fig. 3.

LASSO regression-based screening of variables. Risk factor identification for mixed-etiology SAP with LASSO regression. A Coefficient profiles of candidate variables derived from LASSO regression. Eight variables were identified at the optimal lambda value. B After validating the LASSO model’s optimal lambda parameter, a vertical dashed line was drawn based on the 1-standard error criterion. The figure also shows the correlation between log (lambda) values and the binomial likelihood deviance curve

Fig. 4.

Assessment of (A) ALB, (B) Ca²⁺, and (C) D-D levels for differentiating SAP and NSAP patients. ALB: albumin; Ca²⁺: serum calcium; NSAP: non-severe acute pancreatitis; D-D: D-dimer; SAP: severe acute pancreatitis. ***P < 0.001

Table 5.

Multivariate logistic regression analysis in patients with mixed-etiology SAP

Variables	Univariate logistic regression analysis					Multivariate logistic regression analysis
	β	SE	Z	P	OR (95%CI)	β	SE	Z	P	OR (95%CI)
HR	0.03	0.01	2.81	0.005	1.03 (1.01 ~ 1.05)
T	0.43	0.32	1.36	0.172	1.54 (0.83 ~ 2.86)
Lym	0.05	0.03	1.32	0.187	1.05 (0.98 ~ 1.12)
ALB	−0.20	0.04	−4.59	0.000	0.82 (0.76 ~ 0.89)	−0.14	0.05	−3.01	0.003	0.87 (0.79 ~ 0.95)
Ca2+	−3.93	0.92	−4.26	0.000	0.02 (0.00 ~ 0.12)	−2.58	0.91	−2.83	0.005	0.08 (0.01 ~ 0.45)
BUN	0.15	0.05	3.09	0.002	1.16 (1.06 ~ 1.27)
AST	0.01	0.00	1.97	0.049	1.01 (1.01 ~ 1.02)
D-D	0.20	0.05	3.72	0.000	1.22 (1.10 ~ 1.36)	0.12	0.06	2.24	0.025	1.13 (1.02 ~ 1.26)
Intercept						8.95	2.27	3.95	0.000

The bold values mean the P < 0.05

HR Heart rate, T Temperature, Lym Lymphocyte, ALB Albumin, Ca²⁺ Serum calcium, BUN Blood urea nitrogen, AST Aspartate transaminase, D-D D-dimer, OR Odds ratio, CI Confidence interval

Development of a new mixed-etiology SAP prediction model

According to multivariate analysis outcomes, the following logistic regression equation was established: Logit = 8.95 − 0.14 ALB (g/L) − 2.58 Ca²⁺ (mmol/L) + 0.12 D-D (mg/L). A nomogram for predicting the progression of mixed-etiology AP to SAP was constructed from three independent determinants—ALB, Ca²⁺, and D-D (Fig. 5). By using the integrated point scale, the individual points assigned to each variable were combined to derive the total score; SAP development risk was then determined according to the corresponding probability on the nomogram axis.

Fig. 5.

Construction of a novel prediction model for mixed-etiology SAP. A nomogram was developed for predicting the risk of mixed-etiology SAP. Patients were assigned points according to their ALB (g/L), Ca²⁺ (mmol/L), and D-D (mg/L) levels by locating the corresponding values on the “ALB,” “Ca²⁺,” and “D-D” axes and drawing a vertical line to the top “Points” scale. These individual points were combined to derive a total score, which was mapped to the “Total Points” scale. Finally, a vertical line from the “Total Points” scale to the “Risk of mixed-etiology SAP” axis yielded the predicted risk. A representative clinical case is provided to validate the applicability of the scoring system in clinical settings: a male patient (age: 58 years) scored 150 points, corresponding to a 94.6% probability of mixed-etiology SAP, indicating a high-risk status

Evaluating and validating the mixed-etiology SAP prediction model

The AUC values for the mixed-etiology SAP prediction model, ALB, Ca²⁺, D-D, APACHE II, SOFA, and BISAP were 0.857, 0.783, 0.798, 0.785, 0.745, 0.764, and 0.748, respectively. From a clinical perspective, the model exhibited more meaningful performance advantage when combined with other performance metrics, including Youden’s Index. Based on data shown in Table 6, the Youden’s index of each variable was as follows: ALB (0.488), Ca²⁺ (0.483), D-D (0.481), APACHE II (0.433), SOFA (0.494), and BISAP (0.394). Notably, this mixed-etiology SAP prediction model achieved a Youden’s index of 0.636; this value was substantially higher than that for all individual predictors (range: 0.481–0.494) and existing clinical scoring systems (range: 0.394–0.494) in the context of this internal cohort. APACHE II (a traditional severity scoring system) showed a high sensitivity of 0.868 but a low specificity of 0.565, probably resulting in overidentification of non-severe cases and unnecessary clinical interventions. Although BISAP and SOFA scores were relatively balanced, they exhibited lower overall accuracy (AUC < 0.77). In comparison to these predictors, this model maintained a high sensitivity of 0.830 with a markedly improved specificity of 0.806. Additionally, the PPV (0.786) and NPV (0.847) of the model were consistently higher than those of the individual predictors and existing scoring systems, indicating its potential stronger ability to accurately stratify high-risk and low-risk patients with mixed-etiology SAP (Table 6).

Table 6.

Comparison of independent predictors and predictive models for mixed-etiology SAP

Index	AUC	Cut-off value	Youden’s Index	Sensitivity	Specificity	PPV	NPV	P-value
ALB	0.783	31.75 g/L	0.488	0.698	0.790	0.740	0.754	0.000
Ca2+	0.798	1.955 mmol/L	0.483	0.660	0.823	0.761	0.739	0.000
D-D	0.785	3.42 mg/L	0.481	0.755	0.726	0.702	0.776	0.000
APACHE II	0.745	5.5	0.433	0.868	0.565	0.630	0.833	0.000
SOFA	0.764	2.5	0.494	0.849	0.645	0.672	0.833	0.000
BISAP	0.748	1.5	0.394	0.717	0.677	0.655	0.737	0.000
SAP model	0.857	0.448	0.636	0.830	0.806	0.786	0.847	0.000

The bold values mean the P < 0.05

ALB Albumin, Ca²⁺ Serum calcium, D-D D-dimer; APACHE II Acute physiology and chronic health evaluation II, SOFA Sequential organ failure assessment, BISAP Bedside index for severity in acute pancreatitis, AUC Area under the curve, PPV Positive predictive value, NPV Negative predictive value

ROC curve analyses revealed that, within the internal cohort, the novel mixed-etiology SAP prediction model outperformed individual biomarkers (ALB, Ca²⁺, and D-D) and conventional scoring systems (APACHE II, SOFA, and BISAP) in SAP progression prediction in mixed-etiology AP patients (P < 0.001) (Figs. 6 and 7).

Fig. 6.

ROC curves of independent predictors and prediction models for mixed-etiology SAP. The area under the curve (AUC) values of the mixed-etiology SAP model, albumin (ALB), serum calcium (Ca), and D-dimer (D-D) were 0.857, 0.783, 0.798, and 0.785, respectively

Fig. 7.

ROC curves of the mixed-etiology SAP predictive model, APACHE II, BISAP, and SOFA scores. The AUC values of the mixed-etiology SAP model and the BISAP, APACHE II, and SOFA scores were 0.857, 0.748, 0.745, and 0.764, respectively

A calibration curve was plotted to evaluate model fit (Fig. 8). The Hosmer–Lemeshow test revealed good concordance between predicted SAP development and actual SAP occurrence in mixed-etiology AP patients (X² = 6.8248, P = 0.5556 > 0.05). DCA was conducted to further validate the nomogram’s predictive accuracy (Fig. 9). The nomogram outperformed the indicators ALB, Ca²⁺, and D-D and the APACHE II, SOFA, and BISAP clinical scores in terms of net clinical benefits and applicable high-risk threshold range. By using the bootstrap method to perform internal validation, 1000 repeated samplings were conducted for the internal validation of the model. Following this calibration, the model retained superior accuracy (AUC = 0.845).

Fig. 8.

Calibration curve corresponding to the mixed-etiology SAP predictive model. Calibration curve of the mixed-etiology SAP model, albumin (ALB), serum calcium (Ca²⁺), and D-dimer (D-D) A; Calibration curve of the mixed-etiology SAP model and APACHE II, SOFA, and BISAP scores B. An ideal reference line is illustrated by the gray diagonal line, and the model’s predictive performance is depicted by the colored curve. Concordance between predicted event outcomes and actual events increases as the black curve approaches the gray reference line

Fig. 9.

Decision curve analysis (DCA) to predict mixed-etiology SAP. Two sets of DCA were performed: one for the mixed-etiology SAP model along with albumin (ALB), D-dimer (D-D), and serum calcium (Ca²⁺) A, and another for the mixed-etiology SAP model compared with APACHE II, SOFA, and BISAP B. Threshold probability and the net benefit rate are plotted on x- and y-axes, respectively. The model is considered to provide a net benefit if its colored line is located above the two solid lines

Discussion

AP is a clinically challenging digestive disorder because of its etiological heterogeneity and varying prognosis. Recent studies [17] have increasingly focused on the impact of etiological complexity of AP on patient prognosis. In this study, the leading causative factor of AP was cholelithiasis. However, in this single-center study, the second most common etiology was acute alcoholic pancreatitis; this observation differs from the current etiological composition of AP in China [14, 15]. This discrepancy may be associated with the failure to measure TG levels promptly in clinical practice when AP is first diagnosed, leading to missing data during the period of elevated TG levels. This may also be related to the exclusion of patients due to missing data or loss to follow-up.

A key observation of this study is that mixed-etiology AP, accounting for 18.3% of the total cohort, was significantly associated with a worse prognosis compared to single-etiology AP. Specifically, mixed-etiology AP patients had greater incidences of SAP, persistent organ failure, and local/systemic complications; these outcomes align with previous studies [16, 17, 39], which noted that mixed-etiology AP, although less prevalent than single-etiology AP, is associated with adverse prognoses, which may be related to synergistic pathogenic effects. Subgroup analysis further refined these observations by identifying subtype-specific prognostic patterns: cholelithiasis + HTG showed a relationship with the highest incidence of persistent organ failure as well as an elevated risk of SAP, while cholelithiasis + alcohol was linked to increased local complications and higher hospitalization costs. These subtype-specific differences highlight the need for tailored risk stratification.

The poor prognosis of mixed-etiology AP (e.g., cholelithiasis + HTG and cholelithiasis + alcohol) may be associated with synergistic amplification of inflammatory and metabolic disturbances. For the cholelithiasis + HTG subtype, bile reflux and pancreatic duct obstruction (from cholelithiasis) directly activate pancreatic enzymes to initiate inflammation, while hypertriglyceridemia exacerbates injury through two key metabolic effects: elevated blood viscosity impairs pancreatic microcirculation, and triglyceride breakdown yields free fatty acids (FFAs) that cause necrosis of acinar cells [40, 41]. Notably, this dual insult may amplify inflammatory signaling: cholelithiasis-induced IL-6 and TNF-α upregulation may further promote lipolysis, while hyperlipidemia-derived FFAs may activate NLRP3 inflammasomes, which may enhance proinflammatory cytokine release. Hyperlipidemia may also impair biliary excretion and may promote gallstone formation, potentially creating a vicious cycle that may sustain inflammation [3, 7, 41–43]. Chronic alcohol intake induces dysfunction of zymogen granules (triggering autodigestion) and oxidative stress through reactive oxygen species (ROS) production [7, 44], while cholelithiasis-induced bile reflux may exacerbate endoplasmic reticulum stress in acinar cells. The synergistic effect may lower the threshold for pancreatic injury—alcohol-induced endothelial dysfunction may enhance bile-induced inflammation, and ROS may amplify cholelithiasis-related necrosis, which might elevate the risk of local abnormalities such as ANC/WON.

Growing evidence confirms that most AP cases have multiple etiologies; moreover, mixed-etiology AP, associated with worse severity and prognosis, requires targeted prediction tools for early intervention. Given this background, the present study developed a mixed-etiology SAP prediction model based on LASSO and logistic regression and identified ALB, Ca²⁺, and D-D as independent predictors—a key distinction from existing etiology-specific models that focus solely on HTG-AP or ABP [36, 37, 45]. For instance, Wang et al. (2025) developed an HTG-AP severity model (AUC = 0.966) [36]; however, this model is limited to HTG alone cases and requires CT-derived visceral adipose tissue index (not routinely available in primary hospitals). Similarly, Liu et al. (2022) constructed a risk prediction score for estimating acute HTG-AP severity (AUC = 0.929) [45], which relies on apolipoprotein A1 (ApoA1) (a non-routine lipid marker) and chest imaging findings. Although these models were developed to predict AP severity, they are limited to a single etiology (HTG or cholelithiasis) and employ predictive indicators that are not routinely obtained, thereby reducing their applicability in cases with mixed etiologies.

Comparison with APACHE II, SOFA, and BISAP scores further validated the superior clinical utility of the developed model to predict mixed-etiology SAP. BISAP was developed for all AP and is simple to use; however, its sensitivity is notably low (pooled sensitivity = 0.67 for AP severity prediction [23]). APACHE II comprises over 12 physiological parameters but ignores their synergistic effects [25, 26], leading to cumbersome use and compromised accuracy in mixed subgroups. SOFA focuses on established organ dysfunction rather than mixed-etiology AP-specific early inflammatory-metabolic cascades; moreover, it is affected by inter-rater variability (e.g., neurological assessment) [29]. In contrast, the developed model targets the unaddressed mixed-etiology population, uses only 3 routine laboratory indicators, and has been validated to outperform the aforementioned tools in predictive accuracy and clinical accessibility.

The reduced level of albumin (< 35 g/L), a negative acute-phase protein, dynamically reflects disease severity through three key mechanisms. First, inflammatory mediators (such as TNF-α and IL-6) inhibit hepatic albumin synthesis while facilitating the preferential production of positive acute-phase proteins (such as CRP); moreover, although the fractional synthesis rate of albumin may remain normal or slightly elevated in acute inflammation, its absolute production decreases due to systemic protein metabolic shifts, particularly in severe inflammation or liver dysfunction [46]. Second, capillary permeability is increased by the inflammation-driven upregulation of vascular endothelial growth factor, leading to massive albumin extravasation into the interstitial space, which lowers serum albumin levels linked to inflammatory edema and may disrupt local redox balance [47]; additionally, a hypercatabolic state accelerates albumin degradation, thereby shortening its half-life from 20 days (normal) to 7–10 days [46]. Notably, several studies [34, 48, 49] have shown that severe hypoalbuminemia independently predicts SAP [49], which agrees with the findings of the present study.

Ca²⁺ is a critical factor for the pathogenesis and prognosis of pancreatitis. Ca²⁺ is predominantly localized in the apical zymogen granule region of acinar cells in the pancreas; it activates pancreatic enzymes (triggering tissue self-digestion) following intracellular elevation under pathological conditions, such as biliary obstruction and hypertriglyceridemia, but not under physiological conditions. This abnormal activation of pancreatic enzymes is amplified by Ca²⁺-mediated inhibition of endoplasmic reticulum Ca²⁺-ATPase (blocking Ca²⁺ reuptake and inducing endoplasmic reticulum stress) and disrupted lysosomal-zymogen granule segregation (enabling cathepsin B to activate trypsinogen) [7, 50]; moreover, Ca²⁺ also binds to FFAs from degraded pancreatic fat to form calcium soaps, thereby lowering serum Ca²⁺ levels, a process more prominent in HTG-AP^[51]. As a central pathogenic link [7], Ca²⁺ overload promotes acinar cell necrosis. Hypocalcemia, a well-established predictor of SAP, HTG-AP, and POF in AP patients [33, 34, 51], is actually pancreatic parenchymal damage rather than simple electrolyte imbalance. In the present study, lower serum calcium correlated with higher mixed-etiology SAP risk, as mixed-etiology (e.g., cholelithiasis + HTG) may synergistically exacerbate Ca²⁺ dysregulation, indicating serum calcium as a pivotal predictor for mixed-etiology SAP.

D-dimer, a unique degradation component of cross-linked fibrin generated through plasmin-mediated fibrinolysis, directly reflects activated coagulation-fibrinolysis cascades; it is widely used for thrombosis assessment and serves as a risk factor to predict thrombosis occurrence and AP severity [49, 52–54]. Notably, the reciprocal interaction of coagulation and inflammation is well documented: inflammatory onset can trigger coagulation activation, while coagulation processes can further stimulate inflammation [33, 55]. This loop is exacerbated in mixed-etiology AP, where synergistic insults (e.g., alcohol-induced endothelial injury + hypertriglyceridemia-related hypercoagulability) enhance tissue factor release and platelet activation, accelerating cross-linked fibrin formation and D-dimer elevation. In the present study, mixed-etiology AP patients showing higher D-dimer levels were more likely to progress to SAP, as the elevated D-dimer level may indicate more severe systemic coagulation disturbance and pancreatic necrosis [53, 54]; these pathological changes may be amplified by combined etiologies, which agrees with the observations of previous investigations.

The newly developed model was also compared with the commonly applied scores (SOFA, BISAP, and APACHE II). APACHE II and SOFA [5, 25, 27–30] are frequently employed in clinical studies and show accurate evaluation performance. In the present study, APACHE II and SOFA exhibited high sensitivity (86.8% and 84.9%, respectively) but low specificity (56.5% and 64.5%, respectively), and their operation was complex. BISAP also demonstrated a predictive sensitivity of 71.7%, which aligns with prior research outcomes [22–24, 56]. In contrast, the newly developed model has high sensitivity (83.0%) and specificity (80.6%), which enables to achieve remarkable accuracy; involves fewer required indicators and has a simpler evaluation workflow; and can be assessed simply through blood tests; these advantages make it a simple and convenient tool.

Clinically, the findings of the present study have potential actionable implications. Given the worse prognosis of mixed-etiology AP, the developed model has the potential to assist in early risk stratification in clinical settings by using routine blood markers, which may support targeted early intensive care monitoring for high-risk subgroups (e.g., cholelithiasis + HTG). For patients with cholelithiasis combined with HTG, timely lipid-lowering therapy could potentially disrupt the synergistic pathogenic cycle and reduce the risk of severe outcomes. Additionally, the relatively high accuracy and simplicity of the developed model may help identify patients likely to gain from the early initiation of enteral nutrition, which might further optimize supportive care and potentially improve prognosis.

Study strengths and limitations

This study’s major strength is the systematic evaluation of the clinical prognosis of both single-etiology and mixed-etiology AP and the development of a high-performance early prediction model (AUC = 0.857) based on three readily available biomarkers: ALB, Ca²⁺, and D-D, which specifically targets individuals with mixed-etiology AP, a somewhat ignored population. This model exhibited enhanced predictive accuracy for mixed-etiology SAP relative to conventional scoring systems and demonstrated good calibration and clinical utility through internal validation and DCA.

However, there are several limitations of this study, listed as follows. (1) Study design: This study’s single-center retrospective design introduces inherent selection bias (e.g., specific mixed-etiology distribution and localized protocols) and fails to control unmeasured confounders. Notably, in this study, two key factors may introduce unmeasured confounding effects: first, the lack of standardization in treatment strategies (e.g., fluid resuscitation strategies, anti-inflammatory medication use, and timing of interventional procedures) among the included patients, which may confound the association between mixed etiologies and prognostic outcomes as well as model performance; second, variability in the timing of laboratory tests (e.g., some indicators were measured at 24 h post-admission, while others were estimated at 36–48 h post-admission) due to differences in the implemented protocols. This variability in timing might cause inconsistent reflection of true biomarker levels at a common disease stage, thereby introducing unmeasured confounding effects that could potentially impact the accuracy of the prediction model. Additionally, other unmeasured confounders (e.g., varied in-hospital monitoring frequency) may still exist despite strict exclusion criteria. (2) Sample size: A small sample size in some mixed-etiology subgroups could weaken the reliability of subgroup-specific analyses. (3) Laboratory measurement: The model relies only on static parameters measured within 48 h of admission and does not consider dynamic changes in these parameters; this further limits its ability to capture real-time disease progression. (4) Generalizability and methodological constraints: (a) external validation was not performed, which is a key methodological limitation that hinders the verification of the model’s reproducibility, applicability, and value in broader populations; (b) the exclusion of idiopathic and ERCP-related pancreatitis, although reasonable for focusing on etiological heterogeneity, reduces study population representativeness and limits result generalizability. Despite using multiple imputation methods to address missing data, the relatively high rate of missing BMI values necessitates improvement in future studies; (c) this single-center Chinese cohort differs from Western populations (where alcohol consumption is more prevalent), and mixed-etiology distribution may vary across other ethnic and regional groups. This etiological heterogeneity could alter predictive factor weights and model performance, affecting the model’s applicability in non-Chinese populations. Thus, external validation is urgently required in cohorts with diverse demographics; healthcare settings (e.g., primary hospitals); and across different populations, regions, and ethnicities.

Conclusions

Mixed-etiology AP may have an association with an elevated possibility of SAP, organ failure, and resource utilization. The present study developed a model based on three biomarkers (albumin, serum calcium, and D-D), which exhibits promising predictive performance for SAP in mixed-etiology AP cases (AUC = 0.857). This model may help clinically identify high-risk patients and enable early personalized treatment, thereby reducing severe complication occurrence and easing disease burden. This potential warrants further confirmation through external validation studies.

Future investigations should (1) conduct multicenter prospective studies to validate the model by including cohorts from different regions, ethnicities, and countries to address etiological heterogeneity; (2) calculate minimum sample size using G*Power (expected incidence of severe cases, α = 0.05, 1 - β = 0.8) to improve rigor; (3) integrate dynamic biomarker changes to enhance sensitivity for late-onset SAP; and (4) perform basic studies to clarify synergistic pathogenic mechanisms of mixed etiologies, reinforcing the biological plausibility of the model.

Abbreviations

AP

Acute pancreatitis

LASSO

Least absolute shrinkage and selection operator

SAP

Severe acute pancreatitis

OR

Odds ratio

ICU

Intensive care unit

SIRS

Systemic inflammatory response syndrome

CI

Confidence interval

AUC

Area under the curve

HTG

Hypertriglyceridemia

APACHE II

Acute physiology and chronic health evaluation II

SOFA

Sequential organ failure assessment

BISAP

Bedside index for severity in acute pancreatitis

TyG

Triglyceride and glucose

ERCP

Endoscopic retrograde cholangiopancreatography

CRP

C-reactive protein

PCT

Procalcitonin

ABP

Acute biliary pancreatitis

D-D

D-dimer

TG

Serum triglycerides

BMI

Body mass index

NSAP

Non-severe acute pancreatitis

APFC

Acute peripancreatic fluid collection

IPN

Infected pancreatic necrosis

CRRT

Continuous renal replacement therapy

PPC

Pancreatic pseudo cyst

HR

Heart rate

ANC

Acute necrotic collection

RR

Respiratory rate

WON

Walled-off necrosis

MV

Mechanical ventilation

MAP

Mean arterial pressure

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

WBC

White blood cell

T

Temperature

RBC

Red blood cell

HCT

Hematocrit

HB

Hemoglobin

PLT

Platelet

RDW

Erythrocyte volume distribution width

Neu

Neutrophil

Lym

Lymphocyte

ALT

Alanine aminotransferase

ALP

Alkaline phosphatase

AST

Aspartate transaminase

TBIL

Total bilirubin

BUN

Blood urea nitrogen

GGT

Gamma-glutamyl transferase

FFAs

Free fatty acids

ALB

Albumin

Cr

Serum creatinine

Ca²⁺

Serum calcium

AMY

Amylase

LPS

Lipase

ROC

Receiver operating characteristic

DCA

Decision curve analysis

IQR

Interquartile range

VIF

Variance inflation factor

PPV

Positive predictive value

NPV

Negative predictive value

ROS

Reactive oxygen species

ApoA1

Apolipoprotein A1

Authors’ contributions

All authors contributed substantially to study conceptualization, study design, implementation, data collection, and data analysis and interpretation, either in individual areas or across all these aspects; the authors were also involved in drafting, revising, or critically examining the article; provided final approval to the published version; concurred on the desired journal of submission; and consented to be responsible for all elements of the study.

Funding

This study was financially supported by the IIT Clinical Research Fund of The Second Affiliated Hospital of Xi’an Jiaotong University (Grant No. Z008), the Shaanxi Provincial Natural Science Foundation (2024JC-ZDXM-49), and the Shaanxi Provincial Natural Science Basic Research Program (2024JC-ZDXM-49).

Data availability

All relevant study data and datasets generated or analyzed as part of this work will be made available to interested researchers upon submitting a valid request.

Declarations

Ethics approval and consent to participate

The First Affiliated Hospital of Xi’an Jiaotong University Review Board approved this study (XJTUAF2025LSYY-654). Informed consent requirement was waived. The study complied with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.