Blood urea nitrogen to albumin ratio as a robust predictor of in-hospital mortality in patients with predicted severe acute pancreatitis: a retrospective multicenter observational study

Abstract

Background:

Despite the established role of blood urea nitrogen-to-albumin ratio (BAR) in critical care, its prognostic value in acute pancreatitis remains unvalidated. This multicenter study assessed BAR’s accuracy as an admission biomarker for predicting in-hospital mortality in predicted severe acute pancreatitis (SAP) cases.

Methods:

This retrospective study enrolled 5384 patients from the LOCAL cohort and 494 patients from a multicenter double-blind randomized controlled trial (TRACE cohort), all of whom were predicted to have SAP (APACHE II ≥8) upon admission. Cox regression models were employed in two independent cohorts to explore the association between BAR and the risk of in-hospital mortality in subjects, and a restricted cubic spline regression was further constructed. The receiver operating characteristic (ROC) curve was drawn, and the area under the curve was determined to evaluate the predictive capacity of BAR, blood urea nitrogen (BUN), albumin (ALB) and traditional scoring systems (APACHE II, SIRS, and BISAP) for in-hospital mortality. Time-dependent ROC analysis was also performed to assess the predictive performance of BUN and BAR at multiple time points.

Results:

In the LOCAL and TRACE cohorts, 320 (5.94%) and 39 (7.89%) patients died during hospitalization, respectively. Multivariable Cox regression models showed a significant positive association between BAR and the risk of in-hospital mortality [HR: LOCAL 1.23 (1.16–1.31); TRACE 1.33 (1.01–1.76)], while the restricted cubic spline analysis suggested a potential nonlinear relationship (P for nonlinearity < 0.001). In prognostic prediction, BAR demonstrated significantly better performance than traditional scoring systems, BUN and ALB, and showed high accuracy in predicting outcomes for patients with biliary acute pancreatitis (AP) in both cohorts (area under the curve: LOCAL 0.8696; TRACE 0.8633). Additionally, time-dependent ROC analysis revealed that BAR demonstrated superior accuracy and stability in predicting mortality risk at 3, 5, 7, 9, 14, 28, 60, and 90 days, compared to BUN alone.

Conclusion:

This study is the first to demonstrate that BAR significantly improves the predictive accuracy of BUN and ALB for in-hospital mortality in predicted SAP patients, with particular effectiveness in predicting outcomes for biliary AP patients.

Background

Acute pancreatitis (AP) arises from premature intrapancreatic enzyme activation, triggering a cascade of autodigestive processes and systemic inflammation. Common causes include gallstone disease, excessive alcohol consumption, and hypertriglyceridemia^[1–4]. AP’s clinical trajectory spans from self-limiting inflammation to lethal systemic failure, with mortality directly correlating with organ dysfunction duration^[5]. Epidemiological studies report a global incidence of AP of approximately 33.74 per 100 000 person-years (95% CI 23.22–48.81). Among these cases, 20% to 40% progress to severe acute pancreatitis (SAP), with an associated in-hospital mortality rate of approximately 15%^[6]. According to the revised Atlanta classification^[7], the majority of patients with mild AP (MAP) recover fully within days to a week with conservative treatment. In contrast, patients with SAP typically have a poor prognosis, characterized by persistent failure of one or more organs, including the kidneys, liver, respiratory system, and circulatory system^[8–10].

Urea nitrogen, derived from the metabolic breakdown of proteins, is synthesized in the liver and eliminated via the kidneys^[11,12]. Historically, blood urea nitrogen (BUN) levels have served as a key indicator of renal metabolic activity. However, recent studies have found that BUN plays a significant role in predicting outcomes in AP patients^[13,14]. Elevated BUN levels may indicate pathological states such as hypovolemia, renal injury, gastrointestinal bleeding, or prerenal azotemia in AP patients and are significantly associated with increased in-hospital mortality^[15–17]. Serum albumin (ALB), an important protein synthesized solely by the liver, reflects both liver function and nutritional status, and has been shown to have prognostic significance in AP patients^[18–20]. Recently, researchers innovatively combined BUN and ALB into a single metric—the blood urea nitrogen-to-albumin ratio (BAR)—and demonstrated its significant prognostic value in critically ill hospitalized patients. BAR has been shown to predict in-hospital mortality in patients with diabetic ketoacidosis, coronary artery disease in intensive care units, and all-cause mortality in those with acute kidney injury^[21–25]. However, the clinical value of BAR in prognostic assessment for SAP remains unclear, particularly in patients initially predicted with SAP upon admission, where timely and accurate risk stratification is essential. Therefore, this study aims to analyze clinical data from the LOCAL cohort and the TRACE cohort to evaluate the predictive ability of BAR for short-term prognosis in patients with predicted SAP at admission, providing a new practical tool for risk assessment and stratification in AP patients. This cohort study has been reported in line with the strengthening the reporting of cohort, cross-sectional and case control studies in surgery (STROCSS) guidelines^[26].

Methods

Study design and data sources

This nationwide, multicenter retrospective cohort study examines clinical data from patients initially predicted with SAP at admission to evaluate the BAR’s efficacy in predicting short-term outcomes in SAP cases. The primary data for this study were derived from the pancreatitis cohort of the Department of Gastroenterology, The First Affiliated Hospital of Nanchang University (the LOCAL cohort), as well as from a nationwide multicenter double-blind randomized controlled trial (TRACE trial)^[27]. The LOCAL cohort, established in 2005 and continuously updated and maintained by trained medical personnel, includes hospitalized patients diagnosed with AP in our institution from 2005 to the present^[28]. The cohort gathers demographic data, current and past medical history, primary diagnoses, laboratory results, in-hospital treatments, and discharge outcomes. For this study, we retrospectively included all hospitalized patients diagnosed with AP in the LOCAL cohort (n = 14 650). According to the study’s objectives, we excluded patients with an APACHE II score < 8 at admission (n = 7782), patients aged under 18 or over 75 years (n = 1076), pregnant patients (n = 87), those with end-stage liver or kidney disease (n = 32), patients with missing BAR or outcome data (n = 283), and those with extreme baseline variables (n = 6). Ultimately, 5384 eligible participants were included. A comprehensive flowchart illustrating the inclusion and exclusion criteria was presented in Figure 1. Additionally, the inclusion and exclusion procedures for the TRACE cohort have been described in detail elsewhere^[27]. Based on that, this study further excluded 14 participants with missing baseline BAR data, ultimately including 494 participants.

Figure 1.

Flow chart of the LOCAL cohort study subjects.

Baseline data collection and measurements

The methods for collecting and measuring clinical baseline indicators in participants of the TRACE cohort have been detailed elsewhere^[27]. For the LOCAL cohort, we collected demographic (age, sex) and anthropometric data (height, weight), lifestyle habits (smoking status, drinking status), past medical history (liver disease, chronic renal insufficiency, diabetes mellitus), and clinical and laboratory data measured within 24 hours of admission [type of AP, temperature, pulse, respirations, systolic blood pressure (SBP), total bilirubin (TBIL), white blood cell count (WBC), neutrophils (NEU), platelets (PLT), aspartate aminotransferase (AST), direct bilirubin (DBIL), ALB, triglycerides (TG), diastolic blood pressure (DBP), total cholesterol (TC), BUN, alanine aminotransferase (ALT), APACHE II score, BISAP score, and SIRS score]. In addition, data on in-hospital therapeutic interventions were collected, including continuous renal replacement therapy (CRRT), low-molecular-weight heparin (LMWH) use, albumin supplementation, mechanical ventilation (MV), and volume of fluid resuscitation (VFR). Baseline demographic and anthropometric data, as well as lifestyle habits and past medical history, were obtained by professional clinicians at admission through physical measurements or face-to-face interviews. Laboratory data were collected after patients fasted for a minimum of 8 hours, with peripheral venous blood samples taken by clinical nurses and analyzed in the hospital’s standardized laboratory using an automated hematology analyzer (COULTER LH 780/LH 785, USA) and an automated biochemistry analyzer (Beckman Coulter AU5800, USA).

Handling of missing data

We applied the fully conditional specification method to impute missing data according to the proportion of missingness. For variables with relatively high missing rates—height (17.9%), weight (15.0%), total bilirubin (TBIL, 4.5%), and total cholesterol (TC, 4.3%)—multiple imputation was performed to generate five complete datasets for pooled analysis. Additionally, two alternative imputation methods, median imputation and K-nearest neighbors (KNN) imputation, were applied to generate two separate complete datasets for validation purposes^[29–31] (see Supplementary Digital Content Table S1, available at: http://links.lww.com/JS9/F195). Notably, the primary analysis was conducted with the existing minimal missing data, while the imputed datasets were used for sensitivity analyses to validate the robustness of the primary findings.

Diagnosis of pancreatitis and outcome events

Based on the 2012 Revised Atlanta Classification, AP is diagnosed when at least two of the following three criteria are present: (1) characteristic symptoms, including acute upper abdominal pain; (2) serum amylase or lipase levels exceeding three times the normal upper limit; (3) imaging findings indicative of AP, such as those observed on CT or MRI^[7]. According to the American College of Gastroenterology guideline on the management of acute pancreatitis, patients with an APACHE II score greater than eight are defined as having predicted SAP^[32]. The primary endpoint of this study was in-hospital mortality among AP patients. The follow-up period spanned from hospital admission to either discharge or in-hospital death, whichever occurred first.

Statistical analysis

Initially, in both the LOCAL and TRACE cohorts, participants were categorized based on survival status and BAR tertiles. Baseline characteristics were described as mean (standard deviation), median (interquartile range), or n (%) depending on the type and distribution of variables. Subsequently, the t-test or Mann–Whitney U test was applied to compare the baseline differences of continuous variables between the death and survival groups. For the differences among the BAR tertile groups, one-way analysis of variance or the Kruskal–Wallis H test was utilized. Regarding categorical variables, the chi-square test was employed to assess the differences among groups.

Kaplan–Meier survival curves were generated based on BAR tertiles to depict survival in both cohorts, and the log-rank test was used for group comparisons. Before fitting Cox models, the proportional hazards assumption was tested using Schoenfeld residuals, and multicollinearity among covariates was assessed by multiple linear regression. Variables with a variance inflation factor (VIF) > 5 were considered collinear^[33,34]. In the Cox regression models, BAR was incorporated as both a continuous and a categorical variable (categorized by tertiles). The final model was adjusted for multiple factors, including age, sex, height, weight, SBP, body temperature, pulse, respiration, etiology of AP, APACHE II, WBC, HCT, TBIL, TC, and TG. To further explore the relationship between BAR and the risk of in-hospital mortality in predicted AP patients, we used a restricted cubic spline (RCS) model with four knots based on the fully adjusted model to visualize the shape of the association, and employed a segmented Cox regression model to automatically identify the optimal inflection points on the curve. In addition, exploratory subgroup analyses were conducted based on sex, age (≤ 60, > 60 years), BMI (<24, ≥ 24 kg/m2), smoking status, drinking status, history of hyperlipidemia, diabetes, and etiology of AP. The log-likelihood ratio test was used to assess the effect modification by each subgroup variable. Moreover, to explore treatment heterogeneity across different BAR tertile groups, we assessed the association between VFR and in-hospital mortality within each BAR subgroup^[35]. First, the Schoenfeld residual test was used to verify whether VFR satisfied the proportional hazards assumption. Then, RCS models were constructed to analyze the potential nonlinear relationship between VFR and mortality risk. Finally, a three-dimensional fitted surface plot was generated to visualize the interaction between BAR and VFR in relation to mortality risk.

Receiver operating characteristic (ROC) curves were generated. To evaluate and compare the predictive abilities of BAR, its components BUN and ALB, as well as traditional severity scores for pancreatitis—APACHE II, SIRS, and BISAP—for in-hospital mortality, we calculated the area under the curve (AUC), sensitivity, specificity, and optimal cutoff values. The DeLong test was used to compare the differences in predictive accuracy between these indices. Subsequently, in the LOCAL cohort, the predictive accuracy and stability of BAR and BUN for death events at different time points were further compared through time-dependent ROC analysis. Additionally, subgroup ROC analyses were conducted to investigate the predictive value of BAR for in-hospital mortality in patients with different etiologies of AP, including hypertriglyceridemia-associated AP (HTG-AP), alcoholic AP (AAP), biliary AP (BAP), and other causes of AP.

Several sensitivity analyses were carried out to confirm the reliability of our findings: (1) Due to missing values in some variables of the LOCAL cohort, we applied multiple imputation, median imputation, and KNN imputation to handle the missing data. Subsequently, we repeated the primary analyses on the imputed datasets. (2) To avoid the influence of ultra-acute disease progression, patients with a hospital stay of 24 hours or less were excluded. (3) Based on the fully adjusted model, additional adjustments were made for CRRT, MV, albumin supplementation, LMWH, and VFR volume in the LOCAL cohort. (4) E-values were calculated in fully adjusted models of the LOCAL and TRACE cohorts to assess the potential influence of unmeasured confounders on the BAR-mortality association.

In this study, all statistical analyses were carried out using R software (version 4.3.1), Free Statistics (version 2.1), and Empower (R) (version 4.1). Statistical significance was determined at a two-sided P value < 0.05.

Results

Baseline characteristics and survival rates of subjects

In the LOCAL cohort, involving 5384 individuals, the median hospital stay duration was 9 days. During this period, 320 participants (5.94%) passed away. As for the TRACE cohort, the median length of hospital stay was 15 days, and 39 participants (7.89%) died. Table 1 presents baseline characteristics of both cohorts, grouped by survival status. Overall, the two cohort populations exhibited similar characteristics. Compared to survivors, deceased participants had significantly higher levels of BAR, TBIL, Cr, CRP, APACHE II, and BUN, as well as lower levels of SBP, DBP, TC, PLT, and HCT. Additionally, deceased participants in the LOCAL cohort had higher age, body temperature, pulse, respiratory rate, TG, WBC, and NEU levels, and lower ALB levels. Table 2 summarizes clinical outcomes and complications by BAR tertiles. As BAR tertiles increased, the incidence of ICU stays, hospital stays, organ failure, infected pancreatic necrosis, local complications, and mortality rose significantly in both cohorts. Mortality rates across BAR tertiles were 0.7%, 2.1%, and 15.0% in the LOCAL cohort, and 1.2%, 3.0%, and 19.4% in the TRACE cohort.

Table 1.

Baseline characteristics of subjects grouped according to follow-up outcomes

	LOCAL cohort			TRACE cohort
Variables	Survival (n = 5064)	Death (n = 320)	P	Survival (n = 455)	Death (n = 39)	P
Female	2099 (41.4)	131 (40.9)	0.857	170 (37.4)	10 (25.6)	0.144
Age, year	52.4 ± 14.0	54.2 ± 13.6	0.025	44.5 ± 13.2	46.9 ± 15.4	0.286
Height, cm	164.4 ± 7.9	164.3 ± 7.7	0.867	167.0 ± 8.0	167.8 ± 6.8	0.522
Weight, kg	65.4 ± 13.8	67.0 ± 11.9	0.093	74.6 ± 14.4	75.4 ± 13.6	0.758
Temperature, °C	37.0 ± 0.7	37.4 ± 1.0	<0.001	37.2 ± 0.8	37.2 ± 0.9	0.988
Pulse, times/min	94.7 ± 20.8	114.4 ± 22.4	<0.001	108.3 ± 23.0	112.1 ± 29.7	0.337
Respirations, times/min	21.8 ± 5.2	27.7 ± 8.0	<0.001	23.8 ± 6.4	23.8 ± 6.2	0.948
SBP, mmHg	131.3 ± 24.7	125.5 ± 27.9	<0.001	130.7 ± 21.6	121.8 ± 25.7	0.016
DBP, mmHg	83.5 ± 33.4	77.5 ± 18.6	0.002	80.2 ± 15.2	73.6 ± 17.3	0.011
TBIL, umol/L	17.9 (11.4, 30.1)	24.0 (13.3, 44.4)	<0.001	19.3 (13.2, 30.1)	28.4 (23.4, 69.7)	<0.001
Cr, umol/L	65.0 (50.3, 83.5)	181.6 (91.9, 322.2)	<0.001	64.0 (51.0, 97.3)	197.0 (110.0, 352.0)	<0.001
CRP, mg/L	118.0 (29.0, 210.0)	213.6 (145.5, 320)	<0.001	169.6 ± 102.5	207.1 ± 114.1	0.03
ALB, g/L	36.7 ± 6.3	31.0 ± 5.6	<0.001	31.5 ± 5.3	30.9 ± 6.4	0.451
TC, mmol/L	5.4 ± 3.5	4.9 ± 4.4	0.026	4.4 (3.1, 6.5)	3.1 (2.6, 4.9)	0.001
TG, mmol/L	1.7 (1.0, 4.9)	2.2 (1.3, 4.9)	<0.001	3.9 (1.5, 7.8)	3.7 (1.4, 5.1)	0.235
APACHEII	11.1 ± 2.9	15.6 ± 5.1	<0.001	11.0 ± 4.3	14.8 ± 6.1	<0.001
BUN, mmol/L	5.0 (3.7, 6.9)	12.7 (7.4, 18.9)	<0.001	2.1 (1.4, 3.3)	4.3 (3.3, 6.6)	<0.001
WBC, 109/L	12.6 ± 5.8	15.0 ± 7.4	<0.001	11.6 ± 5.0	11.8 ± 4.5	0.722
NEU, 109/L	10.8 ± 5.5	13.0 ± 7.0	<0.001	9.9 ± 4.7	10.6 ± 4.7	0.332
PLT, × 109/L	208.7 ± 91.8	179.2 ± 95.6	<0.001	174.4 ± 71.7	130.2 ± 54.2	<0.001
LYM, 109/L	1.1 (0.7, 1.5)	0.9 (0.6, 1.4)	0.056	1.0 ± 0.5	0.9 ± 0.5	0.221
HbA1c, %	7.0 ± 2.3	6.7 ± 1.7	0.477	7.4 ± 2.9	7.3 ± 1.6	0.86
HCT, %	39.0 ± 8.4	38.0 ± 11.3	0.042	36.3 ± 8.2	32.6 ± 10.3	0.008
Etiology, n (%)			0.696			0.811
Biliary	2485 (49.1)	165 (51.6)		176 (38.7)	18 (46.2)
HTG	1585 (31.3)	91 (28.4)		227 (49.9)	17 (43.6)
Alcoholic	282 (5.6)	20 (6.2)		30 (6.6)	2 (5.1)
Others	712 (14.1)	44 (13.8)		22 (4.8)	2 (5.1)

ALB, Albumin; APACHEII, Acute Physiology and Chronic Health Evaluation II; BUN, Blood Urea Nitrogen; Cr, Creatinine; SBP, Systolic Blood Pressure; CRP, C-Reactive Protein; DBP, Diastolic Blood Pressure; HbA1c, Glycated Hemoglobin A1c; HCT, Hematocrit; HTG, Hypertriglyceridemia; LYM, Lymphocyte; NEU, Neutrophil; PLT, Platelet; TBIL, Total Bilirubin; TC, Total Cholesterol; TG, Triglyceride; WBC: White Blood Cell.

Table 2.

Clinical outcomes of subjects grouped according to BAR tertiles in two cohorts

LOCAL cohort	Tertile 1 (n = 1795)	Tertile 2 (n = 1794)	Tertile 3 (n = 1795)	P
LOS in hospital (days)	8.0 (6.0, 12.0)	9.0 (6.0, 14.0)	13.0 (8.0, 24.0)	<0.001
Pancreatic local complications, n (%)				<0.001
No	885 (49.3)	738 (41.1)	445 (24.8)
Yes	910 (50.7)	1056 (58.9)	1350 (75.2)
IPN, n (%)				<0.001
No	1731 (96.4)	1685 (93.9)	1533 (85.4)
Yes	64 (3.6)	109 (6.1)	262 (14.6)
Respiratory failure, n (%)				<0.001
No	1556 (86.7)	1392 (77.6)	894 (49.8)
Yes	239 (13.3)	402 (22.4)	901 (50.2)
Renal failure, n (%)				<0.001
No	1778 (99.1)	1754 (97.8)	1348 (75.1)
Yes	17 (0.9)	40 (2.2)	447 (24.9)
Circulatory failure, n (%)				<0.001
No	1772 (98.7)	1746 (97.3)	1518 (84.6)
Yes	23 (1.3)	48 (2.7)	277 (15.4)
Persistent MOF, n (%)				<0.001
No	1772 (98.7)	1739 (96.9)	1348 (75.1)
Yes	23 (1.3)	55 (3.1)	447 (24.9)
Death, n (%)				<0.001
No	1782 (99.3)	1757 (97.9)	1525 (85)
Yes	13 (0.7)	37 (2.1)	270 (15)
TRACE cohort	Tertile 1 (n= 165)	Tertile 2 (n= 164)	Tertile 3 (n= 165)
LOS in ICU (days)	7.0 (3.0, 11.0)	8.0 (5.0, 15.0)	15.0 (7.0, 25.0)	<0.001
LOS in hospital (days)	12.0 (7.0, 18.0)	14.0 (8.8, 22.0)	20.0 (13.0, 33.0)	<0.001
New-onset persistent OF, n (%)				<0.001
No	144 (87.3)	147 (89.6)	118 (71.5)
Yes	21 (12.7)	17 (10.4)	47 (28.5)
Respiratory failure, n (%)				0.467
No	146 (88.5)	151 (92.1)	146 (88.5)
Yes	19 (11.5)	13 (7.9)	19 (11.5)
Renal failure, n (%)				<0.001
No	164 (99.4)	162 (98.8)	151 (91.5)
Yes	1 (0.6)	2 (1.2)	14 (8.5)
Circulatory failure, n (%)				<0.001
No	162 (98.2)	160 (97.6)	140 (84.8)
Yes	3 (1.8)	4 (2.4)	25 (15.2)
Death, n (%)				<0.001
No	163 (98.8)	159 (97)	133 (80.6)
Yes	2 (1.2)	5 (3)	32 (19.4)
IPN, n (%)				<0.001
No	154 (93.3)	146 (89)	111 (67.3)
Yes	11 (6.7)	18 (11)	54 (32.7)

BAR, Blood urea nitrogen to albumin ratio; LOS, Length of Stay; LOS in ICU, Length of Stay in Intensive Care Unit; MOF, Multiple Organ Failure; OF, Organ Failure; IPN, Infected Pancreatic Necrosis.

The Kaplan–Meier curves, shown in Figure 2 and Supplementary Digital Content Figure S1, available at: http://links.lww.com/JS9/F194, depict the survival rates of participants grouped by different BAR tertiles in both cohorts. The curves indicate that, in comparison to participants with lower BAR levels, those with higher BAR levels faced a significantly elevated mortality risk (All Log-rank test P < 0.05).

Figure 2.

The Kaplan–Meier survival curves of subjects in LOCAL cohort grouped by BAR tertiles. BAR, blood urea nitrogen to albumin ratio.

Association of BAR with in-hospital mortality risk and treatment heterogeneity in patients with predicted SAP

Supplementary Digital Content Tables S2 and S3, available at: http://links.lww.com/JS9/F195 show the collinearity analysis for the LOCAL and TRACE cohorts. In both, BUN had a VIF > 5, indicating collinearity and was therefore excluded from further models. Schoenfeld residuals confirmed that BAR met the proportional hazards assumption in both cohorts (all P > 0.05; Supplementary Digital Content Figures S2 and S3, available at: http://links.lww.com/JS9/F194). In both cohorts, BAR was significantly associated with increased mortality risk in the unadjusted model and remained an independent predictor after full adjustment. Each one-standard-deviation increase in BAR was linked to a 23% (HR 1.23, 95% CI 1.16–1.31, P < 0.0001) and 33% (HR 1.33, 95% CI 1.01–1.76, P = 0.041) higher risk in the LOCAL and TRACE cohorts, respectively. Participants in Tertile 3 had a 4.26-fold and 4.95-fold higher mortality risk compared to Tertile 1 in the respective cohorts (Table 3). Using the fully adjusted model, the restricted cubic spline (RCS) curve provided a more detailed visualization of the relationship between BAR and the mortality risk within the LOCAL cohort, as depicted in Figure 3. The results indicated a nonlinear association between BAR and mortality risk (P for nonlinearity < 0.001). While the overall trend showed a positive correlation, a clear inflection point was observed on the curve, where the relationship between BAR and mortality risk changed significantly on either side of the point. Using the segmented Cox regression model, we identified the inflection point on the RCS curve at BAR = 0.44, with a 95% confidence interval of 0.31–0.5, as determined via bootstrap (Table 4). Notably, before BAR reached 0.44, the HR for mortality risk per unit increase in BAR was 371.52 (91.55, 1507.70; P < 0.0001), but after BAR exceeded 0.44, the association with mortality risk was no longer significant, suggesting that 0.44 may represent the saturation point for the relationship between BAR and mortality risk.

Table 3.

Association between admission BAR and in-hospital death in the LOCAL and TRACE cohort

	HR (95% CI)
	LOCAL cohort		TRACE cohort
Variable	Crude model	Adjusted model	Crude model	Adjusted model
BAR (Per SD)	1.28 (1.24, 1.33) < 0.001	1.23 (1.16, 1.31) < 0.001	1.44 (1.16, 1.78) 0.001	1.33 (1.01, 1.76) 0.041
BAR Tertiles
Tertile 1	1(Ref)	1(Ref)	1(Ref)	1(Ref)
Tertile 2	2.18 (1.15, 4.11) < 0.001	2.04 (0.91, 4.55) 0.082	1.57 (0.3, 8.16)	1.17 (0.21, 6.54)
			0.589	0.858
Tertile 3	10.21 (5.82, 17.91) < 0.001	5.26 (2.51, 11.01) < 0.001	6.43 (1.52, 27.24)	5.95 (1.31, 26.97)
			0.011	0.021

BAR, Blood urea nitrogen to albumin ratio; HR, Hazard ratio; SD, Standard Deviation; others as in Table 1.

Crede Model was adjusted for: None.

In the Adjusted model, both cohorts have jointly adjusted Age, Sex, Height, Weight, SBP, Body Temperature, Pulse, Respiration, Etiology of AP, APACHE II, WBC, HCT, TBIL, TC, and TG.

Figure 3.

Visualizing the relationship between BAR and the in-hospital mortality risk of subjects in the LOCAL cohort using a 4-knots RCS. BAR, blood urea nitrogen to albumin ratio; RCS, restricted cubic spline.

Table 4.

Piecewise Cox regression examining thresholds for BAR-related death risk in LOCAL cohort

	Death (HR, 95% CI)
	BAR	P
Fitting model by multivariate Cox regression
	3.03 (2.17, 4.21)	<0.0001
Fitting model by two-piecewise Cox regression
The best inflection point	0.44 (0.31, 0.5)
< inflection point	371.52 (91.55, 1507.70)	<0.0001
> inflection point	0.92 (0.47, 1.78)	0.8002
Log-likelihood ratio test		<0.001

HR, hazard ratios; CI, confidence interval; other abbreviations as in Table 1.

Moreover, we observed significant treatment heterogeneity in the effect of fluid resuscitation across different BAR tertile groups. In patients with lower BAR levels (Tertile 1), VFR alone had little impact on in-hospital mortality. However, in the high BAR group (Tertile 2 and Tertile 3), elevated VFR was significantly associated with increased mortality risk. The three-dimensional surface plots further illustrated the interaction between BAR and VFR on mortality risk across tertiles. Specifically, a relatively safe threshold of VFR appeared to exist in the Tertile 1 group, whereas in Tertile 2 and Tertile 3, the combined elevation of BAR and VFR was associated with a markedly increased risk of death (Supplementary Digital Content Figure S4, available at: http://links.lww.com/JS9/F194).

Predictive value of BAR for in-hospital mortality in patients with predicted SAP

Figure 4 presents the ROC curves for BAR and its components, ALB and BUN, in predicting in-hospital mortality in the LOCAL cohort. The calculated AUC values, sensitivity, specificity, optimal threshold points, and Youden’s Index were summarized in Table 5. The results indicated that the AUC values for BAR, BUN, and ALB were 0.8572, 0.8291, and 0.7624, respectively. Following the application of the Delong test, it was evident that BAR exhibited significantly greater predictive accuracy than its individual components of BUN and ALB (All P < 0.05). Moreover, BAR outperformed established severity scoring systems for acute pancreatitis, including APACHE II, SIRS, and BISAP (Supplementary Digital Content Figure S5, available at: http://links.lww.com/JS9/F194 and Supplementary Digital Content Table S4, available at: http://links.lww.com/JS9/F195). In the TRACE cohort, we further confirmed that BAR had high predictive accuracy for future mortality events, particularly in the treatment subgroup, where the predictive accuracy reached 85.86% (Supplementary Digital Content Table S5, available at: http://links.lww.com/JS9/F195). Furthermore, we conducted ROC analysis in different subgroups of AP etiology within both cohorts to explore the patient group for which BAR would serve as the most appropriate predictor of in-hospital mortality (Supplementary Digital Content Figure S6, available at: http://links.lww.com/JS9/F194). The results indicated that, both in the LOCAL and TRACE cohorts, BAR demonstrated exceptionally high and stable predictive accuracy for in-hospital mortality in patients with BAP, with AUC values of 0.8696 and 0.8633, respectively (Supplementary Digital Content Tables S6 and S7, available at: http://links.lww.com/JS9/F195). Additionally, Table 6 and Figure 5 illustrate the predictive performance of BAR and BUN for outcome events at various time points. The results indicated that, in predicting mortality risk at 3, 5, 7, 9, 14, 28, 60, and 90 days, BAR exhibited superior accuracy and stability compared to BUN alone.

Figure 4.

The ROC curves for predicting in-hospital mortality events in the LOCAL cohort subjects using BAR, ALB, and BUN. BAR, blood urea nitrogen to albumin ratio; BUN, blood urea nitrogen; ALB, albumin; ROC, receiver operating characteristic.

Table 5.

Comparison of BAR and its components, BUN and ALB, in predicting in-hospital mortality of subjects

Test	AUC	95% CI low	95% CI up	Best threshold	Specificity	Sensitivity	Youden’s Index
BAR	0.8572	0.8355	0.8789	0.2398	0.8468	0.7312	0.578
BUN	0.8291a	0.8031	0.8550	7.1200	0.7737	0.7719	0.5456
ALB	0.7624a	0.7364	0.7884	33.9500	0.6556	0.7531	0.4087

AUC, Area under the curve; others as in Table 1.

^aAUC Compared with BAR, DeLong test P < 0.05

Table 6.

Time-dependent ROC analysis of BAR and BUN in predicting in-hospital mortality at different time points in subjects

Variable	Time point (days)	AUC (%)	OCP	SE (%)	SP (%)	Survival (%)
BAR
	3	77.70	0.19394	72.68	72.18	98.69
	5	75.41	0.19241	70.18	71.81	98.33
	7	75.41	0.19023	70.62	71.55	97.92
	9	75.57	0.19394	70.65	72.37	97.35
	14	74.23	0.19413	67.74	72.73	95.23
	28	68.38	0.17507	59.82	68.19	87.63
	60	68.96	0.12392	83.32	47.62	69.63
	90	65.74	0.10976	92.45	42.48	60.68
BUN
	3	76.85	6.03	77.88	63.51	98.69
	5	73.97	6.7	67.60	71.14	98.33
	7	74.04	6.6	68.68	70.53	97.92
	9	74.21	6.45	70.60	68.48	97.35
	14	71.03	6.4	66.54	68.66	95.23
	28	64.48	6.9	47.28	74.22	87.63
	60	58.72	6.9	38.04	75.83	69.63
	90	52.58	3	96.23	18.50	60.68

AUC, Area under the curve; OCP, Optimal Cutoff Point; SE, Sensitivity; SP, Specificity; others as in Table 1.

Figure 5.

Time-dependent ROC analysis of BUN and BAR in the LOCAL cohort participants. BAR, blood urea nitrogen to albumin ratio; BUN, blood urea nitrogen; ROC, receiver operating characteristic.

Subgroup analysis

To further confirm the consistency of the relationship between BAR and the risk of in-hospital mortality, and to investigate potential subgroups worthy of attention, we conducted association analyses across various subgroups based on sex, age (≤ 60, > 60 years), smoking status, drinking status, BMI (< 24, ≥ 24 kg/m2), history of hyperlipidemia, history of diabetes, and AP etiology (Fig. 6). The results indicated that the association between BAR and mortality risk varied significantly across age, BMI, and smoking/alcohol consumption subgroups (All P for interaction < 0.05), with participants over 60 years old, those with smoking and alcohol habits, and those with a BMI < 24 kg/m2 showing a higher BAR-associated mortality risk.

Figure 6.

Stratified analysis based on sex, age, BMI, smoking and drinking status, history of hyperlipidemia, history of diabetes, and AP etiology grouping. BMI, body mass index; AP, acute pancreatitis.

Sensitivity analysis

To assess the robustness of the primary findings, we conducted several sensitivity analyses. In the LOCAL cohort, we applied multiple imputation methods to address missing data, excluded patients with a hospital stay of less than 24 hours, and further adjusted for treatment interventions on top of the fully adjusted model. Across all analyses, the association between BAR and in-hospital mortality remained stable, with HRs consistently above 1.2, closely aligning with the main results (Supplementary Digital Content Tables S8–10, available at: http://links.lww.com/JS9/F195). Additionally, to evaluate the potential impact of unmeasured confounding, we calculated E-values. The results showed that for each standard deviation increase in BAR, the adjusted HRs corresponded to E-values of 1.76 and 1.99 in the LOCAL and TRACE cohorts, respectively. When BAR was analyzed as a categorical variable, the E-values for Tertile 3 were 9.99 and 11.38, respectively (Supplementary Digital Content Table S11, available at: http://links.lww.com/JS9/F195 and Supplementary Digital Content Figure S7, available at: http://links.lww.com/JS9/F194). These relatively high E-values suggest that only a confounder of considerable strength could fully account for the observed associations, supporting the overall robustness of our findings.

Discussion

This nationwide multicenter retrospective cohort study is the first to evaluate the prognostic value of BAR, a composite biomarker derived from BUN and ALB, in predicting in-hospital mortality in patients with predicted SAP. The findings demonstrated that there was a significant association between a higher BAR value and an elevated risk of in-hospital mortality, with a nonlinear relationship and a risk saturation point at a BAR of 0.44. In terms of prognostic performance for in-hospital mortality in patients with predicted SAP, BAR demonstrated significantly higher predictive accuracy (AUC = 0.8572) compared to BUN and ALB alone. Additionally, time-dependent ROC analysis demonstrated that BAR maintained high predictive accuracy and stability for mortality events occurring on the 3rd, 5th, 7th, 9th, 14th, 28th, 60th, and 90th days of hospitalization. Moreover, within both the LOCAL and TRACE cohorts, BAR consistently demonstrated a high level of predictive accuracy for in-hospital mortality in patients with BAP. Specifically, the AUC values were 0.8696 for the LOCAL cohort and 0.8633 for the TRACE cohort.

It is well-known that BUN is an important marker reflecting renal function and nitrogen metabolism^[11,36]. In AP, substantial fluid loss and hypovolemia often lead to renal hypoperfusion and decreased glomerular filtration, resulting in elevated BUN levels^[34,37]. Additionally, systemic inflammatory response syndrome (SIRS) further aggravates renal dysfunction, making elevated BUN a reflection of both impaired renal function and systemic inflammation^[7]. This can lead to azotemia, metabolic waste accumulation, and disturbances in acid-base balance, all of which contribute to increased disease severity and mortality risk^[1,38]. These pathophysiological changes significantly increase the likelihood of disease severity and in-hospital death among patients. Previous studies have shown that higher BUN levels upon admission are notably linked to extended stays in the ICU and a greater risk of mortality^[39–41]. Moreover, Wang et al discovered that the pattern of changes in BUN levels in critically ill patients with AP could be used to distinguish those with unfavorable prognoses. Specifically, patients who exhibited a trajectory characterized as “moderate azotemia with a rapid increase” had a notably higher risk of mortality compared to those with severe azotemia^[17].

ALB, the most abundant plasma protein, plays critical roles in maintaining oncotic pressure and regulating immune function^[42]. In the early stages of AP, SIRS can increase capillary permeability, causing large amounts of ALB to leak into the extravascular space, thereby reducing plasma ALB concentration^[3,43]. Additionally, inflammatory mediators released during AP, such as interleukin-6 (IL-6), can inhibit hepatic synthesis of ALB, while the systemic inflammatory and hypercatabolic state further accelerates ALB degradation^[44,45]. Therefore, low ALB levels not only indicate poor nutritional status but also suggest the presence of severe systemic inflammation and multi-organ dysfunction. Extensive research has shown that low ALB levels at admission are independently associated with an increased risk of persistent organ failure and mortality in AP patients, and may help predict the severity of AP^{[18,19,46–48]}.

BAR is a newly proposed prognostic indicator derived from the simple clinical parameters BUN and ALB. It integrates the risk assessment and predictive capabilities of two important biochemical markers and has played a significant role in evaluating and predicting the risk of in-hospital mortality in patients with acute critical illnesses^{[21–23,49,50]}. Cai et al demonstrated in a study of 13 464 critically ill septic patients that higher BAR levels were significantly associated with increased in-hospital mortality^[49]. Zhang et al further proposed BAR as a novel predictor of mortality in patients in the coronary care unit, with AUC values of 0.671, 0.673, and 0.685 for predicting in-hospital mortality, 28-day postdischarge mortality, and 1-year postdischarge mortality, respectively^[21]. In addition, BAR has also been shown to be associated with all-cause mortality in acute kidney injury patients, in-hospital mortality in critically ill diabetic ketoacidosis patients, and short-term prognosis in critically ill patients with chronic heart failure^[22,50]. However, the prognostic value of BAR for hospitalized AP patients has not been previously reported. This study, based on the LOCAL cohort and TRACE cohort, is the first to explore the prognostic value of BAR, a novel biomarker for acute severe disease, in assessing and predicting the in-hospital mortality risk in patients predicted with SAP at admission. Additionally, we compared BAR with BUN and ALB alone to assess whether it provides superior predictive performance. Similarly, findings from both cohorts indicated that a higher BAR at admission was significantly related to an elevated risk of in-hospital mortality (LOCAL: HR = 1.23, 95% CI: 1.16, 1.31; TRACE: HR = 1.33, 95% CI: 1.01, 1.76), and this association was nonlinear (P for nonlinearity < 0.001). In predicting in-hospital mortality, BAR (AUC = 0.857) exhibited significantly greater predictive accuracy than BUN (AUC = 0.829) and ALB (AUC = 0.762) (All DeLong test P < 0.05).

In further subgroup ROC and stratified analyses, we identified several findings with important clinical implications. As previously reported^[17], Wang et al demonstrated that the BUN trajectory could predict poor outcomes in AP patients in the intensive care unit, suggesting that both BUN and BAR may have a significant prognostic impact throughout the disease course. Therefore, we conducted a time-dependent ROC analysis to evaluate the predictive accuracy of BAR and BUN for mortality at 3, 5, 7, 9, 14, 28, 60, and 90 days after admission. The results indicated that BAR demonstrated superior predictive performance at all time points compared to BUN and maintained strong predictive accuracy for longer-term outcomes (90 days AUC: BAR: 65.74 vs. BUN: 52.58). Additionally, we classified the participants in the LOCAL and TRACE cohorts based on the etiology of pancreatitis, including HTG-AP, AAP, BAP, and other causes of AP, and subsequently calculated the AUC for BAR in predicting in-hospital mortality within these four subgroups. The results revealed that, across both cohorts, BAR exhibited very high predictive accuracy for in-hospital mortality in patients with BAP, with AUCs of 0.8696 (LOCAL cohort) and 0.8633 (TRACE cohort). Nowadays, BAP remains the most prevalent form of pancreatitis worldwide, accounting for the largest proportion of patients in the pancreatitis population, thus imposing a significant burden on both patients and healthcare systems^[1,51,52]. By combining the easily accessible clinical parameters of BUN and ALB, BAR facilitates accurate and stable prognostic prediction for these patients, enabling clinicians to promptly implement appropriate interventions for high-risk individuals, ultimately improving patient outcomes.

Additionally, stratified analysis showed the in-hospital mortality risk associated with BAR was significantly higher in patients with smoking and drinking histories than those without (P for interaction < 0.05). This suggests that prolonged nicotine and alcohol exposure may be significant risk factors contributing to poor prognosis in AP patients. Previous studies have shown that alcohol plays a multifaceted role in the pathogenesis and progression of acute pancreatitis. Alcohol consumption alters the function of pancreatic acinar cells, reducing their defense mechanisms against trypsinogen activation, thereby promoting premature trypsinogen activation^[53,54]. Moreover, alcohol modifies the immune response in the pancreatic microenvironment. It suppresses anti-inflammatory factor production while promoting the release of proinflammatory cytokines like tumor necrosis factor-alpha and IL-1ß. This exacerbates both local and systemic pancreatic inflammatory responses^[55]. More complexly, the effects of alcohol extend beyond pancreatic damage. Long-term alcohol consumption increases intestinal permeability, allowing endotoxins to enter the pancreas through the bloodstream, further activating immune responses. This process not only worsens local pancreatic inflammation but also triggers systemic SIRS, increasing the risk of multiorgan failure, which is strongly associated with poor outcomes in hospitalized AP patients^[56,57]. Therefore, long-term alcohol consumption not only increases the risk of developing AP but also exacerbates pancreatic damage and systemic inflammation through multiple mechanisms, significantly elevating the risk of in-hospital mortality. Furthermore, our analysis revealed considerable treatment heterogeneity across BAR tertiles with respect to fluid resuscitation volume. In patients with low to moderate BAR levels, increased VFR had minimal impact on mortality, whereas in the high BAR group, elevated VFR was significantly associated with increased risk of death. These findings suggest that BAR may aid in guiding individualized fluid management in predicted SAP. Specifically, patients with high BAR may benefit from more conservative fluid resuscitation strategies to avoid volume overload–associated complications. Integrating BAR into early triage may help optimize resuscitation intensity and improve outcomes in high-risk patients.

Strengths and limitations

Strengths: (1) Compared to other pancreatitis-related cohort studies, it benefits from a large sample size (n = 5384), and its findings have been validated in a high-quality multicenter RCT cohort, thereby enhancing the robustness and external validity of the results. (2) This study is the first to clearly establish the significant role of BAR in predicting the prognosis of patients with predicted SAP upon admission, and it identifies BAP as the most suitable patient group for BAR application. (3) As a prognostic indicator for AP, BAR is easy to calculate, readily available in clinical practice, and provides efficient risk stratification and predictive performance.

However, this study also has several limitations: (1) As a retrospective cohort study, it cannot establish a causal relationship between BAR and the risk of in-hospital mortality in AP patients. (2) Although key confounders were adjusted for, the possibility of residual confounding cannot be excluded. To address this, we calculated E-values, which were sufficiently high to support the robustness of the findings. (3) The study only evaluated the association between BAR levels at admission and patient outcomes, and assessing the dynamic changes in BAR during hospitalization might provide additional prognostic information. (4) While the study provides valuable insights into predicted SAP patients’ high-risk clinical subgroups, its exclusive focus on a Chinese population may limit the generalizability of findings to other ethnic groups and broader acute pancreatitis cohorts.

Conclusion

In conclusion, this study is the first to demonstrate, within a nationwide multicenter cohort, that BAR serves as an effective prognostic indicator for short-term outcomes in patients with predicted SAP upon admission. It shows particularly high prognostic value in patients with BAP, with an accuracy of 86.96% in predicting in-hospital mortality, making it a highly cost-effective prognostic factor.

Supplementary Material

js9-112-1295-001.docx ^{(50KB, docx)}

js9-112-1295-002.docx ^{(4.3MB, docx)}

Ethical approval

This study received ethical approval from the Institutional Review Boards of the First Affiliated Hospital of Nanchang University (Ethics Approval No: 2011001) and the Nanjing Military General Hospital of Jiangsu Province (NCT number: NCT02473406).

Consent

Informed consent from the participants has been obtained for this study.

Sources of funding

This work was supported by the National Natural Science Foundation of China (No. 82370661, No. 81960128); the Double-Thousand Plan of Jiangxi Province (No. jxsq2019201028); the Jiangxi Medicine Academy of Nutrition and Health Management (No. 2022-PYXM-01); and the Science and Technology Innovation Team Cultivation Project of the First Affiliated Hospital of Nanchang University (YFYKCTDPY202202).

Author contributions

Y-Z: Conceptualization, methodology, supervision, and project administration. MB-K: Writing—original draft preparation, validation, and formal analysis. YY-Z: Writing—review & editing and software. YT-L: Writing—review & editing and conceptualization. MB-K, L-D, and HF-X: Software. C-H, NS-L, YP-L, LY-L, and HJ-Z: Writing—review & editing, formal analysis, and validation. NH-L, WH-H, X-H, L-X, L-K, and WJ-M: Data curation and validation. All authors read and approved the final manuscript.

Conflicts of interest disclosure

The authors declare that they have no competing interests.

Guarantor

Yin Zhu.

Research registration unique identifying number (UIN)

None.

Provenance and peer review

Not commissioned, externally peer-reviewed.

Data availability statement

The dataset used and/or analyzed during the current study is available from the corresponding author upon reasonable request.