The association between blood urea nitrogen to albumin ratio and short- and long-term all-cause mortalities in acute pancreatitis: Insights from MIMIC-IV database.

Our findings indicate that the BAR accurately predicts short- and long-term mortality in pancreatitis patients, suggesting its potential as a prognostic tool for AP. Nonetheless, additional large-scale prospective multicenter studies are necessary to assess and validate its applicability.

Acute pancreatitis (AP) is an inflammatory disorder of the pancreas that may progress to severe systemic complications, carrying life-threatening risks. The reported incidence ranges from 13 to 45 cases per 100,000 person-years. [ 1 ] This represents a significant public health burden. Current evidence suggests no substantial gender-related differences in disease prevalence or outcomes. [ 2 ] The pathogenesis of AP involves a complex interplay of premature pancreatic enzyme activation, dysregulated cytokine release, gut microbial translocation, and apoptotic cell death Clinically, approximately 20% of patients progress to severe AP, with in-hospital mortality rates reaching 15% to 30%. [ 3 – 5 ] Mortality is primarily driven by persistent organ failure, infected pancreatic necrosis, and systemic inflammatory response. In particular, in patients with multiorgan dysfunction or persistent organ failure, mortality may approach or exceed 40%. [ 6 – 8 ] These findings underscore the considerable clinical burden of AP and highlight the urgent need for early and reliable prognostic tools. Despite advances in intensive care, mortality in severe AP remains unacceptably high, underscoring the urgent need for reliable, easily applicable biomarkers to facilitate timely risk stratification. Indeed, recent studies have reported that an elevated blood urea nitrogen (BUN) to albumin ratio (BAR) is significantly associated with increased mortality in AP, suggesting its potential as a simple and accessible prognostic indicator. [ 9 , 10 ] Although most cases of AP follow a mild clinical course and resolve with conservative management [ 11 , 12 ] a significant proportion (20–30%) progress to moderate or severe disease, characterized by pancreatic necrosis, multiorgan dysfunction, and adverse clinical outcomes. Given this risk, early severity assessment and timely intervention are crucial for improving patient outcomes. Current clinical practice relies on several scoring systems to guide management decisions in both general and intensive care settings. These include the Ranson criteria, [ 13 ] the Balthazar computed tomography severity index, [ 14 ] the Acute Physiology and Chronic Health Evaluation II (APACHE-II), [ 15 ] the Bedside Index for Severity in Acute Pancreatitis, [ 6 ] and the Sequential Organ Failure Assessment (SOFA) [ 16 ] score. Despite their widespread use, these scoring systems face several limitations. First, their complexity, often requiring the evaluation of multiple clinical and laboratory parameters, poses challenges in rapid bedside application. Second, many of the included variables exhibit delayed positivity and necessitate repeated measurements, limiting their utility for early disease stratification. Finally, subjectivity in interpretation among clinicians may introduce variability in scoring accuracy. These factors collectively increase clinical workload and may delay timely decision-making, highlighting the need for more efficient and objective early assessment tools. Given the limitations of existing prognostic markers, there remains a critical need for a clinically accessible and reliable early-stage assessment tool for AP. BUN, a routinely measured renal function parameter, reflects not only glomerular filtration but also extrarenal factors including dehydration, protein catabolism, and gastrointestinal hemorrhage. [ 17 ] In AP, third-space fluid losses contribute to prerenal azotemia, making serial BUN measurements valuable for guiding fluid resuscitation. [ 18 ] While elevated BUN levels correlate with AP severity and prognosis, [ 19 , 20 ] its isolated use lacks sufficient predictive reliability due to multifactorial influences. Similarly, serum albumin (Alb), a negative acute-phase reactant, demonstrates both antiinflammatory properties and vascular integrity maintenance functions. Hypoalbuminemia (<3.5 g/dL) occurs in approximately 60% of AP cases and independently predicts organ failure progression and mortality. [ 21 , 22 ] However, Alb concentrations may be confounded by chronic malnutrition or inflammatory states, limiting its standalone predictive value. Emerging evidence suggests that composite biomarkers outperform singular parameters in critical care prognostication. The BAR has demonstrated prognostic utility in acute ischemic stroke and respiratory failure. [ 23 , 24 ] Taken together, these findings suggest that BAR, by integrating BUN and albumin (2 routinely available laboratory parameters) may offer superior prognostic performance compared to either alone in AP. However, despite emerging evidence of its association with mortality, further validation using advanced analytical approaches remains warranted. To address this knowledge gap, we conducted a retrospective cohort study using the Medical Information Mart for Intensive Care IV (MIMIC-IV v2.2) database. This study aimed to investigate the association between the BAR and all-cause mortality in patients with AP.

Conceptualization: Min Zhang. Data curation: Min Zhang, Xingyi Yang. Formal analysis: Min Zhang, Lihong Lv, Guangdong Wang, Lang Gao, Yuanshuo Ge. Software: Liyin Jin. Validation: Guangdong Wang. Visualization: Liyin Jin, Guangdong Wang, Yuanshuo Ge. Writing – original draft: Min Zhang, Lihong Lv, Xingyi Yang. Writing – review & editing: Xingyi Yang.

The data for this study was sourced from the MIMIC-IV (v2.2) database. [ 25 ] This extensive database comprehensively documents data from over 70,000 ICU patients at Beth Israel Deaconess Medical Center from 2012 to 2019, encompassing test results, prescription schedules, vital signs, and hospital stay durations. The Institutional Review Board at Beth Israel Deaconess Medical Center granted ethical approval for this database (2001P-001699/14). To safeguard privacy, all identifiable patient information was removed and substituted with randomly generated numerical codes. Given this anonymization process, neither additional ethical clearance nor individual consent was required. The author obtained access to the MIMIC database after completing the Collaborative Institutional Training Initiative Program (certificate number 67058598). In this retrospective cohort study, patients diagnosed with AP were included. The diagnosis of AP is determined by International Classification of Diseases codes. The study included 5894 patients with AP, with 1276 requiring ICU admission. Patients were excluded if they were under 18 years old, spent <24 hours in the ICU, lacked BUN and Alb data within 24 hours of admission, or had previous ICU admissions (Fig. 1 ). A flow diagram of study participants. We collected demographic details, initial clinical vital signs, laboratory data, comorbidities, and treatment outcomes from the database, focusing on the critical first 24 hours after ICU admission. We monitored vital signs such as blood pressure, heart rate, respiratory rate, and body temperature. We also documented extensive laboratory parameters, including blood glucose, WBC and platelet counts, sodium, potassium, BUN, creatinine, bilirubin, albumin, prothrombin time, international normalized ratio, lactic acid, and various complications. Data on the SOFA scores were gathered. The BAR was chosen as the main study variable. The model was derived from the average of parameters recorded during the first 24 hours of ICU admission for all patients. The cohort in the study met established diagnostic criteria for AP and was followed for at least 365 days to assess long-term outcomes. Table 1 provides a detailed list of the extracted variables. The study’s primary endpoint was all-cause mortality at 28, and 90 days, as well as 365 days following hospital admission. We assessed the ICU and hospital mortality during this period. Characteristics of included patients. AKI = acute kidney injury, ALT = alanine aminotransferase, AST = aspartate aminotransferase, BAR = blood urea nitrogen to albumin ratio, BUN = blood urea nitrogen, Ca = calcium, CRRT = continuous renal replacement therapy, DBP = diastolic blood pressure, ERCP = endoscopic retrograde cholangiopancreatography, INR = international normalized ratio, MAP = mean arterial pressure, MV = mechanical ventilation, PT = prothrombin time, RBC = red blood cell, RDW = erythrocyte distribution width, SBP = systolic blood pressure, SOFA = Sepsis-Related Organ Failure Assessment Score, WBC = white blood cell. Continuous variables were expressed as mean ± standard deviation (normally distributed data) or median with interquartile range (non-normally distributed data), while categorical variables were reported as frequencies and percentages. Baseline characteristics were compared using Student t tests or 1-way ANOVA for continuous variables and Pearson χ ² tests or Fisher exact tests for categorical variables, as appropriate. We employed the Boruta machine-learning [ 26 ] algorithm for predictive feature selection and subsequently stratified patients into low-risk (BAR < 9.62) and high-risk (BAR ≥ 9.62) groups using X-tile software (v3.6.1, Yale University, New Haven) based on 28-day mortality thresholds. Multivariable Cox proportional hazards regression analysis identified independent prognostic factors for mortality at multiple time points (28, 90, and 365 days post-admission), with results expressed as adjusted hazard ratios and 95% confidence intervals. Kaplan–Meier (KM) survival analysis with log-rank testing compared mortality outcomes between risk groups. Predictive performance was evaluated through receiver operating characteristic (ROC) curve analysis for BAR, BUN, albumin, and SOFA scores, with corresponding sensitivity, specificity, and area under the curve (AUC) values calculated. Restricted cubic spline regression assessed potential nonlinear relationships between BAR and clinical outcomes. Comprehensive subgroup analyses examined potential interaction effects by age, sex, acute kidney injury (AKI) status, sepsis diagnosis, and procedural interventions including endoscopic retrograde cholangiopancreatography and continuous renal replacement therapy (CRRT). A 2-sided P -value < .05 was considered statistically significant. All statistical analyses were performed using R software (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria).

This study enrolled 492 patients, with an optimal BAR cutoff of 9.62 determined by X-tile software based on 28-day mortality. Patients were stratified into a high-BAR group (≥9.62) and a low-BAR group (<9.62). Comparative analysis revealed that the high-BAR group had significantly higher values for age, BUN, serum potassium, creatinine, red cell distribution width, international normalized ratio, prothrombin time, anion gap, and SOFA score (all P  < .05). Conversely, this group demonstrated significantly lower platelet counts, Alb levels, mean arterial pressure, and body temperature compared to the low-BAR group. The high-BAR group also showed a greater prevalence of comorbidities including AKI, sepsis, and chronic kidney disease. In terms of clinical outcomes, the high-BAR group exhibited significantly elevated mortality rates across all measured timepoints: ICU mortality (21.8% vs 3.5%; P  < .001), in-hospital mortality (30.5% vs 6.3%; P  < .001), 28-day mortality (27.6% vs 5.7%; P  < .001), 90-day mortality (39.7% vs 10.4%; P  < .001), and 365-day mortality (43.1% vs 15.7%; P  < .001). Detailed baseline characteristics of both groups are presented in Table 1 . This study systematically evaluated prognostic factors in AP using the Boruta algorithm (a robust machine-learning feature selection method based on random forests) to identify key predictive variables. Feature importance analysis identified the BAR, serum lactate levels, requirement for CRRT, total bilirubin, and patient age as variables with significant independent prognostic value. These findings underscore the pivotal role of the BAR in prognostic evaluation for AP. It not only validates its predictive utility but also provides a reliable quantitative basis for risk stratification in patients with this condition (Fig. 2 ). The feature importance analysis based on Boruta pinpoints critical predictors for 28-day mortality in AP patients. AP = acute pancreatitis. Cox regression analysis showed that BAR ≥ 9.62 was significantly associated with higher short- and long-term mortality in AP patients, even without adjusting for other factors (28-day HR = 5.5, 95% CI: 3.2–9.45, P  < .001; 90-day HR = 4.63, 95% CI: 3.06–7.02, P  < .001; and 365-day HR = 3.45, 95% CI: 2.41–4.94, P  < .001). In Model 1 (adjusted for age and sex), patients with BAR ≥ 9.62 continued to exhibit significantly elevated mortality risk: 28-day HR = 4.67 (95% CI: 2.69–8.1, P  < .001), 90-day HR = 3.88 (95% CI: 2.54–5.92, P  < .001), and 365-day HR = 2.86 (95% CI: 1.98–4.12, P  < .001). Further adjustments in Model 2 maintained this association, reinforcing that BAR ≥ 9.62 is an independent predictor of mortality (28-day HR = 3.30, 95% CI: 1.74–6.25, P  < .001; 90-day HR = 2.97, 95% CI: 1.81–4.88, P  < .001; 365-day HR = 2.29, 95% CI: 1.47–3.57, P  < .001). For complete results, see Table 2 . Association between BAR and mortality in patients with AP. Model 1: Adjusted age and gender. Model 2: Model 1 + SBP, creatinine, lactate, INR, AKI, sepsis, ERCP, octreotide, and CRRT. KM analysis demonstrated consistently elevated mortality rates in the BAR ≥ 9.62 cohort relative to the BAR < 9.62 group across all evaluated time intervals: 28-day mortality (27.5% vs 5.6%; P  < .001), 90-day mortality (39.6% vs 10.3%; P  < .001), and 365-day mortality (43.1% vs 15.7%; P  < .001) (Fig. 3 ). Subsequent ROC curve evaluation of 4 prognostic indicators (BUN, Alb, BAR ratio, and SOFA score) identified the BAR as exhibiting significantly greater discriminatory power for all-cause mortality prediction (all P  < .05 vs comparator metrics) (Fig. 4 ). Notably, the BAR achieved optimal predictive performance for 90-day mortality, with AUC of 0.748. Comprehensive comparative data are detailed in Table 3 . Information of ROC curves in Figure 4 . Alb = albumin, AUC = area under the curve, BAR = blood urea nitrogen to albumin ratio, BUN = blood urea nitrogen, CI = confidence interval, ROC = receiver operating characteristic, SOFA = Sepsis-related Organ Failure Assessment score; Threshold indicates the optimal cutoff value derived from ROC analysis. Kaplan–Meier survival analysis curves for all-cause mortality in patients with AP at 28 days (A), 90 days (B), and 365 days (C) of hospital admission. AP = acute pancreatitis. ROC curves for predicting all-cause mortality in patients with AP at 28 days (A), 90 days (B), and 365 days (C) after admission. AP = acute pancreatitis, ROC = receiver operating characteristic. Using RCS analysis, we demonstrated distinct temporal patterns in the association between the BAR and AP prognosis. A linear relationship was observed at both 28 days ( P for nonlinearity = .179) and 365 days ( P for nonlinearity = .062; P for overall < .001). In contrast, a nonlinear association was evident at 90 days ( P for nonlinearity = .018; P for overall < .001). These results indicate that elevated BAR levels predict higher risks of adverse clinical outcomes. Further details are illustrated in Figure 5 . Association between BAR and survival with the RCS function at 28 days (A), 90 days (B), and 365 days (C) after admission. BAR = blood urea nitrogen to albumin ratio, RCS = restricted cubic splines. To evaluate potential subgroup-specific effects, we performed stratified analyses for age, sex, AKI, sepsis, endoscopic retrograde cholangiopancreatography, and CRRT. No significant interaction was observed between the BAR and any subgroup ( P  > .05), indicating consistent predictive performance across all tested populations. These findings support the robustness of BAR as a prognostic marker, with detailed subgroup comparisons illustrated in Figure 6 . Forest plots of subgroup analysis of the relationship between all-cause mortality and BAR in patients with AP admitted 28 days (A), 90 days (B), and 365 days (C). AP = acute pancreatitis, BAR = blood urea nitrogen to albumin ratio.

AP is a prevalent clinical condition, yet the standardization of therapeutic approaches and precise evaluation of disease severity continue to pose substantial challenges, particularly in prognosticating moderate-to-severe cases. While numerous scoring systems have been employed in recent years to assess AP severity, their clinical utility is frequently hampered by cumbersome indicators and computational inefficiency, restricting their practical application in routine care. Emerging evidence suggests that simplified prognostic ratios (such as the total bilirubin-to-albumin ratio, [ 27 ] red cell distribution width-to-platelet ratio [ 28 ] ) may serve as practical clinical predictors of AP outcomes. In this study, we demonstrate that the BAR independently predicts all-cause mortality in AP patients. KM survival analysis revealed a markedly poorer prognosis in patients with a BAR ≥ 9.62 compared to those with a BAR < 9.62 ( P  < .001). RCS analysis further confirmed a significant positive correlation between the BAR and clinical outcomes. Subgroup analyses revealed no significant interaction effects, reinforcing the predictive stability of the BAR across different patient populations. Notably, as evaluated by the AUC, BAR demonstrated higher ROC values than individual biomarkers (albumin, BUN) and composite scoring systems SOFA in both short- and long-term prognostic assessments, exhibiting moderate discriminative capacity. The consistent performance of BAR strongly supports its reliability and generalizability as a prognostic tool in AP. In AP, elevated BUN levels may indicate renal dysfunction resulting from either inflammation-induced increased vascular permeability or renal hypoperfusion. [ 6 , 29 ] Moreover, dehydration can lead to enhanced renal urea reabsorption, further contributing to BUN elevation. [ 30 ] Clinical evidence consistently demonstrates that BUN elevation correlates with disease severity and poorer prognosis in AP, particularly in severe cases. [ 18 , 31 ] Albumin, the predominant plasma protein, reflects both hepatic synthetic capacity and nutritional status. [ 32 ] Hypoalbuminemia (which may arise from inflammatory extravascular loss, impaired hepatic synthesis, or malnutrition) serves as a predictor for increased infection risk and potential organ failure. [ 33 , 34 ] Numerous studies have confirmed its association with AP severity and adverse outcomes. [ 35 , 36 ] To improve prognostic accuracy in AP patients, our approach minimizes the confounding effect of individual biomarkers by utilizing inverse variations from 2 distinct pathological pathways, thereby reducing reliance on any single indicator in the predictive model. Emerging evidence highlights the BAR as a robust prognostic biomarker across multiple diseases. Feng et al [ 37 ] retrospectively analyzed 1558 pneumonia patients and identified BAR as an independent predictor of poor prognosis, with a hazard ratio of 3.87 (95% CI: 2.174–6.893, P  < .001) and an AUC of 0.685. Similarly, Min et al [ 38 ] reported consistent findings in 10,578 sepsis patients, confirming BAR as an independent risk factor (HR = 1.832, 95% CI: 1.659–2.022, P  < .001). Additionally, Bae et al [ 39 ] validated BAR’s prognostic utility in gastrointestinal bleeding patients (AUC = 0.682). These findings align with our study, reinforcing BAR’s significant role in disease prognosis assessment. Consistent with our findings, 2 recent studies specifically examined the prognostic role of BAR in AP. Li et al [ 9 ] and Xia et al [ 10 ] both reported that elevated BAR levels were independently associated with increased mortality in AP patients, supporting its clinical relevance as a simple and effective biomarker. Compared with these studies, our analysis adds methodological innovations, including the use of Boruta feature selection to robustly identify BAR among a broader set of candidate predictors, and RCS modeling to delineate the dose–response relationship between BAR and mortality. These approaches not only validate BAR’s prognostic value in AP but also provide more granular insights into its clinical applicability. BAR integrates 2 key biomarkers (BUN and albumin) providing a more comprehensive prognostic evaluation than either parameter alone. Our results demonstrate that elevated BAR levels correlate with AP severity and poor outcomes, establishing it as a valuable clinical indicator. Mechanistically, increased BAR reflects concomitant rises in BUN (indicating disease severity or renal dysfunction) and decreases in Alb (suggesting inflammatory extravasation, impaired hepatic synthesis, or malnutrition). This dual-parameter approach enhances prognostic accuracy by simultaneously assessing inflammation, nutritional status, and potential dehydration. Clinically, BAR offers several advantages: its calculation relies solely on routine BUN and Alb measurements, obviating additional costs or tests; it facilitates rapid risk stratification: AP patients with BAR values exceeding the established threshold exhibited worse prognoses, necessitating intensified therapeutic interventions, whereas those below the cutoff may receive standard care. This stratification optimizes resource allocation and improves clinical efficiency. In conclusion, BAR is a practical, cost-effective prognostic tool that synthesizes BUN and Alb levels to enhance AP outcome prediction. Its integration into clinical practice could aid in early risk identification, tailored treatment strategies, and improved patient management. This study employed large-scale, real-world data from public databases to analyze disease prognostic factors, offering methodological strengths in epidemiological generalizability, though notable limitations must be recognized. The absence of standardized AP severity grading precluded stratified analyses across severity subgroups, while unavailable etiological classifications constrained prognosis evaluation by disease subtype or causative mechanism. Mortality outcomes were classified as all-cause due to insufficient death attribution granularity (a limitation masking potential etiology-specific survival patterns) though critically ill mortality likely reflects multifactorial influences. Additionally, the retrospective registry-based design inherently risks selection bias.