Prognostic value of blood urea nitrogen to serum albumin ratio for acute kidney injury and in-hospital mortality in intensive care unit patients with intracerebral haemorrhage: a retrospective cohort study using the MIMIC-IV database

Abstract

Objective

We sought to evaluate the prognostic ability of blood urea nitrogen to serum albumin ratio (BAR) for acute kidney injury (AKI) and in-hospital mortality in patients with intracerebral haemorrhage (ICH) in intensive care unit (ICU).

Design

A retrospective cohort study using propensity score matching.

Setting

ICU of Beth Israel Deaconess Medical Center.

Participants

The data of patients with ICH were obtained from the Medical Information Mart for Intensive Care IV (V.1.0) database. A total of 1510 patients with ICH were enrolled in our study.

Main outcome and measure

The optimal threshold value of BAR is determined by the means of X-tile software (V.3.6.1) and the crude cohort was categorised into two groups on the foundation of the optimal cut-off BAR (6.0 mg/g). Propensity score matching and inverse probability of treatment weighting were performed to control for confounders. The predictive performance of BAR for AKI was tested using univariate and multivariate logistic regression analyses. Multivariate Cox regression analysis was used to investigate the association between BAR and in-hospital mortality.

Results

The optimal cut-off value for BAR was 6.0 mg/g. After matching, multivariate logistic analysis showed that the high-BAR group had a significantly higher risk of AKI (OR, 2.60; 95% confidence index, 95% CI, 1.86 to 3.65, p<0.001). What’s more, a higher BAR was also an independent risk factor for in-hospital mortality (HR, 2.84; 95% confidence index, 95% CI, 1.96 to 4.14, p<0.001) in terms of multivariate Cox regression analysis. These findings were further demonstrated in the validation cohort.

Conclusions

BAR is a promising and easily available biomarker that could serve as a prognostic predictor of AKI and in-hospital mortality in patients with ICH in the ICU.

Strengths and limitations of this study.

• Propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) were performed to minimise the bias of confounding factors and facilitate the comparability between groups.

• In clinic, serum albumin ratio is an easily available biomarker.

• This is a retrospective cohort study, the data generated during routine clinical visits.

• Potential bias may not fully eliminate and undetected confounding factors might exist although subgroup analyses, as well as adjusted, IPTW, and PSM models reached consistent results.

• Multicentre studies are needed to confirm our findings.

Introduction

Stroke is a major cause of death and disability worldwide. In the near future, stroke will continue to be one of the three major causes of death globally.¹ Most studies still predominantly focus on common neurological complications after stroke, while ignoring non-neurological complications, such as acute kidney injury (AKI), which occurs in 11.6–31% of patients after stroke.^{2 3}Whether haemorrhagic or ischaemic stroke, AKI is a powerful predictor of worse outcomes. AKI is closely associated with an increased rate of moderate-to-severe disability, new composite cardiovascular events, average hospitalisation expenses and prolonged hospitalisation. Previous studies have shown that dialysis is required in 0.6% of patients.^{4 5} Therefore, it is necessary to lay more emphasis on the occurrence of AKI for the patients who suffer from stroke. Recently, a non-invasive and easy-to-access prognosis tool based on blood urea nitrogen (BUN) to serum albumin ratio (BAR) has been proposed, which is defined as the quotient between BUN and albumin. BAR has been investigated as a useful predictor for poor outcomes and intensive care in patients with conditions including gastrointestinal bleeding, Escherichia coli bacteraemia and so on.^6–8 Because of its utility, BAR was also suggested as a strong target for predicting the crucial illness in patients with contagious coronavirus disease 2019 (COVID-19).⁹ Of the commonly used laboratory biomarkers, an observational study concluded that decompensated cirrhosis with AKI had significantly lower serum albumin and higher BUN levels than those without AKI.¹⁰ In addition, compared with patients with mild COVID-19, patients that were diagnosed with critical illness or those who did not survive usually have lower albumin levels but higher BUN levels.¹¹ However, the importance of combining the BUN-to-BAR has not been assessed simultaneously, and the potential association between BAR and AKI in patients with intracerebral haemorrhage (ICH) is still unclear.

Therefore, the purpose of our study was to evaluate the incidence of AKI after cerebral haemorrhage and determine the feasibility of using BAR to predict the occurrence of AKI and other adverse outcomes.

Material and methods

Source of data

As a retrospective research, We obtained data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database,¹² a freely accessible intensive care unit (ICU) data resource which contains overall and high-quality data with regard to over 70 000 patients received by the ICU of Beth Israel Deaconess Medical Center in Boston, Massachusetts, USA, during the period from 2008 to 2019 and it is distinguished on the basis of the Health Insurance Portability and Accountability Act Safe Harbour provision.

Study population and definition

Patients diagnosed with ICH based on the 9th and 10th revision of the International Classification of Disease code were included in this study. The inclusion criteria were as follows: (1) first admission of the patients to the ICU and hospital; (2) patients who stayed in the hospital for longer than 48 hours; and (3) patients without end-stage renal disease. In addition, patients with missing serum albumin and BUN on the first day of admission to the ICU were also excluded from the final cohort.

Kidney Disease Improving Global Outcomes (KDIGO) 2012 creatinine criteria were used to define and stage the severity of AKI. Chronic kidney disease (CKD) refers to the abnormal kidney function which lasts longer than 3 months. Indicators consist of albuminuria, urine sediment abnormalities, acid-base derangements and so on. End-stage renal refers to the glomerular filtration rate which is lower than 15 mL/min.¹³

Data collection and outcomes

Baseline characteristics and admission information: demographic characteristics, basic vital signs, comorbidities, laboratory indicators and scoring systems. Demographic characteristics included age, sex and weight. Basic vital signals included mean arterial pressure (MAP), heart rate, respiratory rate (RR) and saturation of percutaneous oxygen. Comorbidities included hypertension, diabetes, myocardial infarction, atrial fibrillation, peripheral vascular disease and hyperlipidaemia and so on. Laboratory indicators included white blood cell count (WBC), haemoglobin (HGB), platelet (PLT), bilirubin, anion gap, bicarbonate, chloride, creatinine, glucose, potassium and sodium. Scoring systems included Sequential Organ Failure Assessment Score (SOFA), Acute Physiology Score III (APSIII) and Simplified Acute Physiology Score II (SAPSII). According to published recommendations,^14–17 all the above scores were calculated with clinical information (including Glasgow Coma Score, hypotension, PLT, serum bilirubin, serum bilirubin, serum creatinine and oxygenation). In addition, BAR (mg/g) was calculated using the initial serum BUN (mg/dL)/serum albumin (g/dL).

The primary outcome of the study was AKI, and the secondary outcome was in-hospital mortality.

Statistical analysis

Continuous variables were summarised as mean±SD, and categorical covariates were presented as frequencies (percentages). The crude cohort was categorised into two groups on based on the optimal cut-off BAR (low-BAR group <6.0 mg/g and high-BAR group ≥6 mg/g). The clinical features between the two groups were analysed using either Student’s t-test or χ² test. In order to adjust the imbalance of the covariates between the two groups, propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) were conducted to adjust the imbalance of the covariates between the two groups. Patients were matched 1:1 using the nearest-neighbour algorithm. The standardised mean difference (SMD) between the low and high-BAR groups was calculated. If the SMD of a variable is greater than 0.1, it can be considered unbalanced.¹⁸ We calculated the adjusted ORs for AKI using multivariable logistic regression. Multivariate Cox regression and adjusted HR for in-hospital mortality were also conducted in the original, matched and weighted cohorts. Kaplan-Meier curves, waterfall plots and heatmaps were generated and compared between the different cohorts. The forest plot was applied to describe the results of subgroup analysis for AKI and in-hospital mortality. What’s more, decision curve analysis (DCA) was performed to evaluate the potential clinical usefulness of BAR. Finally, the receiver operating characteristics curve was used to evaluate the predictive performance of BAR for AKI and in-hospital mortality. All statistical investigation were performed using the X-tile (V.3.6.1) and R software (V.4.1.0). Statistical significance was set at p<0.05.

Patient and public involvement

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Results

Baseline patient characteristics

A total of 1510 patients were enrolled in the original cohort (figure 1). The best cut-off value for BAR, determined using X-tile software, was 6.0 mg/g. The included patients were separated into two groups on the basis of the measurements of BAR, and the baseline characteristics of the research population are shown in online supplemental table 1. Before PSM, 31 of 37 covariates were imbalanced in the original cohort, including SAPSII, APSIII, SOFA, CKD, OASIS (Oxford acute severity of illness score), creatinine, Charlson Index, HGB, mechvent, vasopressors, hypertension, bicarbonate, congestive heart failure, Glasgow Coma Scale (GCS), diabetes, glucose, heart rate, chloride, age, RR, PT (Prothrombin time), INR (International standard ratio), total bilirubin, PLTs, anion gap, potassium, myocardial infarct, MAP, sex, WBC and sodium. After PSM and IPTW, the SMD all turned out to be less than 0.1, indicating that the baseline variables in the matched cohort and weighted cohort have similar distributions (online supplemental figure 1). Otherwise, the relationship between BAR and severity scores, clinical factors and clinical outcomes were described in the original cohort and in the matched cohort (online supplemental figure 2).

Figure 1.

The flowchart of the study. BUN, blood urea nitrogen; ESRD, end-stage renal disease; ICD-9/10, 9th and 10th revision International Classification of Disease; ICU, intensive care unit; IPTW, inverse probability of treatment weighting; MIMIC-IV, Medical Information Mart for Intensive Care IV; PSM, propensity score matching.

BAR as a predictor of the primary end point

Compared with patients in the low-BAR group, those in the high-BAR group were more likely to develop AKI in the original cohort and the matched cohort. Our study showed that the incidence of stage I, stage II and stage III AKI in the high-BAR group was 20.8%, 37.3% and 11.3%, respectively, which was higher than that in the low-BAR group (online supplemental table 1). Univariate analysis indicated that high BAR was associated with increased AKI, with a crude OR of 3.65 (95% confidence index, 95% CI, 2.93 to 4.57, p<0.001) (table 1). The results remained robust after the PSM and IPTW. Multivariate analysis revealed that high BAR was still an independent predictive factor of AKI in the original cohort (OR, 2.69, 95% confidence index, 95% CI, 2.05 to 3.55, p<0.001), matched cohort (OR, 2.60, 95% confidence index, 95% CI, 1.86 to 3.65, p<0.001) and weighted cohort (OR, 2.78, 95% confidence index, 95% CI, 2.09 to 3.71, p<0.001) when adjust for age, sex, weight, comorbidities, score system, interventions, Charlson Index, vital signs and laboratory results (table 1).

Table 1.

Univariate and multivariate logistic analysis of serum albumin ratio for the development of acute kidney injury

	Original cohort		Matched cohort		Weighted cohort
	OR (95% CI)	P value	OR (95% CI)	P value	OR (95% CI)	P value
Unadjusted	3.65 (2.93 to 4.57)	<0.001	2.22 (1.64 to 2.99)	<0.001	2.40 (1.85 to 3.12)	<0.001
Model 1	3.68 (2.93 to 4.64)	<0.001	2.24 (1.65 to 3.05)	<0.001	2.45 (1.89 to 3.22)	<0.001
Model 2	3.75 (2.95 to 4.78)	<0.001	2.25 (1.65 to 3.07)	<0.001	2.45 (1.87 to 3.21)	<0.001
Model 3	2.66 (2.04 to 3.47)	<0.001	2.47 (1.78 to 3.44)	<0.001	2.71 (2.04 to 3.61)	<0.001
Model 4	2.69 (2.05 to 3.55)	<0.001	2.60 (1.86 to 3.65)	<0.001	2.78 (2.09 to 3.71)	<0.001

Model 1 adjusted for age, gender, weight. Model 2 adjusted for model one plus comorbidities. Model 3 adjusted for model 2 plus score system, interventions and Charlson Index. Model 4 adjusted for model 3 plus vital signs and laboratory results.

BAR value as a predictor of secondary outcome

As is shown in figure 2, patients in the original cohort were categorised into two groups (high-BAR group and low-BAR group) (figure 2A). The survival status of the two groups was shown in figure 2B. The relationship between the severity scores, clinical factors and BAR in two groups was shown in figure 2C. Our study showed that BAR also has a good accuracy in predicting in-hospital mortality of patients in the original cohort (figure 2D). The DCA demonstrated that as a predictor, BAR owned more net benefit compared with the ‘treat all’ or ‘treat none’ strategies in the original cohort (figure 2E), indicating that BAR could be an effective and useful clinical predictor. Otherwise, the Kaplan-Meier curves revealed that in-hospital mortality was significantly higher in the high-BAR group compared with than in the low-BAR group in the original cohort (figure 2F). These findings were further conformed in the matched cohort (online supplemental figure 3). Univariate Cox regression analysis revealed that the risk of in-hospital mortality increased in the high-BAR group. Moreover, a higher BAR was still an independent risk factor for in-hospital mortality in the multivariate Cox regression analysis (table 2).

Figure 2.

Prognostic value of BAR for in-hospital mortality in the original cohort. APSIII, Acute Physiology Score III; BAR, serum albumin ratio; SAPSII, Simplified Acute Physiology Score II; SOFA, Sequential Organ Failure Assessment Score.

Table 2.

Univariate and multivariate Cox analysis of serum albumin ratio for in-hospital mortality

	Original cohort		Matched cohort		Weighted cohort
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Unadjusted	3.40 (2.62 to 4.41)	<0.001	2.13 (1.51 to 3.00)	<0.001	2.09 (1.43 to 3.06)	<0.001
Model 1	3.22 (2.48 to 4.20)	<0.001	2.23 (1.57 to 3.17)	<0.001	2.13 (1.46 to 3.11)	<0.001
Model 2	3.30 (2.52 to 4.33)	<0.001	2.27 (1.60 to 3.24)	<0.001	2.17 (1.50 to 3.14)	<0.001
Model 3	2.37 (1.79 to 3.15)	<0.001	2.56 (1.79 to 3.67)	<0.001	2.52 (1.83 to 3.48)	<0.001
Model 4	2.40 (1.80 to 3.20)	<0.001	2.84 (1.96 to 4.14)	<0.001	2.61 (1.90 to 3.59)	<0.001

Model 1 adjusted for age, gender, weight. Model 2 adjusted for model 1 plus comorbidities. Model 3 adjusted for model 2 plus score system, interventions and Charlson Index. Model 4 adjusted for model 3 plus vital signs and laboratory results.

AUC, Area under the receiver-operating characteristic curve; CCI, Charlson Comorbidity Index; GCS, Glasgow Coma Scale; OASIS, Oxford acute severity of illness score.

Subgroup analysis and additional analyses

Based on age, sex and comorbidities, subgroup analyses were performed in the original and matched cohorts. The forest plot demonstrated that BAR was still an independent predictive factor in most prespecified subgroups (figure 3 and online supplemental figure 4). In addition, the model performance was assessed based on Integrated Discrimination Improvement (IDI), Net reclassification index (NRI) as well as Area under the receiver-operating characteristic curve (AUC) by adding BAR to a clinical model composed of age, sex, weight, comorbidities, interventions, vital signs and laboratory results, to develop a combined biomarker and clinical model (online supplemental table 2). The results showed that the addition of BAR to these models improved the prediction of AKI and in-hospital mortality; reclassification adding BAR to the unitised biomarker and clinical model showed a significant increase in AUC, IDI and NRI. Replacing BAR with the SAPSII score indicated low prognostic precision for AKI (AUC: 0.612). Moreover, combining the BAR and SAPSII scores to these models also showed good prognostic accuracy for AKI and in-hospital mortality (p<0.001) (online supplemental table 2).

Figure 3.

Subgroup analysis in the original cohort. AKI, acute kidney injury; BAR, serum albumin ratio; CHF, congestive heart failure; PVD, peripheral vascular disease.

Discussion

In this research, the overall incidence of AKI and in-hospital mortality among patients with ICH were 50% and 17.4%, respectively. The results revealed that BAR is an independent predictor of AKI in patients with cerebral haemorrhage. We further demonstrated that BAR level was significantly associated with in-hospital mortality. In addition, according to the DCA, when the threshold probability for a patient was within the range of 0–100%, the BAR added more net benefit than the ‘treat all’ or ‘treat none’ strategies. Hence, those results suggested that the BAR, as an easily accessible indicator, might be a good predictor for identifying patients at high risk of AKI and in-hospital mortality in patients with ICH in ICU.

There were differences in the incidence of AKI between our study and previous studies. Similarly, mortality in our study differs from others, which may be related to the severity of disease, the type of stroke and the period studied. According to the KDIGO criteria, Yang et al reported in a prospective observation research of 164 patients with aneurysmal subarachnoid haemorrhage, AKI and in-hospital mortality occurred in 17 patients (10.4%).¹⁹ Moreover, by using the Acute Kidney Injury Network criteria²⁰ to define AKI, Qureshi et al recently reported that five subjects (9%) developed stage I AKI but none of them had stage II or III AKI.²¹Tao et al found that contrast-induced AKI (CI-AKI), defined by the diagnostic criteria of CI-AKI,²² was observed in 85 (4.2%) of 2015 patients with stroke who underwent cerebral angiography. The in-hospital mortality rate for this cohort was 6.3%, and the 1-year mortality rate was 19.5%.²³ Interestingly, compared with ischaemic stroke, patients with haemorrhagic stroke are more likely to develop AKI. Covic et al²⁴ retrospectively analysed 1090 patients with stroke for the first time in the emergency department and found that the incidence of AKI for patients who suffered from ischaemic and haemorrhagic stroke was 13.2% and 22.2%, respectively. Regardless of the in-hospital mortality and the incidence of AKI, our results are relatively high, which may be related to patients with our haemorrhagic stroke coming from the ICU.

BAR was an effective predictor of AKI in this study. In the clinic, serum creatinine or urine volume is often used to diagnose AKI. In fact, serum creatinine is not extremely reliable for defining AKI due to some confounding factors, such as sex, age and other factors. Previous studies have suggested that kidney damage may significantly precede an increase in serum creatinine levels.^25–27 In addition, AKI is closely associated with poor outcomes in patients with stroke. Therefore, it is important to identify a highly economical and easily measurable marker for predicting AKI to mitigate further risks. A retrospective study showed that AKI was associated with significantly higher BUN levels and lower serum albumin levels in paediatric liver transplantation patients in the ICU.²⁸ Similarly, Cheng et al²⁹ found that higher serum albumin was closely correlated with failure to define AKI in a timely manner, whereas higher BUN was associated with earlier diagnosis. A Korean study on 328 hyperuricaemia patients treated with non-steroidal anti-inflammatory drugs (NSAIDs revealed that low serum albumin level is an important hazard factor for AKI, and that, NSAIDs should be carefully used in hyperuricaemia patients with low serum albumin levels.³⁰

In the present study, BAR effectively predicted in-hospital mortality. In the literature, the positive relation between an elevated BAR level and mortality has been shown in multiple acute diseases, including E. coli bacteraemia, gastrointestinal bleeding, aspiration pneumonia, Fournier’s gangrene and acute pulmonary embolism.7 8 31–33 BAR was confirmed to be a more reliable and powerful predictor than albumin and BUN levels, which increased in-hospital mortality among patients with COVID-19.³⁴ In addition, it was reported that an increased BAR at treatment commencement was closely inter-related to 30-day death rate in patients who are not affected with HIV infection.³⁵Our research indicated that the incidence of in-hospital mortality in the BAR≥6.0 mg/g group was obviously higher than that of BAR<6.0 mg/g group.

Patients with serious illnesses often suffer from hypoalbuminaemia and renal hypoperfusion, leading to increased BUN levels and BAR. It will be easier for clinicians to comprehend the essence of a high-BAR ratio. However, the potential pathophysiological mechanisms and possible reasons for the prognostic values of BAR remain unclear. It has been reported that an increased BUN can predict the death rate in terminally ill patients who are independent of ‘normal’ creatinine.³⁶ Meanwhile, hypoalbuminaemia is a strong predictive factor of 30-day all-cause death rate with regard to intensely admitted patients in an observational cohort research.³⁷

BUN is a nitrogen-containing compound. In the ordinary humdrum case, BUN is mainly filtered by the glomerulus and excreted in the urine. When glomerular filtration function decreases, BUN concentration increases. Therefore, the determination of BUN can be used to estimate the glomerular filtration function. BUN is also a useful biomarker which can obviously reveal the complex correlation among nutritional condition, hypovolaemia and protein metabolism.^{7 32} Serum albumin can prevent kidney damage via several pathways. As an important antioxidant in plasma, albumin can inhibit the apoptosis of renal tubular cells and enhance propagation by scavenging oxygen radicals.^38–40 Albumin also improves glomerular filtration and renal perfusion by prolonging potent renal vasodilation.⁴¹ Albumin is not only a nutritional marker, but also an anti-inflammatory marker. Zhang and Frei⁴² found that albumin selectively inhibits tumour necrosis factor alpha-inducible expression of vascular cell adhesion molecule-1 and the activation of nuclear factor kB and monocyte adhesion in endothelial cells of human being. These results also suggest that a low serum albumin level may have a significant effect on endothelial dysfunction. As albumin synthesis decreases and catabolism increases, the inflammatory state in the body also increases.⁴³In addition to being an antioxidant and anti-inflammatory molecule, albumin also plays other important parts in the body, such as regulating immune function, and adjusting acid-base balance and vascular permeability.⁴⁴ Thus, the possible underlying mechanism of BAR in forecasting the occurrence of AKI and mortality in patients who are affected with cerebral haemorrhage might be as follows: First, the central autonomic network affects renal haemodynamics and excretion. Second, the nutritional status of the body greatly decreased. Third, oxidative stress and inflammation status increase and the ability to scavenge of oxygen radical and inflammatory mediators decrease.⁴⁵

There were some limitations to our study. First, this research was a retrospective study, we were unable to avoid selection bias. However, we strictly set the inclusion criteria in order to reflect actual conditions as accurately as possible. Second, the clinical data for our study was obtained from the MIMIC-IV database, data were collected from patient medical records and we relied on the accuracy of the records, therefore, prospective studies are necessary to conduct in the future. Third, although the PSM and IPTW were used to control for potential confounders, there might exist some other residual confounders, such as different drugs, albuminuria, contrast agents, cerebral haemorrhage volume and surgical procedures, which would not be measured in this study. Fourth, the BAR may only help to recognise critical clinical situations quickly, and does not provide any more information about potential life-threatening pathophysiological mechanisms. Finally, it is necessary to establish longer follow-up data because of the relatively short follow-up time of the original cohort.

Conclusion

A higher level of BAR is associated with an increased risk of AKI and in-hospital mortality, and it could serve as a prognostic predictor for AKI and in-hospital mortality in patients with cerebral haemorrhage in the ICU.

Data availability statement

Data are available upon reasonable request. The data used in this study can be obtained by the corresponding author upon request.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

This study obtained the approval from the Institutional Review Boards of MIT and BIDMC. Informed consent was exempted due to the anonymous data.