ABO Blood Type and Stroke Risk: The REasons for Geographic and Racial Differences in Stroke (REGARDS) Study

Abstract

Background

ABO blood type is an inherited trait associated with coagulation factor levels and vascular outcomes.

Objectives

To assess the association of blood type with stroke and whether blood type contributes to racial disparities in stroke in the United States.

Patients and Methods

The REasons for Geographic and Racial Differences in Stroke (REGARDS) Study recruited 30,239 participants between 2003-07. Using a case-cohort design, blood type was genotyped in 646 participants with stroke and a 1,104 participant cohort random sample. Cox models adjusting for Framingham stroke risk factors assessed the association of blood type with stroke.

Results

Over 5.8 years of follow-up, blood types A or B versus type O were not associated with stroke. Blood type AB versus O was associated with an increased risk of stroke (adjusted HR 1.83; 95% CI 1.01, 3.30). The association of blood type AB versus O was greater in those without diabetes (adjusted HR 3.33; 95% CI 1.61, 6.88) than those with diabetes (adjusted HR 0.49; 95% CI 0.17, 1.44) (p-interaction = 0.02). Factor VIII levels accounted for 60% (95% CI 11%, 98%) of the association of AB blood type and stroke risk.

Conclusion

Blood type AB is associated with an increased risk of stroke that is not attenuated by conventional stroke risk factors and factor VIII levels were associated with 60% of the association. While blood type AB is rare in the U.S. population, it is a significant stroke risk factor and may play an important role in stroke risk in these individuals.

The ABO blood system determines carbohydrate moieties that are ubiquitously expressed on red blood cells and the vascular endothelium[1]. While most research on ABO antigens has focused on issues of transfusion and organ transplantation, there is a growing body of literature associating ABO blood type with diverse disease states including cardiovascular disease[1].

The ABO gene on chromosome 9 encodes a glycosyltransferase which attaches monosaccharides to the H-antigen on red blood cells. Blood type O results from an inactive glycosyltransferase whereas blood types A (and its subtypes A1 and A2) and B result from different monosaccharide chains attached to the H antigen as a result of polymorphisms in the glycosyltransferase. ABO antigens are also expressed on vascular endothelium and through unclear mechanisms, associated with levels of the procoagulant proteins factor VIII and von Willebrand Factor (VWF) and markers of endothelial function such as p-selectin and soluble intracellular adhesion molecule 1[2-7].

Prior studies have related ABO blood type with venous thrombosis and inconsistently with myocardial infarction; genetic studies have identified the ABO locus as relating to myocardial infarction, stroke, and venous thrombosis risk[2, 8-12]. Only one prior study demonstrated an association with ABO blood type and stroke risk but was limited in that stroke events were ascertained only in those with hypertension or in peri-menopausal women[10].

We hypothesized that non-O blood types would be associated with stroke risk in the REasons for Geographic and Racial Differences in Stroke (REGARDS) study and that some of the excess risk would be mediated by factor VIII levels. We further hypothesized that due to racial differences in ABO blood type that blood type would mediate some of the excess stroke risk seen in blacks in the United States (U.S.). While blood type is not a modifiable risk factor, it may help select a population at risk for more aggressive risk factor reduction.

Methods

Cohort

REGARDS recruited 30,239 individuals from the contiguous U.S. between 2003-07 and was designed to study regional and racial differences in stroke[13]. By design, the cohort was 55% female, 41% black, and 56% lived in the southeast (recruitment goal of each was 50%). Participants were contacted over the phone and after verbal informed consent participated in a telephone interview gathering basic demographic and risk factor data. The telephone response rate was 33% and the cooperation rate was 49%[14]. Exclusion criteria were race other than black or white, medical conditions preventing long-term participation, active cancer or active treatment for cancer, resident in or waiting placement in a nursing home, or inability to communicate in English. Participants then underwent an in-home visit where anthropomorphic data was collected and a medication inventory, fasting phlebotomy for laboratory studies, and written informed consent were done (Examination Management Systems Incorporated, Irving, Texas).

Laboratory

Phlebotomy was performed after a 10-12 hour fast in the morning at an in-home visit. Samples were centrifuged locally for 97 ± 127 minutes at varying speeds and kept at −4°C and shipped that night on ice to the study laboratory at the University of Vermont. On arrival, samples were recentrifuged for 30,000 G minutes and stored at −80°C. Factor VIII antigen was measured in plasma using an enzyme linked immunosorbant assay and reported as a percent of normal concentration of pooled plasma (Enzyme Research Laboratories, South Bend, Indiana). The correlation coefficient between ideal processing and REGARDS processing was 0.90. Blood type was inferred by using 4 informative single nucleotide polymorphisms (SNP) (rs507666, rs687289, rs8176704, and rs8176749) measured using probes from TaqMan (Life Technologies, Grand Island, NY)[15].

Probabilities of diplotypes of the four informative SNPs were constructed using PHASE 2.1 (University of Washington, Seattle, WA)[16, 17]. Diplotypes were grouped into four mutually exclusive categories corresponding to blood types A, B, AB, and O based on known mutations[4] and review of Blood Group Antigen Gene Mutation Database for ambiguous diplotypes[18] (Table 1). Individuals missing genotype data were assigned haplotype probabilities based on the population probabilities stratified by race. Four observations were created for each individual for each blood type and weighted by the probability of that blood type to account for diplotype uncertainties.

Table 1.

Haplotype and Associated ABO Blood Type

ABO Blood Type	Haplotype
	rs507666	rs687289	rs8176704	rs8176749
O	G	G	G	C
O	A	G	G	C
O	G	G	G	T
A1	G	A	G	C
A1	A	A	G	C
A2	G	A	A	C
B	G	A	G	T
B	A	A	G	T

Outcome

Participants or proxies were contacted every 6 months by telephone to ascertain stroke or stroke symptoms. Medical records, including neuroimaging and other diagnostic reports were retrieved and centrally reviewed by physicians to confirm the diagnosis, stroke type, and possible etiology. Final adjudication was based on the World Health Organization’s definition of stroke and/or imaging results consistent with a stroke[14, 19]. For this analysis all strokes adjudicated by September 1, 2011 were included.

Definitions

Race was determined by participant self-report as black or white. Baseline stroke was based on participant self-report. Diabetes mellitus was defined as a fasting glucose ≥126mg/dL (≥7.0 mmol/L), a nonfasting glucose ≥200mg/dL (≥11.1mmol/L), or self-reported use of diabetes medications. Atrial fibrillation was defined by baseline electrocardiogram or self-report of a physician diagnosis of atrial fibrillation. Left ventricular hypertrophy was assessed by standard criteria from the electrocardiogram. Baseline cardiovascular disease was defined as participant self-report of baseline coronary heart disease or peripheral artery disease which included revascularization procedures of the legs. Systolic blood pressure was defined as the average of two seated measures taken after a 5 minute rest and use of antihypertensive medications was defined by self-report.

Statistical Analysis

We performed a stratified case-cohort study within REGARDS using methods proposed by Prentice[20]. All stroke cases (n=646) and a stratified cohort random sample (n=1,104) were included (by design, some incident stroke cases are in the cohort random sample). The cohort random sample was stratified by age (20% 45-54, 20% 55-64, 25% 65-74, 25% 75-84, and 10% ≥ 85), race (50% black, 50% white), and gender (50% male, 50% female) to maximize power to detect age and racial differences[21]. Correlates of blood type were determined accounting for the sample weighting in the cohort random sample. Categorical variables were assessed using the Rao-Scott χ² statistic or weighted logistic regression with 95% confidence intervals calculated using the Taylor series with finite population correction. Weighted means were calculated for continuous variables for the cohort random sample and the differences in the means between blood types A, B, and AB with blood type O assessed using weighted Analysis of Variance.

Cox proportional hazard models were used to calculate the hazard ratios (HR) with robust sandwich estimators to calculate the 95% confidence intervals (CI) for blood type with stroke accounting for population weighting and blood type probabilities. Models were adjusted for 1) demographics (age, sex, race, age by race interaction term, and region) and 2) Framingham stroke risk factors (demographic model plus systolic blood pressure, taking antihypertensive medications, diabetes, current smoking, cardiovascular disease, atrial fibrillation, left ventricular hypertrophy). To evaluate mediation of the association of blood type with stroke by factor VIII, factor VIII was added to the Framingham adjusted model. The 95% CI of the percent mediation was calculated by bootstrapping with 1000 replicate samples. The percent of the association of black race with stroke mediated by blood type was assessed by examining the coefficients for race before and after inclusion blood type in a model at ages 45, 55, 65, 75, and 85 (due to the previously described age by race interaction for stroke risk) and the 95% CI of the percent attenuated calculated using bootstrapping with 1000 replicate samples[14, 19]. Sensitivity analyses were performed removing hemorrhagic strokes from the analysis, dividing blood type A into subtypes A1 and A2, and excluding individuals where no blood type probability was ≥90%. The study was approved by the Institutional Review Boards of all participating institutions.

Results

Over a median of 5.8 years (interquartile range 4.1, 7.0), 646 strokes occurred in participants free of stroke at baseline and 989 of 1,104 individuals in the cohort random sample were free of stroke at baseline (baseline stroke = 87 participants and 3 with missing information on baseline stroke) and did not have a stroke during follow-up (incident stroke = 25 individuals). Genotyping predicted blood type with ≥90% certainty in 1,619 of 1,725 participants (94%) included in this analysis. In the cohort random sample, 43.0% of participants were blood type A, 13.5% were blood type B, 3.9% were blood type AB, and 39.6% were blood type O (Table 2). Blacks had a higher prevalence of blood type B (18.3% vs. 10.2%) and AB (6.4% vs. 2.2%) than whites (Table 2, p <0.001). In those with incident stroke vs. those without incident stroke, there was a trend for an increased prevalence of blood type AB (6.1% vs. 3.6%), overall p = 0.08.

Table 2.

Prevalence of ABO Blood Type by Race and Stroke Status in REGARDS

		ABO Blood Type‡
		A		B		AB		O
	N	Weighted N (%)	N	Weighted N (%)	N	Weighted N (%)	N	Weighted N (%)
Cohort Random Sample	464	12,747 (43.0%)	158	4,011 (13.5%)	48	1,155 (3.9%)	434	11,719 (39.6%)
Blacks*	230	5,182 (42.6%)	106	2,227 (18.3%)	33	777 (6.4%)	183	3,992 (32.8%)
Whites	234	7,565 (43.3%)	52	1,784 (10.2%)	15	378 (2.2%)	251	7,727 (44.3%)
Incident Stroke†	283	283 (43.9%)	89	89 (13.8%)	39	39 (6.1%)	234	234 (36.3%)
No Incident or baseline Stroke	409	11,513 (42.5%)	143	3,722 (13.8%)	41	985 (3.6%)	396	10,841 (40.1%)

^*p <0.001 for black versus white, global Rao-Scott χ².

^†p = 0.08 for stroke versus no stroke in those without stroke at baseline, global Rao-Scott χ². Some individuals in the cohort random sample without stroke at baseline developed incident stroke. Rounding due to study design results in numbers of strokes not adding precisely to 646 strokes.

^‡Weighted numbers represent projections for absolute numbers and percentages for the entire cohort accounting for stratified case-cohort design and the sampling frequency

Blood type was not associated with cardiovascular risk factors except for age, with participants of blood type B being younger than those of blood type O and a higher prevalence of diabetes in participants of blood type AB vs. the other blood types (all p < 0.01) (Table 3). These associations became non-significant after adjusting for race (p = 0.16 for diabetes and p = 0.22 for age). Factor VIII was strongly associated with blood type with participants with blood type O having the lowest mean factor VIII levels (105%; 95% CI 102%, 109%) (Table 3). Compared to participants with blood type O, mean factor VIII levels were 42% higher (95% CI 25%, 58%) in participants with blood type AB, 33% higher (95% CI 26%, 40%) in participants with blood type B, and 22% higher (95% CI 16%, 27%) in participants with blood type A. These associations remained significant after adjusting for age, sex, and race (all p <0.001).

Table 3.

Baseline Associations of Cardiovascular Risk Factors with ABO Blood Type in the REGARDS Cohort Random Sample

	Blood Type
	A	B	AB	O	P*
Age (years; mean, 95% CI)	64.9 (64.2, 65.6)	63.6 (62.3, 64.9)	66.8 (64.7, 69.0)	65.4 (64.7, 66.2)	0.004
Male (n, %)†	230 (45.8%)	84 (46.4%)	21 (29.3%)	217 (45.1%)	0.28
Obesity (n, %)†
Underweight	5 (1.4%)	2 (1.4%)	0 (0.0%)	6 (0.8%)	0.27
Normal	144 (28.2%)	35 (21.6%)	8 (19.7%)	116 (22.0%)
Overweight	163 (36.5%)	53 (32.8%)	22 (40.1%)	148 (35.3%)
Obese	150 (33.8%)	65 (44.2%)	16 (40.2%)	159 (41.9%)
Diabetes (n, %)†	94 (20.7%)	32(19.1%)	18 (46.3%)	89 (22.7%)	0.005
Current Smoker (n, %)†	71 (15.6%)	27 (16.5%)	4 (7.4%)	56 (10.9%)	0.12
Cardiovascular Disease (n, %)†	98 (18.4%)	22 (13.8%)	14 (27.9%)	79 (17.9%)	0.27
Heart Disease (n, %)†	94 (18.1%)	19 (11.1%)	12 (22.5%)	74 (17.3%)	0.23
Atrial Fibrillation (n, %)†	40 (8.1%)	11 (8.6%)	3 (7.8%)	41 (11.1%)	0.56
Left Ventricular Hypertrophy (n, %)†	39 (2.9%)	18 (11.1%)	5 (9.9%)	46 (9.7%)	0.29
Hypertension Medications (n, %)†	279 (56.8%)	93(57.6%)	33 (67.9%)	255 (57.5%)	0.66
Systolic Blood Pressure (mmHg; mean 95% CI)	127 (125, 128)	126 (123, 128)	132 (126, 138)	127 (126, 129)	0.11
Factor VIII (%; mean, 95% CI)	127 (123, 132)	139 (131, 146)	147 (127, 168)	105 (102, 109)	<0.001

^*Rao Scott χ² for categorical variables and p-values for linear variables were calculated using weighted linear regression.

^†n’s represent the actual number while percents represent with weighting of the cohort random sample

In a basic model adjusted for age, race, region, and an age by race interaction term, blood types A or B were not associated with stroke compared to blood type O (Table 4). By contrast, blood type AB was associated with a HR of 1.80 for stroke compared with blood type O (95% CI 1.04, 3.12). The association of blood type AB with stroke was slightly higher after adjusting for Framingham stroke risk factors (Table 4). In Model 2, when factor VIII was added to the model as a continuous variable, the association of blood type with stroke decreased. For blood type AB, factor VIII attenuated 60% (95% CI 8%, 97%) of the association with stroke versus blood type O, decreasing the HR from 1.83 to 1.33 (data not shown).

Table 4.

Risk of Stroke for different ABO Blood Types in REGARDS

	Blood Type (Hazard Ratio, 95% CI)
	A	B	AB	O
Model 1*	1.22 (0.94, 1.57)	1.25 (0.87, 1.81)	1.80 (1.04, 3.12)	Reference
Model 2†	1.15 (0.87, 1.53)	1.27 (0.87, 1.87)	1.83 (1.01, 3.30)	Reference

^*Model 1: Age, sex, race, region, age by race interaction term

^†Model 2: Model 1 plus systolic blood pressure, taking antihypertensive medications, diabetes, current smoking, cardiovascular disease, atrial fibrillation, and left ventricular hypertrophy (Framingham Stroke Risk Factors)

There was no evidence of an interaction between blood type and any of the risk factors listed in table 3 including race (all p > 0.10) except for diabetes (p-interaction = 0.02 in Model 2). Blood type AB compared to O was not associated with stroke risk in those with diabetes (HR 0.49; 95% CI 0.17, 1.44) but was a stronger risk factor in those without diabetes (HR 3.33; 95% CI 1.61, 6.88) (Table 5). Figure 1 graphically presents the hazard ratio of black versus white race with stroke with and without blood type in Model 2 at various ages. While the hazard ratio for black race with stroke decreased slightly at all ages, this was not significant at any age with the 95% CI of the percent of the association mediated including 0%.

Table 5.

Association of ABO Blood Type with Stroke Stratified by Diabetes Status^*

Blood Type†	Diabetes (HR, 95% CI) (Incident Strokes = 173)	No Diabetes (HR, 95% CI) (Incident Strokes = 449)
A	1.33 (0.72, 2.45)	1.10 (0.80, 1.52)
B	1.11 (0.51, 2.45)	1.31 (0.83, 2.05)
AB	0.49 (0.17, 1.44)	3.33 (1.61, 6.88)
O	Reference	Reference

^*Adjusted for Covariates in Model 2: Age, sex, race, region, age*race, systolic blood pressure, taking antihypertensive medications, diabetes, current smoking, cardiovascular disease, atrial fibrillation, and left ventricular hypertrophy (Framingham Stroke Risk Factors)

Figure 1.

Mediation of Increased Hazard Ratio of Black vs. White Race for Stroke by ABO Blood Type^*

^*Solid black lines represent hazard ratios (solid grey lines represent 95% CI) without blood type in the model and dashed black lines represent hazard ratios (dashed grey lines represent 95 % CI) with blood type in the model at the ages listed on the horizontal axis. Adjusted for covariates in Risk Factor Model: Age, sex, race, region, age*race, systolic blood pressure, taking antihypertensive medications, diabetes, current smoking, cardiovascular disease, atrial fibrillation, and left ventricular hypertrophy. Blood type did not significantly mediate any of the association of black race with stroke at any age group (all p >0.05).

When hemorrhagic strokes (n = 78) were removed in a sensitivity analysis, the association of blood type AB with stroke increased slightly (HR 1.96; 95% CI 1.08, 3.57) with minimal differences in the associations of blood types A and B (adjusted for risk factors in Model 2). The HR for stroke adjusted for risk factors in Model 2 for blood type A1 (HR 1.25; 95% CI 0.92, 1.70) and A2 (HR 1.03; 95% CI 0.67, 1.59) vs. O was similar to that of blood type A vs. O. There were too few individuals with blood type A2B (n = 4) for meaningful comparisons of different A alleles in blood type AB and stroke risk. When individuals with no blood type probability ≥0.90 were excluded from the analysis, results were similar with blood type AB having an increased risk of stroke (HR 1.83; 95% CI 1.00, 3.35).

Discussion

Blood type AB was associated with stroke risk independent of conventional risk factors with a HR 1.83. The association was stronger in participants without diabetes and despite a higher prevalence of blood type AB in African-Americans, blood type did not significantly mediate any of the association of race with stroke. Factor VIII mediated 60% of the association between blood type AB and stroke risk.

To our knowledge, this is the first report in the medical literature demonstrating that blood type is associated with stroke after adjusting for race, stroke risk factors, and factor VIII levels. In an analysis of stroke cases and controls from a health maintenance organization in the United States, blood type B was associated with ischemic stroke (OR 1.59; 95% CI 1.17, 2.17) but not blood type AB (OR 0.93; 95% CI 0.52, 1.65)[10]. These data were subject to selection bias as cases were all either treated for hypertension or were peri-menopausal women. In a genome-wide association study of 1,544 strokes and 19,602 controls, the ABO locus was not associated with stroke risk though in a more focused analysis of genes associated with coagulation factors and fibrin structure/function a SNP in the ABO locus was associated with ischemic stroke. However, the authors did not do a haplotype analysis and report on the association of blood type with stroke[12, 22]. Other associations noted for blood type and vascular disease include the association of blood groups O and A2 with a reduced risk of VTE[11], the association of non-O blood group with an increased risk of VTE[2], and the association of blood types A1, B, and AB with an increased risk of VTE[10]. Blood group has also been associated with myocardial infarction with increased risk seen with blood type A1[10] and AB[9].

The reason for the association of blood type with vascular disease is unknown. The B haplotype is more common in populations of African and Asian ancestry, and may be a surrogate for risk factors which segregate with race[1]. This was seen in our analysis where the association of blood type B with younger age and blood type AB with diabetes was no longer significant when we adjust for race – as blacks were slightly younger and have a higher prevalence of diabetes in REGARDS. Another hypothesis could be individuals with non-O blood type have higher levels of factor VIII and other procoagulant proteins (such as VWF) associated with vascular risk[5, 23]. The A and B antigen are also found on VWF and enzymatic removal of these antigens from VWF has been found to decrease VWF activity in vitro[24]. The impact of factor VIII levels on the association of blood type with vascular disease has not been previously studied except in venous thrombosis where factor VIII levels reduced the odds ratio for VTE from 1.65 to 1.31 but did not mediate the entire association of blood type with venous thrombosis[2]. The residual association may be due to intraperson variability in factor VIII levels not accounted for with one measurement, the effect of ABO antigen expression on circulating VWF or endothelial cells, or other mechanisms yet to be elucidated[1, 24]. There are no prior studies on a differential association of blood type and vascular disease risk by diabetes status. Our original hypothesis was that blood type would have a neutral effect or potentiate the associations of traditional cardiovascular risk factors with stroke. The hyperglycemia associated with diabetes results in non-enzymatic glycosylation of the red cell membrane structures and other plasma proteins and one potential hypothesis is this non-enzymatic glycosylation could perturb the impact of blood type (determined by enzymatic glycosylation of the red cell H-antigen) on stroke risk[25]. Further study is needed on this issue.

While AB-blood type is associated with an 83% increase in stroke risk, given a population prevalence of AB blood type of 3.9%, we were not surprised that adjustment for blood type did not mediate the black-white differences in stroke risk. Even a large increase in risk in such a small proportion of the population could never account substantial black-white differences in stroke. The role of AB blood-type as a contributor to the racial differences in stroke is further complicated by the post-hoc observation of a potential differential effect of blood type AB among diabetics (where it was non-significantly protective) and non-diabetics (where it even more substantially increased risk). While blacks were more likely to have blood type AB, they are also more likely to be diabetic; and the larger proportion of blacks with diabetes may imply that (on average) blood type AB could be less harmful for blacks than for whites. Our data suggest that the 3.9% of the population in the U.S. with blood type AB is at substantially higher stroke risk; however, this is a relatively small portion of the general population, and as such blood type AB is not a major contributor to stroke risk in the general population (however, it remains quite important to the individuals with the characteristic).

Our main limitations are that we genotyped blood type and we did not measure some covariates of interest such as VWF. The greatest potential source of error is misclassifying participants who are phenotypically blood type O as blood type A1[18, 26, 27]. Blood type O results from a non-functional glycosyltransferases, with most but not all individuals having a unique frame-shift mutation[26]. Rarer mutations in the catalytic domain of the enzyme will also result in the phenotype of blood type O, but we would misclassify these individuals as blood type A1[26]. This misclassification would bias our results towards the null for individuals with blood type AB and A, but would not affect associations for participants genotyped as blood type O and B. Sensitivity analyses separating out blood type A1 from A2 did not change our results. Von Willebrand factor is the carrier protein for factor VIII and the levels of VWF and factor VIII are highly correlated. There is much debate as to whether vascular disease risk relates to VWF levels, factor VIII levels, or if they act independently [5-7, 23, 28, 29]. Unfortunately REGARDS sample collection methods precluded accurate measurement of VWF levels. Another weakness as well as strength is that REGARDS participants are scattered throughout the United States and there was no systemic approach during the stroke event to determine stroke type; in a recent analysis, the ABO gene locus was associated with large vessel and cardio-embolic stroke but not small vessel disease[12].

In conclusion, we demonstrated that blood type AB was associated with stroke risk in a large, biethnic population, and that 60% of the association was mediated by factor VIII levels. Further, despite differences of blood type by race, little of the association of race with stroke was mediated by blood type. Further study of diverse populations is needed to fully determine the association of blood type with vascular disease risk and whether more aggressive risk factor modification is warranted in these individuals.