Association between a 24-h increase in blood urea nitrogen and clinical outcomes in acute non-variceal upper gastrointestinal bleeding: a dual-center retrospective cohort study

Abstract

Background and aim:

Blood urea nitrogen (BUN) is incorporated into admission-based risk scores for acute non-variceal upper gastrointestinal bleeding (NVUGIB); however, the clinical relevance of early in-hospital BUN kinetics remains unclear.

Objectives:

To evaluate whether a 24-h increase in BUN is independently associated with adverse clinical outcomes in patients with acute NVUGIB.

Design:

Dual-center retrospective cohort study.

Methods:

We conducted a dual-center retrospective cohort study of adult patients with endoscopically confirmed NVUGIB admitted between 2018 and 2023. The exposure was defined as any absolute increase in BUN within 18–30 h after the baseline measurement at presentation. The primary outcome was 30-day all-cause mortality. Secondary outcomes included in-hospital mortality, in-hospital rebleeding, red blood cell transfusion, length of hospital stay, and need for radiologic or surgical intervention. Multivariable regression was performed in a propensity score-matched cohort with confirmatory analyses in the original cohort.

Results:

Among 611 patients, 218 (35.7%) demonstrated a 24-h increase in BUN. Propensity score matching yielded 400 patients (200 per group). In the matched cohort, a 24-h BUN increase was independently associated with higher 30-day mortality (adjusted odds ratio (aOR) 3.307; 95% confidence interval (CI) 1.604–6.819) and in-hospital mortality (aOR 2.732; 95% CI 1.208–6.178). No independent associations were observed with rebleeding, transfusion requirements, or radiologic/surgical intervention.

Conclusion:

An increase in BUN within the first 24 h of hospitalization is independently associated with higher short-term mortality in acute NVUGIB. Early BUN kinetics may serve as a complementary risk marker, but prospective validation and formal predictive-performance testing are needed before clinical integration.

Plain language summary

Does a rise in blood urea nitrogen during the first day in hospital predict worse outcomes in patients with upper gastrointestinal bleeding?

Upper gastrointestinal bleeding is a common medical emergency that can be life-threatening. Doctors use several clinical signs and blood tests at the time a patient arrives at the hospital to estimate how severe the bleeding is and to predict outcomes. One commonly measured blood test is blood urea nitrogen (BUN), which can rise due to blood loss, dehydration, or reduced blood flow to the kidneys. However, most existing risk scores rely only on the first BUN value measured on admission and do not consider how this value changes over time. In this study, we examined whether an increase in BUN during the first 24 hours after hospital admission is linked to worse outcomes in patients with non-variceal upper gastrointestinal bleeding. We analyzed data from 611 patients treated at two hospitals and compared patients whose BUN increased within 24 hours with those whose BUN did not increase. To ensure a fair comparison, we used statistical methods to balance differences in patient characteristics between groups. We found that patients whose BUN increased within the first 24 hours had a significantly higher risk of dying during hospitalization and within 30 days after admission. This association remained even after accounting for differences in age, kidney function, illness severity, and other important clinical factors. In contrast, a rise in BUN was not linked to a higher risk of rebleeding or the need for additional procedures. These findings suggest that monitoring changes in BUN during the first day of hospitalization may help identify patients at higher risk of poor outcomes. Because BUN is a simple and widely available blood test, tracking its early changes could support closer monitoring and improved care for patients with upper gastrointestinal bleeding.

Introduction

Acute non-variceal upper gastrointestinal bleeding (NVUGIB) remains a frequent medical emergency worldwide and continues to be associated with considerable short-term morbidity, mortality, and healthcare utilization despite advances in pharmacologic therapy and endoscopic hemostatic techniques.^1,2 Early risk stratification is central to contemporary management, as it informs triage decisions, resuscitative intensity, level-of-care allocation, and timing of endoscopic intervention. Current clinical guidelines from the American College of Gastroenterology, the European Society of Gastrointestinal Endoscopy, and the National Institute for Health and Care Excellence emphasize early identification of high-risk patients and recommend endoscopy within 24 h of presentation as a cornerstone of care.^3–5

Several prognostic scoring systems have been developed to facilitate early risk assessment in NVUGIB.^6,7 The Glasgow–Blatchford Score (GBS) is widely used to predict the need for clinical intervention and incorporates laboratory variables including blood urea nitrogen (BUN), hemoglobin level, and hemodynamic parameters.^8–10 In contrast, the AIMS65 score was designed primarily to predict in-hospital mortality and includes age, serum albumin, international normalized ratio (INR), altered mental status, and systolic hypotension, without incorporating renal or nitrogen-related biomarkers. ¹¹ Importantly, these tools are based solely on admission data and do not account for physiological changes occurring after initial presentation.

BUN is a multifactorial biomarker influenced by renal perfusion, intravascular volume status, protein catabolism, and gastrointestinal absorption of nitrogenous compounds derived from digested blood. ¹² In patients with NVUGIB, elevated baseline BUN has been associated with greater bleeding severity, higher transfusion requirements, and adverse outcomes.^13,14 However, BUN levels may evolve during hospitalization in response to ongoing hemorrhage, resuscitative adequacy, renal hemodynamics, and systemic stress, and the clinical implications of such early changes remain unclear.

Laboratory trajectories over time may reflect evolving pathophysiology not captured by a single measurement at admission. An early rise in BUN could be associated with persistent hypoperfusion, neurohormonal activation, or unresolved systemic stress during the resuscitative phase. Whether early BUN kinetics are independently associated with clinically meaningful outcomes in NVUGIB has not been well established.

Accordingly, we conducted a dual-center retrospective cohort study to evaluate whether an increase in BUN within 24 h of admission is independently associated with adverse clinical outcomes in patients with acute NVUGIB. We specifically examined associations with short-term mortality, rebleeding, transfusion requirements, length of hospital stay, and need for additional interventions. This study aims to clarify the clinical relevance of early BUN kinetics during the initial phase of NVUGIB management.

Methods

Study design and setting

This dual-center retrospective cohort study was conducted at two tertiary care hospitals in Thailand—Hatyai Hospital and Ayutthaya Hospital—between January 2018 and December 2023. Adult patients (⩾18 years) who presented to the emergency department with signs and symptoms of acute upper gastrointestinal bleeding and were subsequently admitted for inpatient management were screened for eligibility.

To ensure diagnostic homogeneity, only patients with endoscopically confirmed NVUGIB were included. Exclusion criteria comprised end-stage renal disease requiring dialysis, absence of baseline or 24-h BUN measurements, death or hospital discharge prior to the 24-h laboratory assessment, and incomplete data required for analysis.

Because the exposure required paired BUN measurements, patients who died before the 24-h BUN assessment window could not be classified and were excluded. Only four patients died within the first 24 h after admission. This design, therefore, represents a 24-h landmark analysis, which may introduce survivor (immortal time) bias; however, the small number of early deaths makes substantial bias unlikely.

Exposure definition

The primary exposure was defined as an increase in BUN at 24 h after admission compared with the baseline value at presentation and was analyzed as a binary variable (increase vs no increase). To account for real-world variability in laboratory timing, the 24-h BUN measurement was operationally defined as any value obtained within 18–30 h following the initial baseline measurement.

Any absolute increase greater than zero compared with baseline was classified as a BUN increase, without applying a minimum absolute or percentage threshold. This approach was chosen to capture early directional changes in BUN during the initial hospitalization period rather than relying solely on a single admission value. Patients were categorized into two groups accordingly.

Outcome measures

The primary outcome was 30-day all-cause mortality. Secondary outcomes included in-hospital mortality, in-hospital rebleeding, need for red blood cell (RBC) transfusion (analyzed as a binary outcome and as transfusion volume), length of hospital stay, and additional interventions, defined as radiologic or surgical hemostatic procedures.

Data collection and covariates

All data were extracted from electronic medical records using a standardized data abstraction form developed a priori. Collected variables included baseline demographic characteristics, comorbidities (including chronic kidney disease not requiring dialysis, cirrhosis, hypertension, diabetes mellitus, and cardiovascular disease), presenting clinical features, and medication use prior to admission (nonsteroidal anti-inflammatory drugs, antiplatelet agents, anticoagulants, proton pump inhibitors, and herbal medications).

Laboratory parameters recorded at admission and at 24 h included hemoglobin, BUN, serum creatinine, serum albumin, INR, and serum electrolytes. Clinical severity at presentation was assessed using hemodynamic parameters, presence of shock, and nasogastric aspirate findings. The GBS was calculated using admission data for descriptive purposes.

Endoscopic variables included timing of upper endoscopy, endoscopic diagnoses, and early management prior to endoscopy, including administration of intravenous proton pump inhibitors and prokinetic agents.

Propensity score matching

To reduce confounding and account for baseline differences between groups, propensity score matching (PSM) was performed. Propensity scores were estimated using a multivariable logistic regression model incorporating clinically relevant covariates selected a priori, including age, sex, hospital center, baseline laboratory values (BUN, hemoglobin, serum creatinine, INR), presence of shock at presentation, comorbidities, and endoscopic findings.

Nearest-neighbor matching without replacement was performed in a 1:1 ratio using a caliper width of 0.2 of the standard deviation of the logit of the propensity score. Balance after matching was assessed using standardized mean differences (SMDs), with values <0.1 considered indicative of adequate covariate balance.

Statistical analysis

Descriptive statistics were used to summarize baseline characteristics and clinical outcomes. Continuous variables were expressed as means with standard deviations or medians with interquartile ranges, as appropriate based on distribution, and compared using Student’s t test or the Mann–Whitney U test. Categorical variables were summarized as counts and percentages and compared using the Chi-square test or Fisher’s exact test, as appropriate.

To evaluate the independent association between a 24-h BUN increase and each outcome, multivariable regression analyses were performed. Binary outcomes were analyzed using logistic regression models, with results reported as adjusted odds ratios (aORs) and 95% confidence intervals (CIs). Continuous outcomes, including transfusion volume and length of hospital stay, were analyzed using linear regression models, with results reported as regression coefficients.

Covariates included in adjusted models were selected based on clinical relevance and potential residual imbalance and included age, sex, body mass index, hospital center, presence of shock at presentation, fresh blood via nasogastric tube, serum albumin, serum creatinine, and GBS. Shock was defined as mean arterial pressure <65 mmHg with heart rate >100 beats/min at initial evaluation. Length of hospital stay was analyzed as a continuous variable measured in days.

Collinearity was assessed using variance inflation factors (VIFs), with values <5 considered acceptable. To minimize model overfitting, covariate selection prioritized physiologically relevant markers of bleeding severity and baseline risk. Events-per-variable ratios were calculated for mortality models by dividing the number of outcome events by the number of predictors included.

Primary multivariable analyses were conducted in the propensity score-matched cohort, with confirmatory analyses performed in the full unmatched cohort.

Prespecified sensitivity analyses included: (1) replacing GBS with the complete Rockall score; (2) modeling systolic blood pressure and heart rate as continuous variables instead of binary shock; (3) adjusting for chart-diagnosed CKD and repeating analyses after excluding CKD; and (4) adjusting for RBC transfusion within 24 h as a continuous variable.

Missing data were minimal in both cohorts (<2% for all variables included in final models; Table S1). Given the low degree of missingness, complete-case analysis was performed without multiple imputation.

All statistical analyses were conducted using STATA, version 15.1 (StataCorp LLC, College Station, TX, USA). Two-sided p values <0.05 were considered statistically significant.

Sample size considerations

The sample size was determined pragmatically based on eligible patients during the study period. Prior to matching, 611 patients with confirmed NVUGIB were included, of whom 218 (35.7%) exhibited a 24-h BUN increase. Formal a priori sample size calculation was not performed given the observational design; effect estimates are interpreted with emphasis on CI precision.

Ethics approval

The study was conducted in accordance with the Declaration of Helsinki and was approved by the institutional review boards of both participating centers (Hatyai Hospital: HYH EC 097-65-01; Ayutthaya Hospital: COA 061/2568). Given the retrospective nature of the study and the use of anonymized data, the requirement for informed consent was waived.

The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. ¹⁵

Results

Study population

During the study period, 611 adult patients with endoscopically confirmed acute NVUGIB met the inclusion criteria and were included in the analysis. Of these, 218 patients (35.7%) exhibited an increase in BUN at 24 h after admission, while 393 patients (64.3%) did not. PSM yielded a balanced cohort of 400 patients, comprising 200 patients with a 24-h BUN increase and 200 patients without a BUN increase (Figure 1).

Figure 1.

Study flow diagram of patient selection and PSM. Flow diagram illustrating patient selection for the dual-center retrospective cohort study. Adult patients presenting with acute upper gastrointestinal bleeding were screened for eligibility. Only patients with endoscopically confirmed NVUGIB and available baseline and 24-h BUN measurements were included. Patients were categorized according to the presence or absence of a BUN increase at 24 h after admission. PSM yielded a final matched cohort of 400 patients (200 per group).

BUN, blood urea nitrogen; NVUGIB, non-variceal upper gastrointestinal bleeding; PSM, propensity score matching; UGIB, upper gastrointestinal bleeding.

Baseline characteristics

Baseline demographic, clinical, laboratory, and endoscopic characteristics of the original and propensity score-matched cohorts are summarized in Table 1.

Table 1.

Baseline characteristics, clinical presentation, laboratory findings, and endoscopic diagnoses of patients with non-variceal upper gastrointestinal bleeding.

Baseline characteristic	Original subjects (n = 611)			Propensity score matching (n = 400)
	Non-increase in BUN (n = 393)	Increase in BUN (n = 218)	p-Value	Non-increase in BUN (n = 200)	Increase in BUN (n = 200)	p-Value
Demographics
Age (years), mean ± SD	60.4 ± 15.2	59.5 ± 16.1	0.500	58.4 ± 14.9	59.9 ± 15.7	0.327
Male sex, n (%)	295 (75.1)	158 (72.5)	0.471	143 (71.5)	148 (74.0)	0.573
Body mass index (kg/m2), mean ± SD	23.7 ± 4.4	23.1 ± 4.6	0.117	24.3 ± 4.2	24.1 ± 4.7	0.655
Previous upper gastrointestinal bleeding, n (%)	69 (17.6)	41 (18.8)	0.709	30 (15.0)	31 (15.5)	0.133
Comorbidities
Hypertension, n (%)	181 (46.1)	86 (39.4)	0.106	83 (41.5)	78 (39.0)	0.578
Diabetes mellitus, n (%)	100 (25.4)	42 (19.3)	0.085	43 (21.5)	34 (17.0)	0.254
Chronic kidney disease (non-dialysis), n (%)	61 (15.5)	38 (17.4)	0.543	22 (11.0)	26 (13.0)	0.532
Cirrhosis, n (%)	43 (10.9)	24 (11.0)	0.975	22 (11.0)	21 (10.5)	0.873
Ischemic heart disease, n (%)	17 (4.3)	15 (6.9)	0.162	6 (3.0)	13 (6.5)	0.103
Congestive heart failure, n (%)	8 (2.0)	7 (3.2)	0.418	3 (1.5)	7 (3.5)	0.198
Dyslipidemia, n (%)	66 (16.8)	28 (13.3)	0.190	22 (11.0)	27 (13.5)	0.444
Cerebrovascular disease, n (%)	34 (8.7)	15 (6.9)	0.424	19 (9.5)	15 (7.3)	0.454
Chronic obstructive pulmonary disease, n (%)	5 (1.3)	9 (4.1)	0.062	2 (1.0)	7 (3.5)	0.173
Clinical presentation
Fresh blood via nasogastric tube, n (%)	71 (18.1)	47 (21.6)	0.295	35 (17.5)	37 (18.5)	0.796
Shock at initial assessment, n (%)	49 (12.5)	41 (18.8)	0.034	29 (14.3)	36 (18.0)	0.353
Glasgow–Blatchford score, mean ± SD	11.0 ± 4.0	10.0 ± 4.2	0.004	10.0 ± 4.4	9.8 ± 4.2	0.641
Medication
Aspirin, n (%)	43 (10.9)	26 (11.9)	0.698	21 (10.5)	21 (10.5)	1.000
Clopidogrel, n (%)	11 (2.8)	18 (8.2)	0.002	5 (2.5)	10 (5.0)	0.189
Any antiplatelet agent, n (%)	39 (9.9)	27 (12.4)	0.344	21 (10.5)	21 (10.5)	1.000
Anticoagulant, n (%)	12 (3.1)	4 (1.8)	0.278	10 (5.0)	4 (2.0)	0.105
NSAIDs, n (%)	84 (21.4)	56 (25.7)	0.418	38 (19.0)	46 (23.0)	0.352
Herbal medication, n (%)	20 (5.1)	18 (8.3)	0.224	8 (4.0)	13 (6.5)	0.262
Proton pump inhibitors, n (%)	57 (14.5)	30 (13.8)	0.807	29 (14.5)	28 (14.0)	0.885
Laboratory values at admission
Hemoglobin (g/dL), mean ± SD	8.1 ± 2.9	8.6 ± 3.2	0.055	8.5 ± 3.0	8.6 ± 3.2	0.748
Platelet count (×103/μL): median (IQR)	248 (181–324)	237 (183–314)	0.554	235 (185–320)	227 (183–311)	0.568
BUN (mg/dL), median (IQR)	39 (26–57)	28 (15–42)	<0.001	32 (20–40)	27 (15–40)	0.060
Serum creatinine (mg/dL), mean ± SD	1.1 ± 0.9	1.9 ± 1.5	<0.001	1.4 ± 0.9	1.4 ± 0.8	1.000
Serum albumin (g/dL), mean ± SD	3.0 ± 0.7	2.8 ± 0.7	0.001	3.0 ± 0.8	2.9 ± 0.8	0.212
Total bilirubin (mg/dL), mean ± SD	1.8 ± 1.1	1.7 ± 1.2	0.308	1.8 ± 1.3	1.7 ± 1.2	0.459
Serum Sodium (mmol/L), mean ± SD	138.5 ± 7.1	138.8 ± 4.5	0.523	138.7 ± 5.0	138.8 ± 4.6	0.835
International normalized ratio, mean ± SD	1.1 ± 0.2	1.2 ± 0.3	<0.001	1.1 ± 0.2	1.1 ± 0.2	1.000
Center, n (%)			<0.001			0.455
Hatyai	344 (87.5)	164 (75.2)		157 (78.5)	151 (75.5)
Ayutthaya	49 (12.5)	54 (24.8)		43 (21.5)	49 (24.5)
Early management before index endoscopy
Time to endoscopy (h), median (IQR)	21 (13–45)	21 (12–47)	0.753	22 (15–48)	21 (12–49)	0.873
Endoscopy within 24 h, n (%)	227 (57.8)	131 (60.0)	0.567	105 (52.5)	119 (59.5)	0.158
Pre-endoscopic PPIs, n (%)	387 (98.5)	214 (98.2)	0.742	198 (99.0)	197 (98.5)	0.652
Pre-endoscopic prokinetic, n (%)	3 (0.8)	2 (0.9)	0.653	0 (0)	2 (1.0)	0.499
Hemoglobin level before endoscopy (g/dL), mean ± SD	8.5 ± 2.4	8.9 ± 2.6	0.062	8.9 ± 2.45	9.0 ± 2.6	0.696
BUN level at 24 h of admission (mg/dL), median (IQR)	28 (17–42)	37 (23–52)	0.002	24 (15–34)	36 (22–51)	0.022
Endoscopic findings
Gastric ulcer, n (%)	167 (42.5)	87 (37.9)	0.521	81 (40.5)	73 (36.5)	0.393
Duodenal ulcer, n (%)	137 (34.1)	53 (24.3)	0.006	51 (25.5)	49 (24.5)	0.822
Mallory–Weiss tear, n (%)	15 (3.8)	13 (6.0)	0.238	6 (3.0)	11 (5.5)	0.216
Cameron, n (%)	1 (0.3)	0 (0)	1.000	0 (0.0)	0 (0.0)	N/A
Gastritis, n (%)	78 (19.8)	46 (21.1)	0.728	49 (24.5)	42 (21.0)	0.401
Duodenitis, n (%)	20 (5.1)	6 (2.8)	0.179	6 (3.0)	6 (3.0)	1.000
Esophagitis, n (%)	38 (9.7)	21 (9.6)	0.966	29 (14.5)	21 (10.5)	0.230
Esophageal ulcer, n (%)	16 (4.1)	11 (5.0)	0.594	10 (5.0)	10 (5.0)	1.000
Tumor bleeding, n (%)	7 (1.8)	8 (3.7)	0.173	3 (1.6)	8 (4.1)	0.126
Dieulafoy lesion, n (%)	1 (0.3)	3 (1.4)	0.309	0 (0)	2 (1.0)	0.499
Helicobacter pylori infection, n (%)	127 (32.3)	40 (18.3)	<0.001	40 (20.0)	39 (19.5)	0.906

Data are expressed as number (%) unless otherwise specified.

BUN, blood urea nitrogen; IQR, interquartile ranges; NSAIDs, non-steroidal anti-inflammatory drugs; PPIs, proton pump inhibitors; SD, standard deviation.

In the original cohort, patients with a 24-h BUN increase exhibited greater baseline severity, reflected by higher GBS, lower serum albumin levels, higher serum creatinine levels, higher INR level, higher rate of shock at initial assessment and clopidogrel use. The prevalence of duodenal ulcer bleeding was higher in the 24-h BUN increase group. Differences in hospital center distribution were also observed. After PSM, baseline characteristics were well balanced between groups, including demographics, comorbidities, presenting features, laboratory values, endoscopic findings, timing of endoscopy, and pre-endoscopic management. SMDs were <0.1 for all matched variables, indicating adequate covariate balance.

Baseline characteristics stratified by 30-day mortality status are shown in Table S2.

Clinical outcomes

In-hospital clinical outcomes and healthcare resource utilization according to 24-h BUN change are presented in Table 2.

Table 2.

In-hospital clinical outcomes and healthcare resource utilization according to 24-h BUN change.

Outcomes	Original patients (n = 611)			Propensity score matching (n = 400)
	Non-increase BUN (n = 393)	Increase BUN (n = 218)	p-Value	Non-increase BUN (n = 200)	Increase BUN (n = 200)	p-Value
In-hospital mortality, n (%)	19 (4.8)	26 (11.9)	0.001	10 (5.0)	23 (11.5)	0.018
30-day mortality, n (%)	29 (7.4)	36 (16.5)	<0.001	12 (6.0)	33 (16.5)	<0.001
In-hospital rebleeding, n (%)	21 (5.3)	13 (6.0)	0.722	9 (4.5)	11 (5.5)	0.651
Endoscopic hemostasis, n (%)	98 (24.9)	65 (29.8)	0.191	35 (17.5)	54 (27.0)	0.021
Failed endoscopic intervention, n (%)	5 (0.5)	2 (0.9)	0.472	1 (0.5)	0 (0.0)	1.000
Need for RBC transfusion, n (%) a	277 (70.5)	155 (71.1)	0.872	129 (64.5)	141 (70.5)	0.197
Total RBC transfusion (units), median (IQR)	2 (0–3)	2 (0–4)	0.287	1.5 (0–3)	2 (0–4)	0.457
Length of stay (days), median (IQR)	4 (3–7)	5 (4–8)	<0.001	4 (3–7)	5 (4–8)	0.043
Additional intervention
Radiologic hemostasis, n (%)	3 (0.8)	3 (1.4)	0.688	1 (0.5)	2 (1.0)	1.000
Surgical hemostasis, n (%)	7 (1.8)	2 (0.9)	0.317	3 (1.5)	2 (1.0)	0.651

Data are expressed as number (%) unless otherwise specified.

^aPercentage calculated among patients with complete transfusion records.

BUN, blood urea nitrogen; IQR, interquartile ranges.

In the unmatched cohort, patients with a 24-h BUN increase had significantly higher 30-day mortality (16.5% vs 7.4%, p < 0.001) and in-hospital mortality (11.9% vs 4.8%, p = 0.001). Length of hospital stay was also longer in this group. Rates of rebleeding, RBC transfusion, and additional interventions were not significantly different between the two groups.

In the propensity score-matched cohort, a 24-h BUN increase was associated with higher 30-day mortality (16.5% vs 6.0%, p = 0.001) and in-hospital mortality (11.5% vs 5.0%, p = 0.018). Median length of stay remained longer among patients with a BUN increase. In contrast, no significant differences were observed in rebleeding, transfusion requirements, total RBC units transfused, or additional interventions after matching.

Multivariable analyses

Results of multivariable regression analyses are shown in Table 3.

Table 3.

Multivariable regression analysis of a 24-h BUN increase and clinical outcomes.

Outcome	Original cohort (n = 611) adjusted effecta,b	95% CI	p Value	PSM cohort (n = 400) adjusted effecta,b	95% CI	p Value
Primary outcome
30-Day mortality	2.549	1.458–4.456	0.001	3.307	1.604–6.819	0.001
Secondary outcomes
In-hospital mortality	2.620	1.354–5.071	0.004	2.732	1.208–6.178	0.016
In-hospital rebleeding	0.945	0.434–2.059	0.887	1.218	0.479–3.102	0.679
Any RBC transfusion	1.902	1.163–3.111	0.010	1.536	0.883–2.673	0.129
Additional intervention: radiologic	1.263	0.204–7.822	0.802	2.173	0.181–26.163	0.541
Additional intervention: surgical	0.420	0.079–2.237	0.309	0.515	0.076–3.485	0.496
Continuous outcomes
Total RBC units transfused (β)	0.541	0.057–1.025	0.028	0.366	−0.159 to 0.890	0.171
Length of stay, days (β)	0.951	−0.374 to 2.276	0.159	0.937	−0.517 to 2.391	0.206

Total RBC transfusion and length of stay were analyzed as a continuous variable (units and days, respectively). The magnitude and direction of the association between a 24-h BUN increase and 30-day mortality were consistent across matched and unmatched cohorts. Effect estimates remained stable in prespecified sensitivity analyses incorporating alternative bleeding severity metrics, granular hemodynamic adjustment, CKD exclusion, and early transfusion adjustment (Tables S4 and S5). CIs were comparable across models, suggesting no substantial loss of precision.

^aBinary outcomes are reported as aORs; continuous outcomes are reported as regression coefficients (β).

^bAdjusted for age, sex, hospital center, shock at presentation, fresh blood via NG, serum albumin, serum creatinine, International normalized ratio, and GBS.

aORs, adjusted odds ratios; BUN, blood urea nitrogen; CI, confidence interval; CKD, chronic kidney disease; GBS, Glasgow–Blatchford Score; PSM, propensity score matching; RBC, red blood cell.

In the original cohort, a 24-h BUN increase was independently associated with higher 30-day mortality (aOR, 2.549; 95% CI, 1.458–4.456; p = 0.001) and in-hospital mortality (aOR, 2.620; 95% CI, 1.345–5.071; p = 0.004). No independent associations were observed with rebleeding, RBC transfusion, transfusion volume, length of stay, or additional interventions.

In the propensity score-matched cohort, the association persisted for 30-day mortality (aOR, 3.307; 95% CI, 1.604–6.819; p = 0.001) and in-hospital mortality (aOR, 2.732; 95% CI, 1.208–6.178; p = 0.016). Associations with secondary non-mortality outcomes remained non-significant after adjustment.

No evidence of problematic multicollinearity was observed (VIF range 1.01–1.48 in the original cohort and 1.02–1.50 in the matched cohort; Table S3). For the primary endpoint (30-day mortality), the events-per-variable ratio was 5.0 in the matched cohort and 7.1 in the original cohort.

Sensitivity analyses

Prespecified sensitivity analyses addressing bleeding severity, hemodynamic instability, renal dysfunction, and early transfusion exposure are presented in Tables S4 and S5. In the matched cohort, the association between a 24-h BUN increase and both 30-day and in-hospital mortality remained consistent after:

• (i) replacing GBS with the complete Rockall score,

• (ii) modeling systolic blood pressure and heart rate as continuous variables,

• (iii) adjusting for and excluding chronic kidney disease, and

• (iv) adjusting for RBC transfusion within 24 h.

Similar results were observed in the original cohort, supporting the robustness of the findings.

Discussion

In this dual-center retrospective cohort study of patients with acute NVUGIB, an increase in BUN within the first 24 h of hospitalization was independently associated with higher short-term mortality. This association remained significant after multivariable adjustment and PSM. In contrast, early BUN elevation was not independently associated with rebleeding, transfusion requirements, or the need for radiologic or surgical intervention.

These findings suggest that early BUN kinetics may reflect evolving physiological stress not fully captured by admission laboratory values alone. While baseline BUN is incorporated into established risk scores, such tools rely exclusively on presentation data and do not account for temporal laboratory changes during hospitalization. Although we did not formally evaluate discrimination metrics, the persistence of the mortality association after adjustment for the GBS indicates that early BUN change may carry information beyond static baseline assessment. However, whether the incorporation of dynamic BUN measurements improves predictive performance requires dedicated model-comparison studies.

The observed association likely reflects pathophysiological processes extending beyond baseline renal function. Despite adjustment for serum creatinine, hemodynamic instability, and bleeding severity markers, the mortality signal persisted. Rising BUN levels during hospitalization may result from prerenal azotemia due to hypovolemia, ongoing renal hypoperfusion, gastrointestinal absorption of nitrogenous compounds from digested blood, or systemic catabolic stress. These mechanisms provide biological plausibility for the association, although causal inference cannot be established in this observational design.

Notably, unlike serum creatinine—which was balanced after matching and included in multivariable models—BUN remained independently associated with mortality. This distinction suggests that BUN kinetics may capture bleeding-related or stress-mediated physiological perturbations not fully explained by conventional markers of renal function. Furthermore, the absence of a significant association between early BUN increase and rebleeding or transfusion burden indicates that rising BUN at 24 h does not necessarily represent persistent hemorrhage. Instead, it may reflect a broader vulnerability to adverse clinical outcomes.

Prior investigations have explored the role of BUN in gastrointestinal bleeding. Kumar et al. ¹⁶ reported that a 24-h increase in BUN was associated with worse outcomes including mortality and prolonged hospitalization, in acute NVUGIB. Similarly, Tomizawa et al. ¹⁷ demonstrated higher BUN levels in upper compared with lower gastrointestinal bleeding, supporting its diagnostic utility. The present study extends these observations by evaluating a larger dual-center cohort and applying rigorous adjustment using both multivariable regression and PSM. The consistency of findings across analytic approaches strengthens the interpretation of early BUN increase as an independent risk marker rather than a simple surrogate of baseline severity.

From a practical perspective, BUN is inexpensive and routinely measured in patients admitted with gastrointestinal bleeding. Monitoring changes during the first 24 h does not require additional testing or resource allocation. Patients exhibiting a rising BUN trajectory may warrant closer observation, particularly in the absence of overt rebleeding. Nevertheless, BUN kinetics should be interpreted as complementary to, rather than a replacement for, established clinical assessment and guideline-directed management.^3–5

This study has several strengths, including its dual-center design, restriction to endoscopically confirmed NVUGIB, and comprehensive adjustment for bleeding severity and renal function. The stability of effect estimates across matched and unmatched cohorts and across prespecified sensitivity analyses supports the robustness of the findings.

Several limitations merit consideration. First, the retrospective observational design precludes causal inference and limits conclusions to associative relationships. Although multivariable adjustment and PSM were applied, residual confounding from unmeasured or incompletely measured variables cannot be excluded. Second, because classification of 24-h BUN change required survival to the measurement window, the study effectively employed a 24-h landmark approach. Although only four patients died within 24 h and were excluded, the possibility of survivor bias or immortal time bias cannot be entirely eliminated. In addition, a substantial proportion of initially screened patients did not have a 24-h BUN measurement and were therefore excluded from the analytic cohort. Because repeat laboratory testing may reflect clinical practice patterns or patient stability, this could introduce selection bias, and the analytic cohort may not fully represent all patients presenting with NVUGIB. Third, detailed information regarding fluid resuscitation volume, diuretic exposure, nutritional status, and neurohormonal or inflammatory markers was unavailable, limiting the mechanistic interpretation of BUN dynamics. Fourth, although multicollinearity was formally assessed and no significant concerns were identified, model misspecification and overfitting remain potential risks in multivariable analyses with limited event counts. Finally, while the dual-center design enhances generalizability within similar healthcare settings, the findings have not undergone external validation in independent populations. Prospective multicenter studies are warranted to confirm reproducibility and generalizability.

Conclusion

In conclusion, an increase in BUN within the first 24 h of hospitalization is independently associated with higher short-term mortality in patients with acute NVUGIB. Early laboratory trajectories may provide clinically relevant information beyond baseline measurements. Prospective studies are warranted to determine whether incorporation of dynamic laboratory assessment can enhance risk stratification and clinical decision-making in acute upper gastrointestinal bleeding.