Chronic Kidney Disease and End-Stage Renal Disease Predict Higher Risk of Mortality in Patients with Primary Upper Gastrointestinal Bleeding

Abstract

Background

The outcome of gastrointestinal bleeding in chronic kidney disease (CKD) and end-stage renal disease (ESRD) patients is difficult to discern from the literature. Many publications are small, single-center series or are from an era prior to advanced interventional endoscopy, widespread use of proton pump inhibitors or treatment for Helicobacter pylori infections. In this study, we quantify the role of CKD and ESRD as independent predictors of mortality in patients admitted to the hospital with a principal diagnosis of primary upper gastrointestinal bleeding (UGIB).

Methods

We used the Nationwide Inpatient Sample that contains data on approximately 8 million admissions in 1,000 hospitals chosen to approximate a 20* stratified sample of all US facilities. Patients discharged with the principal diagnosis of primary UGIB, CKD or ESRD were identified through the ninth revision of the International Classification of Diseases, clinical modification (ICD-9-CM) codes. The outcome variables included frequency and in-hospital mortality of UGIB in CKD and ESRD patients as compared to non-CKD patients and were analyzed using logistic regression modeling.

Results

In 2007, out of a total of 398,213 admissions with a diagnosis of primary UGIB, 35,985 were in CKD, 14,983 in ESRD, and 347,245 in non-renal disease groups. The OR for primary UGIB hospitalization in CKD and ESRD was 1.30 (95* CI 1.17–1.46) and 1.84 (95* CI 1.61–2.09), respectively. The corresponding all-cause mortality OR was 1.47 (95* CI 1.21–1.78) and 3.02 (95* CI 2.23–4.1), respectively.

Conclusion

Patients with CKD or ESRD admitted with primary UGIB have up to three times higher risk of all-cause in-hospital mortality, warranting heightened vigilance by their clinicians.

Introduction

Multiple risk factors have been proposed to explain the observation that patients with chronic kidney disease (CKD) have an increased risk for gastrointestinal (GI) bleeding, and these include the severity of uremia, platelet dysfunction and medication adverse effects. Autopsy studies of uremic patients confirmed that GI problems were common, with two thirds showing edema, congestion or hemorrhagic ulcerations [1]. Of all the causes for blood loss in CKD patients, one of the most common sites was the upper GI tract [2]. Specifically, reports of endoscopic imaging for acute upper GI bleeding (UGIB) in CKD and end-stage renal disease (ESRD) patients have confirmed the high frequency of peptic ulcer disease (gastric and duodenal) as well as emphasized the importance of erosive gastritis and esophagitis. The impact of UGIB on patient outcomes, however, has not been fully investigated across large patient populations in the modern era of advanced therapeutic endoscopy techniques and potent antacid medications (i.e. proton pump inhibitors). Much of the literature on this issue is from prior decades, small, single-center series and has involved acute kidney injury with multi-organ failure and not CKD/ESRD. Recently, Cheung et al. [3] showed ESRD to be a predictor of worse outcomes in GI bleeding associated with peptic ulcer disease, a higher risk of rebleeding, need for multiple transfusions and longer hospital stays when compared to CKD and non-renal disease groups. Studies that proposed an increased risk of (or trend for) death associated with a UGIB were similarly limited by trial design or size. In this study, we thus utilized a large national database of millions of admissions to explore the hypothesis that CKD and ESRD are independent predictors of mortality for patients hospitalized with a principal diagnosis of primary UGIB.

Methods

Data Source

We used the Healthcare Cost and Utilization Project – Nationwide Inpatient Sample (NIS), the largest all-payer inpatient care database publicly available in the US. This is an administrative dataset created by the Agency for Healthcare Research and Quality and contains data on approximately 8 million hospitalizations in 1,000 hospitals that were chosen so as to approximate a 20* stratified sample of all US community hospitals. Both hospital and discharge weights are provided in the NIS database which are then used to extrapolate national estimates. Each hospitalization is treated as an individual entry in the database and is coded with one principal diagnosis, up to 14 secondary diagnoses, and 15 procedural diagnoses associated with that stay. NIS encompasses information on all patients regardless of payer and includes patient information regardless of hospital type: teaching or nonteaching, rural or urban, large or small volume, and private or publicly owned. Data from the year 2007, which was the last year where full information was available on all variables, were used in this study.

Study Population

We used the ninth revision of the International Classification of Diseases, clinical modification codes (ICD-9-CM) to identify adult patients (aged ≥18 years) admitted with a primary diagnosis of UGIB (ICD-9-CM codes 531–534, 578, and 530). Similarly, we used ICD-9-CM codes to identify the primary predictors of outcome which in this study were CKD (ICD-9-CM code 585.9) and ESRD (ICD-9-CM code 585.6). We also used ICD-9-CM codes to identify independent predictors (covariates), which included history of cancer, anemia, atrial fibrillation, cirrhosis, hypertension, diabetes mellitus, congestive heart failure (CHF), coronary artery disease (CAD), stroke, thrombocytopenia, coagulation defect, alcoholism, smoking, venous thromboembolism (VTE), severe sepsis, general surgery or orthopedic surgery. In the appendix, we provide a complete list of ICD-9-CM codes used.

Outcomes

Our primary outcomes of interest were the frequency of admissions with the primary diagnosis of UGIB in CKD and ESRD patients and the all-cause in-hospital mortality in these two population groups as compared to non-CKD patients. We also analyzed the effect of comorbid conditions on outcomes of primary UGIB and all-cause mortality.

Statistical Analysis

Statistical analysis was performed using Stata IC 11.0 (Stata-Corp, College Station, Tex., USA). Univariate analysis was performed using a χ² test for categorical variables and a t test for continuous variables utilizing the same end points (outcomes) as used in multivariate analysis.

We assessed the risk factors for primary UGIB hospitalizations and all-cause in-hospital mortality in such patients by multivariate logistic regression modeling. A total of four different models (separate models for CKD and ESRD groups for each primary outcome) were constructed for this purpose. Using the entire cohort, we compared primary UGIB hospitalizations of non-renal patients with admissions of individuals with CKD (model 1) or ESRD (model 2). Next, we assessed the risk of all-cause in-hospital mortality in patients admitted with primary UGIB separately in the CKD group (model 3) and the ESRD group (model 4) and compared each with the non-renal disease group. The variables found significant on univariate analysis (p < 0.05), with the outcome of interest, were included in the final four models. Interaction terms were generated to account for interaction between independent predictors and only those with a p value <0.05 were included in the final models.

Results

Based on the NIS dataset, utilizing strata weights provided, national estimates were extrapolated using appropriate survey estimation commands. Accordingly, a total of 398,213 admissions of adults >18 years of age with a diagnosis of primary UGIB were identified. The demographic characteristics of the primary UGIB and non-bleeding patients are summarized in table 1. Of those with primary UGIB, there were 35,985 admissions for bleeding CKD patients (out of 1,799,785 total CKD hospitalizations), 14,983 for ESRD patients (out of 840,282 admissions), and the remaining 347,245 were for patients with no renal problems (out of 30,119,186 admissions). Put in a broader perspective, per 10,000 all-cause hospitalizations, there were 200 admissions for UGIB in CKD patients, 178 in the ESRD group, and 115 bleedings in those without reported kidney disease. In the non-renal disease and CKD groups, the commonest cause of UGIB was ‘unspecified’ (ICD-9-CM 578) followed by gastric ulcer (15 and 12.6*, respectively). In the ESRD group, gastritis (13.5*) followed by gastric ulcer (12.9*) were the commonest causes of UGIB. The demographic characterization of the patients revealed that those >65 years of age represented 84* of the CKD, 55* of the ESRD, and 50* of those patients with no renal dysfunction. A total of 55* of patients with UGIB in the CKD and ESRD groups were males (table 2). A majority (73*) of UGIB patients in the CKD group was Caucasian, while in the ESRD group 39* were African-Americans. The majority of those with UGIB in the CKD and ESRD groups were non-smokers and non-alcoholics. A total of 37 and 26* of CKD and ESRD patients, respectively, had some degree of anemia (acute or chronic or both) at the time of admission. In terms of their care, approximately 60* of UGIB patients in the CKD group were admitted to large bed size hospitals as compared to 69* of those with ESRD. A majority (>80*) of UGIB patients in each group were admitted to urban hospitals. A total of 41 and 38* of UGIB patients in the CKD and ESRD groups, respectively, had EGD performed during the hospital stay as compared to 37* in the non-renal group.

Table 1.

Demographic characteristics of primary UGIB patients compared with patients without primary UGIB

Demographic factors	No GIB* (n = 32,361,040)	Primary UGIB (n = 398,213)	p value
Age >65 years	13,022,682 (40*)	247,892 (62*)	<0.001
Male sex	12,580,455 (39*)	201,110 (50*)	<0.001
Race			<0.001
Caucasian	15,931,700 (68*)	204,728 (71*)
Afro-American	3,374,534 (14*)	40,052 (14*)
Hypertension	14,238,643 (44*)	218,762 (55*)	<0.001
Diabetes mellitus	6,826,714 (21*)	101,439 (25*)	<0.001
Alcohol intake	1,367,394 (4.2*)	35,153 (8.8*)	<0.001
Smoking	5,381,328 (17*)	62,377 (15.7*)	<0.001
Platelet abnormalities	755,341 (2.3*)	19,180 (4.8*)	<0.001
Cancer	4,250,872 (13*)	52,510 (13*)	0.31
Coagulation defect	333,131 (1*)	19,438 (4.8*)	<0.001
Packed red blood cell transfusion	2,070,614 (6.4*)	185,841(46.7*)	<0.001
Any major surgery	1,053,751 (3.2*)	3,544 (0.89*)	<0.001
CAD	1,273,080 (3.9*)	103,990 (26*)	<0.001
CVA	625,441 (1.9*)	3,478 (0.9*)	<0.001
Anemia	1,104,725 (3.4*)	121,336 (30.5*)	<0.001
VTE	664,455 (2*)	6,399 (1.6*)	<0.001
Atrial fibrillation/flutter	3,373,186 (10*)	64,034 (16*)	<0.001
Cirrhosis	587,258 (1.8*)	30,717 (7.7*)	<0.001
Severe sepsis	389,056 (1.2*)	2,079 (0.5*)	<0.001
DIC	33,312 (0.13*)	592 (0.15*)	<0.001
Peripheral vascular disease	1,281,208 (3.9*)	20,875 (5.2*)	<0.001
Shock	354,397 (1.1*)	17,246 (4.3*)	<0.001
Vasopressor use	65,003 (0.2*)	678 (0.17*)	0.051
Mechanical ventilation	860,336 (2.7*)	8,930 (2.3*)	<0.001
Esophagogastroduodenoscopy	369,369 (1.1*)	148,600 (37*)	<0.001
Location			<0.001
Rural	4,235,492 (13*)	63,894 (16*)
Urban	28,081,514 (87*)	333,473 (84*)

^*National estimates were extrapolated from the NIS database utilizing strata weights which are provided along with NIS dataset, using appropriate Stata survey estimation commands.

Table 2.

Distribution of demographic/risk factors in primary UGIB stratified by degree of renal function

Demographic factors	Primary UGIB (n = 398,213)
	normal renal function (n = 347,245)	CKD (n = 35,985)	ESRD (n = 14,983)
Age >65 years	209,629 (60*)	30,071 (83*)*	8,192 (54*)*
Male sex	173,215 (49.9*)	19,606 (54.5*)*	8,289 (55*)*
Race
White	181,615 (72*)	18,679 (73*)	4,433 (40*)
Black	31,876 (13*)	3,937 (15*)*	4,239 (38*)*
Insurance status
Medicare	210,732 (61*)	29,547 (82*)*	11,819 (79*)*
Medicaid	25,311 (7*)	1,502 (4*)*	1,302 (8.7*)*
Private	78,251 (22*)	3,930 (11*)*	1,479 (10*)*
Self pay	20,171 (6*)	480 (1.3*)*	166 (1.11*)*
No charge/others	13,727 (3.5*)	497 (1.4*)*	197 (1.21*)*
Hypertension	179,916 (52*)	25,207 (70*)*	13,639 (91*)*
Diabetes mellitus	80,268 (23*)	13,916 (38*)*	7,254 (48*)*
Alcohol intake	33,738 (9.8*)	1,130 (3*)*	285 (1.9*)*
Smoking	57,739 (17*)	3,285 (9*)*	1,335 (9*)*
Platelet abnormalities	16,230 (4.6*)	1,907 (5*)	1,043 (7*)*
Cancer	46,476 (13*)	4,619 (13*)*	1,415 (10*)*
Coagulation defect	16,458 (5*)	2,294 (6.3*)*	686 (4.5*)
CAD	85,200 (24*)	14,102 (39*)*	4,688 (26*)*
CVA	2,891 (0.8*)	41 (1.2*)*	146 (0.8*)
Anemia	104,115 (30*)	13,353 (37*)*	3,868 (26*)*
VTE	5,384 (1.5*)	669 (1.9*)	346 (2.3*)*
Atrial fibrillation	50,860 (15*)	10,877 (30*)*	2,298 (15*)
Cirrhosis	27,823 (8*)	1,907 (5*)*	986 (6.6*)*
Severe sepsis	1,589 (0.5*)	276 (0.80*)*	214 (1.5*)*
DIC	480 (0.14*)	53 (0.15*)	59 (0.40*)*
Peripheral vascular disease	16,140 (4.7*)	3,045 (8.5*)*	1,690 (11*)*
Shock	14,124 (4*)	2,346(6.5*)*	776 (5.2*)*
Mechanical ventilation	7,147 (2*)	1,114 (3.1*)*	669 (4.5*)*
Discharge destination
Home	239,524 (69*)	18,735 (52*)*	8,832 (67*)*
Hospital	8,328 (2.4*)	857 (2.4*)	9,478 (1.9*)*
SNF	57,805 (16*)	9,722 (27*)*	70,636 (21*)*
Home care	27,671 (8*)	4,935 (13.7*)*	34,296 (11*)*
AMA	5,439 (1.6*)	192 (0.5*)*	5,880 (1.7*)
Died	8,259 (2.4*)	1,506 (4.2*)*	10,561 (5.3*)*

^*p value <0.025 using χ² test (comparing CKD or ESRD with normal renal function). SNF = Skilled nursing facility; AMA = against medical advice.

A univariate analysis revealed age, sex, race, presence of platelet abnormalities, coagulation defect, hypertension, diabetes mellitus, major surgery, orthopedic surgery, anemia, cerebrovascular accident (CVA), CAD, CHF, VTE, atrial fibrillation, smoking, alcohol use, cirrhosis, cancer, severe sepsis and disseminated intravascular coagulation (DIC) to be significantly associated with primary UGIB for both CKD and ESRD groups. These variables were then utilized to construct the multivariate logistic models. Based on representation in this broad-based national database, primary UGIB hospitalization in patients with CKD was found to be 30* higher when compared to the non-renal disease group, after controlling for potentially confounding covariates and adjusting for interaction between the independent variables (OR 1.30, 95* CI 1.17–1.46, p < 0.001). Similarly, ESRD patients with primary UGIB were over represented in the database as their hospitalizations were 84* higher than those for patients without renal disease, after controlling for the covariates and adjusting for the independent variable interactions (OR 1.84, 95* CI 1.61–2.09, p < 0.001).

The strongest positive predictors of primary UGIB in both models were female sex, age >50 years, Asian race, presence of acute and chronic anemia, diagnosis of cirrhosis, presence of coagulation defects and alcohol use (all p < 0.001, table 3). The database, however, could not allow for an analysis of how the outpatient monitoring or treatment of acute or chronic concurrent conditions might have impacted the extent of bleeding or the decision for hospitalization. Thus, it was difficult to interpret the finding that the presence of CVA, cancer, VTE, CHF and surgery had a statistically significant independent negative association with admissions for primary UGIB (p < 0.001, table 3) in both renal disease models.

Table 3.

Summary of models 1 and 2 and the OR and 95* CI values for primary UGIB hospitalizations in CKD and ESRD patients compared to the non-CKD, non-ESRD group

Risk factors/interaction termsa	Primary UGIB hospitalizations/CKD		Primary UGIB hospitalizations/ESRD
	OR (95* CI)	p value	OR (95* CI)	p value
CKD or ESRD	1.30 (1.17–1.46)	<0.001	1.84 (1.61–2.09)	<0.001
Age 50–65 years vs. <50 years	2.0 (1.96–2.14)	<0.001	2.0 (1.98–2.16)	<0.001
Age >65 years vs. <50 years	2.74 (2.6–2.9)	<0.001	2.8 (2.7–2.9)	<0.001
Sex (male vs. female)	0.64 (0.62–0.66)	<0.001	0.65 (0.64–0.67)	<0.001
Race (Afro-American vs. Caucasian)	1.04 (1.0–1.09)	0.026	1.03 (0.99–1.07)	0.007
Race (Asian vs. Caucasian)	1.42 (1.26–1.61)	<0.001	1.40 (1.25–1.59)	<0.001
Smoking	0.95 (0.92–0.99)	0.011	0.96 (0.92–0.99)	0.028
Alcohol	1.97 (1.88–2.03)	<0.001	1.99 (1.90–2.09)	<0.001
Hypertension	1.02 (1.00–1.05)	0.04	0.99 (0.97–1.02)	0.91
Diabetes mellitus	0.96 (0.93–0.98)	0.003	0.95 (0.92–0.97)	<0.001
Thrombocytopenia	0.88 (0.82–0.93)	<0.001	0.87 (0.82–0.93)	<0.001
Coagulation defect	2.20 (2.1–2.3)	<0.001	2.2 (2.1–2.3)	<0.001
CAD	0.99 (0.95–1.0)	0.16	0.97 (0.94–1.0)	0.06
CVA	0.25 (0.23–0.29)	<0.001	0.26 (0.23–0.29)	<0.001
Acute anemia	12.1 (11.5–12.6)	<0.001	12.2 (11.612.8)	<0.001
Chronic anemia	3.17 (2.9–3.35)	<0.001	3.1 (2.9–3.3)	<0.001
Cancer	0.73 (0.69–0.76)	<0.001	0.73 (0.69–0.76)	<0.001
Venous thromboembolism	0.58 (0.53–0.63)	<0.001	0.58 (0.54–0.62)	<0.001
Atrial fibrillation/flutter	1.10 (1.07–1.13)	<0.001	1.12 (1.09–1.15)	<0.001
Cirrhosis	2.54 (2.40–2.69)	<0.001	2.56 (2.23–2.60)	<0.001
General surgery	0.16 (0.14–0.19)	<0.001	0.16 (0.14–0.18)	<0.001
Orthopedic surgery	0.03 (0.02–0.05)	<0.001	0.03 (0.02–0.05)	<0.001
Sepsis	0.29 (0.26–0.33)	<0.001	0.29 (0.25–0.33)	<0.001
DIC	1.08 (0.86–1.35)	0.50	1.08 (0.86–1.36)	0.48
Peripheral vascular disease	0.94 (0.89–0.97)	0.003	0.92 (0.88–0.96)	0.001
CHF	0.78 (0.76–0.80)	<0.001	0.76 (0.74–0.89)	<0.001
Insurance status
Medicaid vs. medicare	0.74 (0.70–0.79)	<0.001	0.75 (0.71–0.80)	<0.001
Private vs. medicare	0.91 (0.87–0.94)	<0.001	0.92 (0.89–0.96)	<0.001
Self pay vs. medicare	1.30 (1.19–1.37)	<0.001	1.31 (1.22–1.40)	<0.001
CKD or ESRD/age >65 years	0.92 (0.84–1.0)	<0.001	0.83 (0.74–0.93)	0.002
CKD or ESRD/sex	1.31 (1.24–1.40)	<0.001	1.22 (1.10–1.36)	<0.001
CKD or ESRD/thrombocytopenia	0.84 (0.72–0.96)	0.017	non significant b
CKD or ESRD/coagulation defect	0.91 (0.78–1.02)	0.126	0.90 (0.71–1.15)	0.43
CKD or ESRD/CAD	0.52 (0.45–61)	<0.001	0.74 (0.58–0.93)	0.012
CKD or ESRD/acute anemia	0.92 (0.84–0.98)	<0.001	0.63 (0.57–0.70)	<0.001
CKD or ESRD/chronic anemia	0.68 (0.59–0.79)	<0.001	0.67 (0.49–0.92)	0.016
CKD or ESRD/atrial fibrillation	1.12 (1.04–1.20)	0.001	non significant b
CKD or ESRD/cirrhosis	0.63 (0.54–0.73)	<0.001	0.91 (0.74–1.11)	0.39
CKD or ESRD/alcohol	0.81 (0.67–0.97)	0.026	0.67 (0.47–0.96)	0.029
CKD or ESRD/hypertension	0.78 (0.73–0.84)	<0.001	non significant b
CKD or ESRD/diabetes mellitus	0.94 (0.88–1.00)	0.06	non significant b

^aInteraction terms were generated to account for interaction between independent predictors. Only the interaction terms found significantly interacting with each other (at p <0.05) were included in the model. In the above table, only interaction terms with the primary predictor variable CKD or ESRD are reported.

^bInteraction not found significant on univariate analysis with primary predictor variable ESRD, and hence the interaction term was not included in that model.

For the analyses of all-cause mortality, mechanical ventilation and blood product transfusions were additionally found to be significant on univariate analysis and were incorporated into the final models. The risk of all-cause in-hospital mortality was 47* higher in the CKD group when compared to the non-CKD, non-ESRD group, after controlling for the potential confounders and accounting for the interaction amongst the independent variables (model 3, OR 1.47, 95* CI 1.21–1.78, p < 0.001). Furthermore, the risk of all-cause mortality in primary UGIB patients with ESRD was three times higher when compared to the non-renal disease group (model 4, OR 3.02, 95* CI 2.23–4.1, p < 0.001). The strongest predictors of all-cause mortality in primary UGIB patients in both models were the need for mechanical ventilation, severe sepsis, presence of DIC, presence of cancer and age >65 years (all p < 0.001, table 4). As in models 1 and 2, the database could not provide insights into how the chronicity, severity or outpatient care of the various diagnoses may have impacted either the decision for admission, the treatment plan or, ultimately, the inpatient death. This makes it difficult to interpret the finding that the listing of ICD-9-CM codes for either acute or chronic anemia was protective for all-cause mortality in UGIB in both these models (p < 0.001, table 4). Other significant positive predictors were presence of coagulation defect, history of stroke, VTE and atrial fibrillation (all p < 0.001).

Table 4.

Summary of models 3 and 4 and OR and 95* CI values for all-cause in-hospital mortality in primary UGIB patients compared to the non-CKD, non-ESRD group

Risk factors/interaction terms a	Mortality (all-cause)/CKD		Mortality (all-cause)/ESRD
	OR (95* CI)	p value	OR (95* CI)	p value
CKD or ESRD	1.47 (1.21–1.78)	<0.001	3.02 (2.23–4.1)	<0.001
Age >65 years vs. <65 years	4.90 (3.60–6.90)	<0.001	5.57 (4.1–7.7)	<0.001
Sex (male vs. female)	0.69 (0.51–0.93)	0.17	0.72 (0.53–0.97)	0.34
Race (Afro-American vs. Caucasian)	1.04 (0.88–1.24)	0.59	0.97 (0.82–1.16)	0.78
Race (Asian vs. Caucasian)	0.76 (0.53–1.10)	0.15	0.76 (0.52–1.08)	0.13
Smoking	0.70 (0.51–0.95)	0.026	0.72 (0.0.53–0.98)	<0.001
Alcohol	0.98 (0.70–1.38)	0.9	1.04 (1.0.74–1.47)	0.78
Hypertension	0.45 (0.36–0.56)	<0.001	0.42 (0.0.33–0.51)	<0.001
Diabetes mellitus	0.85 (0.73–1.0)	0.05	0.78 (0.68–0.91)	<0.001
Coagulation defect	3.79 (2.41–5.90)	<0.001	3.83 (2.43–6.1)	<0.001
CVA	1.69 (0.91–3.12)	0.09	1.48 (0.79–2.80)	0.22
Acute anemia	0.58 (0.51–0.65)	<0.001	0.58 (0.52–0.66)	<0.001
Chronic anemia	0.31 (0.19–0.48)	<0.001	0.31 (0.19–0.49)	<0.001
Cancer	7.13 (5.24–9.71)	<0.001	7.44 (5.46–10.1)	<0.001
VTE	2.66 (1.60–4.41)	<0.001	2.63 (1.58–4.37)	<0.001
Atrial fibrillation/flutter	1.68 (1.38–2.04)	<0.001	1.69 (1.40–2.1)	<0.001
Cirrhosis	2.59 (1.90–3.53)	<0.001	2.60 (1.91–3.55)	<0.001
CHF	2.32 (1.9–2.76)	<0.001	2.33 (1.96–2.77)	<0.001
Sepsis	22.62 (14.9–34.2)	<0.001	21.1 (13.9–31.8)	<0.001
DIC	41.08 (15.3–109.9)	<0.001	36.9 (14.1–96.1)	<0.001
Mechanical ventilation	33.4 (24.6–45.2)	<0.001	33.1 (24.3–44.8)	<0.001
Packed red blood cell transfusion	1.57 (1.21–2.04)	0.001	1.57 (1.20–2.10)	0.002
Insurance status
Medicaid vs. medicare	1.26 (0.97–1.65)	0.08	1.32 (1.01–1.72)	0.04
Private vs. medicare	0.91 (0.74–1.11)	0.34	0.95 (0.78–1.16)	0.62
Self pay vs. medicare	1.24 (0.87–1.77)	0.22	1.28 (0.90–1.83)	0.16
CKD/diabetes mellitus a	0.53 (0.35–0.81)	0.004	not significant b
ESRD/sex	not significant		0.54 (0.34–0.86)	0.01
ESRD/CVA	not significant		0.013 (0.001–0.22)	0.003

^aInteraction terms were generated to account for interaction between independent predictors. Only the interaction terms found significantly interacting with each other (at p <0.05) were included in the model. In the above table, only interaction terms with the primary predictor variable CKD or ESRD are reported.

^bInteraction not found significant on univariate analysis with primary predictor variable ESRD and hence the interaction term was not included in that model.

Discussion

It has been widely reported, but not rigorously studied, that the risk for GI bleeding is increased with chronic renal disease. Although the incidence has not been well-established [4], Holden et al. [5] reported that hemorrhage in ESRD patients occurs at 2.5* per person year and is of GI origin in most patients, especially those being treated with aspirin or aspirin-warfarin combined anticoagulation. Wasse et al. [6] utilized the United States Renal Data System (USRDS) and reported approximately 23 bleedings per 1,000 ESRD patients per year. The highest rates were in Caucasians, diabetics, smokers, and those with cardiovascular disease, impaired independent ambulation and malnutrition. In our study a true incidence of UGIB cannot be calculated because we utilized a database of hospital admissions and the population at risk is difficult to ascertain. Proposed etiologic factors for UGIB in CKD include hypergastrinemia, high levels of gastric acidity, abnormalities in the gastric mucosa, ischemia due to splanchnic vascular disease, platelet dysfunction (i.e. due to uremia, medications or secondary to anemia), ulcerogenic drugs (e.g., NSAIDs), anticoagulation that is chronic (i.e. aspirin for cardioprotective purposes or warfarin) or intermittent (during hemodialysis treatments), and unintentional systemic effects from leakage of heparin-'locked' catheters [7]. Conversely, other investigators used USRDS morbidity and mortality studies for relevant ICD-9 codes, and regression analyses did not support UGIB risk from aspirin, NSAIDs, antiplatelet or anticoagulant medications or dialysis modality; however, there was an association with cardiovascular disease, smoking and disability [6]. In our study, the regression analyses did account for coded coagulopathies but did not include explicit data concerning aspirin; however, the latter is expected to have significant use in the CKD/ESRD population because of multiple cardiac risk factors. Thus, we were instead able to control for CAD, diabetes mellitus, stroke and hypertension which are extremely strong correlates of aspirin use. In that the database included acquired coagulation defects and such arrhythmias as atrial fibrillation/flutter, history of deep vein thrombosis/pulmonary embolism and history of CVA, we believe that we were able to control for warfarin use, without having specific results for prothrombin time/international normalized ratio values. The role played by transient elevations in partial thromboplastin time from heparin use during hemodialysis or from leaks from heparin locks would not be controlled for in our analyses as the partial thromboplastin time may have returned to normal by the time the patients presented to the emergency room with the primary UGIB. Lastly, the presence of anemia can itself cause platelet dysfunction, thereby worsening a UGIB and initiating a vicious cycle of further bleeding. It has been speculated that anemia causes dysfunctional crowding of platelets in the middle of a lamellar blood stream and thereby makes them inaccessible for interaction with endothelium to initiate the coagulation cascade [8].

UGIB is most commonly due to gastric or duodenal peptic ulcer disease [9], and recent endoscopy series have stressed a disproportionately high rate of erosive gastritis, esophagitis, vascular ectasias and angiodysplasia [3,10,11]. This is consistent with the pattern of ICD-9-CM coding demonstrated in our results, given that ‘unspecified category’ (code 578) was the most common designation.

The outcome from GI bleeding in CKD patients is difficult to discern from the literature since many of the publications were from an era prior to advanced interventional endoscopic techniques, widespread use of proton pump inhibitor medications or treatment for Helicobacter pylori infections. Most were small series or single-center experiences. Chalasani et al. [2], for example, described their center's experience of 727 patients with UGIB. Sixty had CKD and they had the same rates for rebleeding, transfusions, surgical intervention and mortality as those with normal renal function. Other groups have described poor clinical outcomes from UGIB in patients with either CKD or acute kidney injury [12,13]. The latter studies stressed the additional risk factors of concurrent cirrhosis or thrombocytopenia. Cheung et al. [3] reported that compared to those with normal renal function, CKD patients had a higher incidence of rebleeding, transfusions and longer length of stay. There was only a trend for higher mortality, which may have been impacted by the small study size. Similarly, Boyle et al. [9] described CKD patients having increased risk for uncontrolled bleeding or rebleeding, blood transfusions and longer hospitalization. In that small study of 20 patients, there was only a trend for increased mortality. Klebl et al. [14], however, reviewed their 363 patients with UGIB and found that risk factors for hospital mortality included renal disease (defined as acute or chronic conditions with a serum creatinine at least 1.25 times normal), coagulopathy and glucocorticoid use. Tsai et al. [11] reported that the mortality rate for their 58 ESRD patients with UGIB was 13*, compared to 2* in the 640 individuals with normal kidney function. The renal patients were older and more likely to have bleeding from erosive gastritis, erosive esophageal disease, or esophageal ulcers.

Our study has, for the first time, used a very large multi-center database to demonstrate a profound increase in the risk for all-cause mortality for CKD and ESRD patients admitted with primary UGIBs, when compared to those without renal dysfunction. CKD patients had a nearly 50* increase in risk of mortality, and this risk reached approximately 300* for those with ESRD. This finding remains significant after controlling for the potential confounding factors of severe sepsis, mechanical ventilation, DIC, cirrhosis, cancer, stroke, and age; all of which are independently associated with higher mortality in this setting. The strongest predictors of mortality were the need for mechanical ventilation, severe sepsis, DIC, cancer, age >65 years, coagulation defect, and VTE. Whether aggressive measures to control infection, minimize anticoagulants, or correct anemia would lead to improved survival remains to be fully investigated.

Our study has several limitations. This analysis is limited to episodes of primary UGIB requiring inpatient care (not including outpatient or observation-status hospitalization) and thus likely underestimates the true frequency of UGIB in the CKD/ESRD population. Our dataset does not provide information as to whether any of the admissions were due to rebleeding episodes. As described above, rebleeding rates are higher in ESRD patients, and it is conceivable that a higher proportion of rebleeding admissions in our dataset may be strengthening the association of ESRD with both primary UGIB and higher mortality (which is common with rebleedings [3]). Our dataset did not contain any information on UGIB in peritoneal dialysis patients. Availability of this information would have further dissected the role of ESRD pathophysiology in causing UGIB with no contribution from anticoagulant medications. In addition, the CKD group was not further characterized. This would have helped draw conclusions on the role of worsening renal failure in causing UGIB and its association with mortality. Our dataset also did not have information on steroid and NSAID use, as these are potentially ulcerogenic. The former is used in some causes of CKD (e.g. glomerulonephritides) and could cause a higher incidence of UGIB. However, we anticipate that this effect would have been quite small: our dataset was representative of the national population and thus would be expected to have a small proportion of CKD secondary to glomerulonephritis.

Lastly, we had anticipated potential confounding effects from the intensity and breadth of care prior to the index admission. We hypothesized that robust outpatient care for multiple chronic illnesses might identify individuals at risk for severe GI bleeding and ultimately could theoretically improve their survival (i.e. through early use of acid reducing medications). We believe that this phenomenon would explain the paradoxical negative association of mortality with the variables of smoking, hypertension, diabetes mellitus, some other chronic diseases, and the presence of chronic or acute anemia. Many of these nonintuitive associations were only marginally statistically significant. In fact, after controlling for the insurance status in the logistic model, the negative association of these covariates with the outcomes of interest seems to diminish (tables 3 and 4).

Conclusion

Patients with CKD or ESRD admitted with primary UGIB have a profound increase in risk of all-cause in-hospital mortality. With the risk of death being as high as three-fold higher than in patients with normal kidney function, clinicians need to be exceptionally vigilant in the care of these renal patients.

Disclosure Statement

The authors declare that they have no conflicts of interest.