Chronic Kidney Disease and Risk for Gastrointestinal Bleeding in the Community: The Atherosclerosis Risk in Communities (ARIC) Study

Abstract

Background and objectives

Patients on dialysis are known to have higher risk for gastrointestinal (GI) bleeding. However, data on mild to moderate CKD, particularly elevated albuminuria, are limited.

Design, setting, participants, & measurements

Among 11,088 participants in the Atherosclerosis Risk in Communities (ARIC) Study, we investigated the association of eGFR and urinary albumin-to-creatinine ratio (ACR) with risk for hospitalization with GI bleeding. Kidney measures were assessed at visit four (1996–1998), and follow-up was continued through 2011.

Results

During a median follow-up of 13.9 years, 686 first incident hospitalizations with GI bleeding were observed (incidence rate, 4.9 per 1000 person-years [95% confidence interval (95% CI), 4.5 to 5.3]). Multivariable Cox proportional hazards models revealed that both lower eGFR and higher ACR were associated with higher risk for GI bleeding. With eGFR≥90 ml/min per 1.73 m² as a reference, risk for GI bleeding was significant in moderately decreased eGFR of 30–59 ml/min per 1.73 m² (hazard ratio [HR], 1.51; 95% CI, 1.13 to 2.02), and was highest in severely decreased eGFR<30 ml/min per 1.73 m² (HR, 7.06; 95% CI, 3.91 to 12.76). Compared with ACR<10 mg/g, risk for GI bleeding became significantly higher in mild albuminuria with ACR 10–29 mg/g (HR, 1.36; 95% CI, 1.08 to 1.69), and was nearly double in moderate and severe albuminuria (HR, 2.13; 95% CI, 1.66 to 2.71 for ACR 30–299 mg/g, and HR, 2.07; 95% CI, 1.33 to 3.22 for ACR≥300 mg/g). These results were largely consistent in demographic and clinical subgroups and independent of incident cardiovascular events or dialysis during follow-up.

Conclusions

Individuals with even mild to moderate CKD warrant clinical attention regarding the risk of hospitalization with GI bleeding.

Introduction

Gastrointestinal (GI) bleeding is the leading cause of bleeding-related hospitalizations, and >300,000 adults in the United States are hospitalized with this diagnosis every year (1). Incidence of GI bleeding is approximately three-fold higher than that of another major bleeding event, hemorrhagic stroke (2,3). Of note, some anticoagulation drugs influence the risk of GI bleeding and hemorrhagic stroke differently (4).

GI bleeding is an important complication for patients with ESRD, with five-fold higher risk as compared with those without CKD (5,6). Anticoagulant agents used for extracorporeal circulation during hemodialysis may explain high risk of GI bleeding in hemodialysis patients, but the risk is also high in peritoneal dialysis (7) and transplant patients (8). Antiplatelet agents are frequently prescribed to patients with ESRD (9) and may also contribute to the risk of bleeding in this population (10). In addition, platelet dysfunction is present among individuals with advanced CKD (11,12).

A few prospective studies have explored risk for GI bleeding among CKD individuals not requiring dialysis (5,6,13,14). However, most of these studies used dichotomous definitions of CKD (5,6,13), and the only study assessing bleeding risk in finer CKD categories exclusively investigated patients after cardiac surgery (14). In addition, to our knowledge, no studies have evaluated whether the presence of albuminuria, a key element of defining and staging CKD (15), contributes to risk for GI bleeding. Therefore, the aim of the study was to comprehensively assess whether measures of CKD, eGFR and albuminuria, were associated with risk for incident GI bleeding in a biethnic community-based cohort, the Atherosclerosis Risk in Communities (ARIC) Study.

Materials and Methods

Study Participants

The ARIC Study is a population-based cohort study, which consists of individuals aged between 45 and 64 years at the time of enrollment (1987–1989) from four United States communities (Forsyth County, NC; Jackson, MS; suburban Minneapolis, MN; and Washington County, MD). There were three short-term follow-up visits approximately every 3 years (visit two in 1990–1992, visit three in 1993–1995, and visit four in 1996–1998). Details of the ARIC Study have been described elsewhere (16). The study was conducted complying with the Declaration of Helsinki. Written informed consent was obtained from all participants, and the institutional review board at each study site approved the study.

Both eGFR and albuminuria were assessed for the first time in the ARIC Study at visit four, and thus we set this visit as the baseline. Of 11,656 participants at visit four, we excluded participants with nonblack/nonwhite ethnicities (n=31), prior GI bleeding events (as defined below) (n=160), prevalent ESRD (n=15), and missing serum creatinine or cystatin C (n=320) (Supplemental Figure 1). The remaining 11,088 participants were included in primary analysis.

Kidney Function and Damage Markers

GFR was estimated by the CKD Epidemiology Collaboration equation incorporating serum creatinine and cystatin C as well as age, sex, and race (17). Serum creatinine was measured using a modified kinetic Jaffé method, calibrated to the Cleveland Clinic laboratory measurements (18), and then standardized to the isotope-dilution mass spectrometry traceable method (19). Serum cystatin C was measured using an enhanced immunonephelometric assay (Siemens Healthcare Diagnostics). Urine albumin-to-creatinine ratio (ACR) was used as a measure of albuminuria. Urine albumin and urine creatinine were measured by nephelometry and the Jaffé method, respectively, using spot urine samples stored at −70°C immediately after collection.

Outcome Measurements

The primary outcome of interest was first incident hospitalization with a diagnosis of GI bleeding (we will call “hospitalization with GI bleeding”) regardless of their position in the discharge record. The ARIC Study conducts active surveillance to capture hospitalizations of study participants and abstracts comprehensive lists of diagnostic codes of International Classification of Diseases Ninth Revision (ICD-9) from discharge records. The following ICD-9 codes were considered as GI bleeding: esophageal bleeding (456, 456.2, 530.7, 530.21, and 530.82), gastric and duodenal bleeding (531.XX, 532.XX, 533.XX, 534.XX, 535.01, 535.11, 535.21, 535.31, 535.41, 535.51, 535.61, 537.83, 537.84, and 578), intestinal bleeding (562.02, 562.03, 569.85, and 569.86), colon and rectal bleeding (562.12, 562.13, and 569.3), and bleeding from unspecified GI sources such as blood in stool (578.1 and 578.9) (Supplemental Table 1). Participants were censored when they were lost to follow-up, died, or were administratively censored on December 31, 2011.

Other Variables of Interest

All covariates were assessed at visit four except years of education, which were assessed at visit one. Age, sex, race, smoking status, and alcohol consumption were determined on the basis of a self-reported questionnaire. Body mass index (BMI) was calculated by dividing weight in kilograms by height in meters squared. BP was taken twice and their average was recorded. Hypertension was defined as use of antihypertensive medications, systolic BP ≥140 mmHg, or diastolic BP ≥90 mmHg. Diabetes mellitus was defined as a fasting glucose level ≥126 mg/dl, casual blood glucose level ≥200 mg/dl, use of antidiabetic medications, or a self-reported diagnosis of diabetes. Prevalent atherosclerotic cardiovascular disease (CVD) included coronary heart disease or stroke and was defined by a self-reported history at visit one or an adjudicated event between visits one and four. History of cancer, gastroesophageal reflux disease, and peptic ulcer were defined by hospital discharge records with ICD-9, 140–289, 530.81, and 531–533, respectively. Abnormal liver function was defined as aspartate transaminase and alanine transaminase greater than three times the upper limit of normal (20), or history of liver cirrhosis (ICD-9, 571 and 456.1). Use of aspirin, anticoagulant agents, nonsteroidal anti-inflammatory drugs, H₂ blockers, and proton pomp inhibitors were ascertained by a self-report at visit four with verification by checking drug containers.

Statistical Analyses

Baseline characteristics of the study participants were compared using t tests and chi-squared tests. The incidence rate of GI bleeding according to linear spline terms of eGFR (knots at 30, 45, 60, 75, and 90 ml/min per 1.73 m²) and log-transformed ACR (knots at 10, 30, and 300 mg/g) was estimated using Poisson regression models adjusted for age, sex, and race. Cox proportional hazards models were used to estimate the hazard ratio (HR). Model 1 adjusted for age, sex, race, BMI, smoking status, alcohol consumption, education level, aspirin, anticoagulant agent, nonsteroidal anti-inflammatory drugs, H₂ blockers, proton pomp inhibitors, hypertension, diabetes, abnormal liver function, prior CVD, prior neoplasm, and prior gastroesophageal reflux disease and peptic ulcer. Model 2 additionally included eGFR for analyses of ACR and ACR for analyses of eGFR. In addition, HRs of GI bleeding across categories of eGFR and ACR were quantified (15). Potential interaction was assessed by sex, race, diabetes, hypertension, smoking status, use of aspirin, use of anticoagulant agents, and prior CVD using likelihood ratio tests. The subgroup analyses were reported for eGFR and ACR dichotomized at 60 ml/min per 1.73 m² and 30 mg/g, respectively, to obtain reliable estimates.

Several sensitivity analyses were performed. First, we assessed the risk separately according to the site of GI bleeding (upper, lower, and unspecified GI bleeding). Second, we further adjusted for incident CVD or ESRD during follow-up as a time varying covariate. Records of incident ESRD were retrieved through the linkage to the United States Renal Data System. Third, we analyzed the data only using GI bleeding as a primary discharge diagnosis. Fourth, we assessed the risk of mortality associated with GI bleeding as an underlying cause in death certificates. Fifth, by the linkage to the Centers for Medicare and Medicaid Services, we analyzed the data including outpatient GI bleeding events within a subcohort of individuals aged ≥65 years. Finally, we assessed risk for non-GI-bleeding hospitalization, which was defined as hospitalizations without ICD-9 codes for GI bleeding.

We also assessed whether the information on CKD measures could improve risk discrimination in an established prediction scheme for bleeding risk. We used Hypertension, Abnormal renal/liver function, Stroke, Bleeding history or predisposition, Labile international normalized ratio, Elderly (65 years or older), Drugs/alcohol concomitantly (HAS-BLED) (20) as the base model because it has demonstrated better performance among others (21,22). Covariates in HAS-BLED were slightly modified as summarized in Supplemental Table 2. Harrell c-statistics were used as a measure of discrimination (23). Change in c-statistics was assessed with the use of the delta methods. We performed complete-case analysis as missing values of each variable of interest were found in <1% of participants. A two-sided P value of <0.05 was considered statistically significant. All statistical analyses were conducted using STATA version 13 (StataCorp., College Station, TX).

Results

Of 11,088 participants, 686 participants (6.2%) had hospitalization with GI bleeding during a median follow-up of 13.9 years (140,517 person-years). Compared with individuals without GI bleeding, those with GI bleeding were more likely to be older, of black race, current/former smokers, less educated, on aspirin and anticoagulant agents, and have higher BMI and a greater prevalence of hypertension, diabetes, abnormal liver function, and CVD (Table 1). When baseline characteristics were assessed across eGFR or ACR categories, individuals in the lower eGFR or higher ACR category were more likely to be hypertensive, diabetic, taking aspirin or anticoagulant agents, and have prevalent CVD compared with those with high eGFR or low ACR, respectively (Supplemental Tables 3 and 4).

Table 1.

Baseline characteristics of the study population

Characteristic	Without GI Bleeding (n=10,402)	With GI Bleeding (n=686)	P Value
Mean age (SD), yr	62.7 (5.7)	64.8 (5.5)	<0.001
Female, n (%)	5854 (56)	363 (53)	0.09
Black race, n (%)	2285 (22)	190 (28)	<0.001
Mean body mass index (SD), kg/m2	28.7 (5.6)	29.7 (5.9)	<0.001
Mean BP (SD), mmHg
Systolic	127.3 (18.9)	132.2 (20.0)	<0.001
Diastolic	71.0 (10.3)	71.2 (11.3)	0.59
Current/former smoker, n (%)	6000 (58)	440 (65)	<0.001
Current/former alcohol consumer, n (%)	8217 (79)	513 (76)	0.02
12 yr or more of education, n (%)	8436 (81)	498 (73)	<0.001
Medication use, n (%)
Aspirin	5840 (56)	443 (65)	<0.001
Anticoagulant	175 (2)	41 (6)	<0.001
NSAIDs	2940 (28)	213 (31)	0.12
H2 blocker	977 (9)	78 (11)	0.09
Proton pomp inhibiter	313 (3)	32 (5)	0.02
Medical history, n (%)
Hypertension	3861 (37)	341 (50)	<0.001
Diabetes	1667 (16)	171 (25)	<0.001
Abnormal liver function	58 (1)	14 (2)	<0.001
Prior cardiovascular disease	983 (9)	127 (19)	<0.001
Prior neoplasm	850 (8)	61 (9)	0.51
Prior GERD/peptic ulcer	173 (2)	14 (2)	0.46

GI, gastrointestinal; NSAIDs, nonsteroidal anti-inflammatory drugs; GERD, gastroesophageal reflux disease.

Complete-case analysis was performed for the following variables with missing data: body mass index (n=22), systolic BP (n=2), diastolic BP (n=2), smoking status (n=70), alcohol consumption (n=68), years of education (n=17), abnormal liver function (n=63), hypertension (n=63), diabetes (n=55), use of aspirin (n=32), and use of anticoagulant agents (n=31).

The crude incidence rate of GI bleeding was 4.9 per 1000 person-years (95% confidence interval [95% CI], 4.5 to 5.3). Figure 1 shows the age-, sex-, and race-adjusted incidence rate across eGFR or log ACR. There was a nonlinear relationship in eGFR, with a shallow risk gradient above eGFR 45 ml/min per 1.73 m² and a steeper risk gradient below eGFR 30–45 ml/min per 1.73 m². In contrast, the relationship was nearly linear for log ACR.

Figure 1.

Incidence rate of GI bleeding was higher in lower eGFR and higher ACR. Incidence rate for GI bleeding overlaid with histogram according to (A) eGFR and (B) ACR. The solid line indicates the point estimate, and the shaded area indicates corresponding 95% confidence intervals. Bars in the background indicate the histogram for distribution of eGFR and ACR. The models were adjusted for age, sex, and race. ACR, albumin-to-creatinine ratio; GI, gastrointestinal.

After additionally adjusting for other potential confounders, the associations were consistent and significant for both lower eGFR and higher ACR (Model 1 in Table 2). As compared with eGFR≥90 ml/min per 1.73 m², eGFR 30–59 ml/min per 1.73 m² was associated with 50% greater risk for GI bleeding (HR, 1.51; 95% CI, 1.13 to 2.02), and eGFR<30 ml/min per 1.73 m² was associated with seven-fold higher risk (HR, 7.06; 95% CI, 3.91 to 12.76). As compared with ACR<10 mg/g, moderate (ACR 30–299 mg/g) and severe (ACR≥300 mg/g) albuminuria were associated with approximately two-fold higher risk of GI bleeding (HR, 2.13; 95% CI, 1.68 to 2.71 and HR 2.07; 95% CI, 1.33 to 3.22, respectively). Mild albuminuria (ACR 10–29 mg/g) was also associated with incident GI bleeding (HR, 1.35; 95% CI, 1.08 to 1.69). When further adjusting for both eGFR and ACR, all associations remained significant except eGFR 60–89 ml/min per 1.73 m² (P=0.06) and ACR≥300 mg/g (P=0.14) (Model 2 in Table 2).

Table 2.

Hazard ratios of GI bleeding events according to eGFR and ACR categories

Category	Total No.	No. of Events	Hazard Ratio (95% CI)
			Model 1	Model 2
eGFR
≥90 ml/min per 1.73 m2	4586	214	1 (Reference)	1 (Reference)
60–89 ml/min per 1.73 m2	5543	368	1.19 (1.00 to 1.43)	1.18 (0.99 to 1.42)
30–59 ml/min per 1.73 m2	715	70	1.51 (1.13 to 2.02)	1.40 (1.04 to 1.88)
<30 ml/min per 1.73 m2	40	12	7.06 (3.91 to 12.76)	5.64 (2.87 to 11.10)
ACR
<10 mg/g	8628	449	1 (Reference)	1 (Reference)
10–29 mg/g	1278	97	1.35 (1.08 to 1.69)	1.36 (1.09 to 1.70)
30–299 mg/g	704	88	2.13 (1.68 to 2.71)	2.05 (1.61 to 2.61)
≥300 mg/g	182	22	2.07 (1.33 to 3.22)	1.49 (0.91 to 2.43)

Model 1 was adjusted for age, sex, race, body mass index, alcohol consumption, smoking status, education level, aspirin, anticoagulant agents, nonsteroidal anti-inflammatory drugs, H₂ blocker, proton pomp inhibitor, hypertension, diabetes, abnormal liver function, cardiovascular disease, neoplasm, and gastroesophageal reflux disease/peptic ulcer. Model 2 was adjusted for ACR/eGFR in addition to the model 1. ACR was missing in 92 participants. GI, gastrointestinal; ACR, albumin-to-creatinine ratio; 95% CI, 95% confidence interval.

In the context of the Kidney Disease Improving Global Outcomes 2012 guidelines for CKD staging (15), we assessed the risk of hospitalizations with GI bleeding in crosscategories of eGFR and ACR (Supplemental Table 5). These two CKD measures were multiplicatively associated with GI bleeding risk without a significant interaction (P for interaction, 0.44). Individuals with both eGFR<30 ml/min per 1.73 m² and ACR≥300 mg/g had the highest HR (HR, 10.06; 95% CI, 4.85 to 20.86).

In subgroup analyses, we found no significant interaction by sex, diabetes, hypertension, smoking status, use of aspirin, use of anticoagulant, or history of CVD with eGFR and ACR in the risk of GI bleeding (Figure 2). ACR showed a stronger association with GI bleeding among black participants compared with white participants (HR, 3.42; 95% CI, 2.53 to 4.64 versus HR, 2.36; 95% CI, 1.82 to 3.07; P for interaction, 0.05) (Figure 2B). However, the interaction was attenuated when the model was adjusted for incident ESRD as a time-varying covariate (P for interaction, 0.14).

Figure 2.

In subgroup analyses, no significant interaction was observed except for race in the analysis of ACR. Hazard ratios for GI bleeding according to (A) eGFR (< versus ≥60 ml/min per 1.73 m²) and (B) ACR (≥ versus <30 mg/g). Models were adjusted for age, sex, race, body mass index, alcohol consumption, smoking status, education level, aspirin, anticoagulant agents, nonsteroidal anti-inflammatory drugs, H² blocker, proton pomp inhibitor, hypertension, diabetes, abnormal liver function, cardiovascular disease, neoplasm, and gastroesophageal reflux disease/peptic ulcer. The circles represent the point estimate of relative hazard, and the horizontal lines indicate corresponding 95% confidence intervals. 95% CI, 95% confidence interval; ACR, albumin-to-creatinine ratio; GI, gastrointestinal; HR, hazard ratio.

Site specific analyses for upper, lower, or unspecified GI bleeding risk yielded consistent results (Supplemental Table 6). The additional adjustment for incident CVD as a time-varying covariate did not substantially alter the association (Supplemental Table 7). Similarly, the adjustment for incident ESRD as a time-varying variable did not alter the association except for ACR≥300 mg/g (Supplemental Table 8). Restricting to 346 cases of GI bleeding as a primary discharge diagnosis did not alter the association (Supplemental Table 9). Also, the association was consistent when assessing the risk of mortality associated with GI bleeding (Supplemental Table 10). When taking into account GI bleeding events at outpatient settings within a subcohort aged ≥65 years (n=3906), the association was attenuated but largely consistent (Supplemental Table 11). The HRs for non-GI-bleeding hospitalization were smaller than those for hospitalization with GI bleeding (Supplemental Table 12). Seemingly unrelated regression (24), a test to compare the two coefficients for two outcomes (non-GI bleed versus GI bleed), demonstrated statistical significance (P=0.007 for eGFR and P=0.004 for log-transformed ACR as a continuous variable).

Finally, we assessed whether CKD measures could improve risk discrimination of GI bleeding beyond an established prediction scheme of major bleeding, HAS-BLED (20). In the base model, severely decreased kidney function (eGFR<30 ml/min per 1.73 m²) demonstrated the highest HR (HR, 8.22; 95% CI, 4.51 to 15.00; left column in Table 3). Moderately decreased kidney function (eGFR 30–59 ml/min per 1.73 m²), which was not included in HAS-BLED, demonstrated a similar HR (HR, 1.57; 95% CI, 1.23 to 2.02) as hypertension (HR, 1.59; 95% CI, 1.22 to 2.08) and the use of aspirin (HR, 1.42; 95% CI, 1.21 to 1.67) (a middle column in Table 3). When the presence of albuminuria was added, the HR was one of the highest among other variables (HR, 2.23; 95% CI, 1.80 to 2.77; right column in Table 3) and was similar to stroke (HR, 1.97; 95% CI, 1.37 to 2.81) and older age (HR, 1.82; 95% CI, 1.56 to 2.12). Reflecting their strong associations, inclusion of moderately decreased eGFR and the presence of albuminuria significantly improved the risk discrimination of GI bleeding (Table 3). Of note, addition of albuminuria increased the c-statistic by 0.018, whereas each of hypertension, history of stroke, and aspirin used contributed to a c-statistic increment of 0.006, 0.006, and 0.021, respectively, within the base model.

Table 3.

Hazard ratios of GI bleeding in traditional risk factors

Risk Factor	Hazard Ratio (95% CI)
	Base Model	Base + Moderately Decreased eGFR	Base + Albuminuria	Base + Moderately Decreased eGFR + Albuminuria
Abnormal kidney function
Moderately decreased		1.57 (1.23 to 2.02)		1.41 (1.09 to 1.81)
Severely decreased	8.22 (4.51 to 15.00)	8.62 (4.73 to 15.73)	4.63 (2.49 to 8.60)	4.93 (2.65 to 9.19)
Presence of albuminuria			2.32 (1.87 to 2.87)	2.23 (1.80 to 2.77)
Age>65 yr	1.90 (1.63 to 2.22)	1.84 (1.57 to 2.14)	1.87 (1.60 to 2.18)	1.82 (1.56 to 2.12)
Hypertension	1.62 (1.24 to 2.11)	1.59 (1.22 to 2.08)	1.36 (1.03 to 1.78)	1.35 (1.03 to 1.77)
Abnormal liver function	4.11 (2.37 to 7.12)	3.97 (2.29 to 6.88)	3.94 (2.27 to 6.83)	3.87 (2.23 to 6.71)
Stroke	2.14 (1.50 to 3.06)	2.08 (1.46 to 2.98)	2.01 (1.41 to 2.88)	1.97 (1.37 to 2.81)
Use of anticoagulant agents	3.22 (2.31 to 4.49)	3.13 (2.24 to 4.36)	3.12 (2.24 to 4.36)	3.07 (2.20 to 4.28)
Use of aspirin	1.44 (1.23 to 1.68)	1.42 (1.21 to 1.67)	1.43 (1.22 to 1.68)	1.42 (1.22 to 1.67)
Harrell's c-statistic	0.647	0.653	0.665	0.666
Change in c-statistic from base model (95% CI)		0.01 (0.01 to 0.01)	0.02 (0.01 to 0.03)	0.02 (0.01 to 0.03)

Moderately decreased kidney function was eGFR 30–59 ml/min per 1.73 m². Severely decreased kidney function was eGFR<30 ml/min per 1.73 m². Presence of albuminuria was albumin-to-creatinine ratio ≥30 mg/g. Hypertension was systolic BP >160 mmHg regardless of treatment status. Abnormal liver function was liver enzyme greater than three times upper limit of normal, or the presence of liver cirrhosis. GI, gastrointestinal; 95% CI, 95% confidence interval.

Discussion

In this community-based study of >10,000 participants with a median follow-up of 14 years, we demonstrated that moderately decreased eGFR was associated with nearly 50% greater risk for GI bleeding as compared with normal kidney function, and the risk was highest in severely decreased eGFR. In addition, risk for GI bleeding became significantly higher even in mild albuminuria, and moderate and severe albuminuria doubled the risk for GI bleeding independent of eGFR. The association was consistent across demographic and clinical subgroups and was regardless of incident CVD or incident ESRD during follow-up.

Although underlying pathophysiology is quite different between upper and lower GI bleeding (25–28), we observed consistent associations of CKD measures across sites of GI bleeding. In patients with ESRD, accumulation of uremic toxins impairs platelet aggregation and normal platelet-vessel interactions (29), increasing bleeding risk (30). Furthermore, high prevalence of gastric vascular ectasia (31) and ischemic colitis (32) among individuals with CKD may also contribute to the higher risk. Also, because CKD is a risk factor for hospitalization (33), CKD individuals may have higher chances of developing complications of GI bleeding during their hospital stay. In the previous reports of nondialysis populations, severe renal insufficiency was significantly associated with risk for hospitalization with GI bleeding (5,14), whereas the results for mild to moderate renal insufficiency have been inconclusive (5,14,34,35). Our finding that the bleeding risk is highest among participants with severely decreased eGFR is consistent with the previous studies. The significant associations with moderately decreased eGFR (30–59 ml/min per 1.73 m²) in our study may reflect the influence of uremic toxin on bleeding risk even before reaching an advanced stage of CKD.

To our knowledge, this is the first study demonstrating the association of higher albuminuria with risk for GI bleeding. Although the mechanisms linking albuminuria to GI bleeding are not clear, high albuminuria has been related to hemorrhagic stroke (36) and hemorrhagic transformation after the ischemic stroke (37). High albuminuria may reflect systemic endothelial dysfunction. Animal models of albuminuria showed impaired function of endothelial glycocalyx (38) and reduced production of endothelial-derived nitric oxide (39). Future studies are needed to elucidate the pathophysiologic pathways linking albuminuria and GI bleeding.

In subgroup analyses, ACR tended to be more strongly associated with GI bleeding risk in blacks than in whites. Black persons have higher risk for bleeding after thrombolytic therapy than white persons (40,41), raising a possibility of genetic determinants. Another potential explanation is that among those with ACR>30 mg/g, the proportion of severe albuminuria (ACR≥300mg/g) was greater in blacks than whites (25% versus 20%). Furthermore, black race is associated with higher risk of ESRD, a condition conferring especially high bleeding risk (42), and thus the racial difference might be driven by blacks who developed ESRD during follow-up. A previous study reported no difference in the risk of bleeding events between whites and blacks on dialysis (43). Indeed, race-ACR interaction was attenuated when we accounted for incident ESRD during follow-up.

Given that GI bleeding is the leading cause of bleeding-related hospitalizations, this study may have several clinical implications. First, physicians should be aware of high risk of GI bleeding in individuals with reduced kidney function even at a moderate stage. Compared with severely decreased kidney function, as implemented in HAS-BLED (20), less attention may be drawn to individuals with moderately decreased kidney function. With one exception of a recently proposed scheme, Outcomes Registry for Better Informed Treatment (44), none of the representative prediction tools for major bleeding account for moderately decreased kidney function (20–22). In our study, the association of moderately decreased kidney function was nearly equivalent to that of hypertension and the use of aspirin. Furthermore, our findings demonstrated albuminuria as a potent predictor of GI bleeding. Although the change in c-statistic by adding albuminuria may look small (Δc-statistics= 0.018), its contribution to the risk discrimination was similar or even greater than most of the established predictors. Assessment of albuminuria is already recommended for persons with diabetes, hypertension, and CKD (15). Thus, when available, both CKD measures should be taken into consideration. This is particularly the case when physicians prescribe antiplatelet or anticoagulant agents because the evidence of these medications in CKD patients is not necessarily robust (45,46).

Our study has several limitations. First, the outcome ascertainment of GI bleeding relied on the ICD-9 codes. Nevertheless, the validity of ICD-9 codes to identify GI bleeding has been reported to be high. The positive predictive value of ICD-9 codes was 88%–98% for nonvariceal upper GI bleeding (47,46), and 95%–98% for lower GI bleeding when validated with charts reviews (46). Similarly, the positive predictive value was 87% for acute, nonvariceal upper GI hemorrhage when validated with endoscopic diagnoses (47). Second, as is true in any observational cohort study, we cannot exclude the possibility of residual confounding. For example, we could not adjust for platelet count. Finally, our study population is restricted to a middle-aged biethic population; therefore, whether our results can be generalizable to other races and age ranges is yet to be determined.

In conclusion, in a biethnic middle-aged community-based cohort, lower eGFR and higher albuminuria were associated with risk of hospitalization with GI bleeding independently of each other and other potential confounders. These results suggest that even persons with mild to moderate CKD should warrant attention regarding the risk of GI bleeding.

Disclosures

None.