Edinburgh Research Explorer Association of apolipoprotein E with intracerebral hemorrhage risk by race/ethnicity

IMPORTANCE Genetic studies of intracerebral hemorrhage (ICH) have focused mainly on white participants, but genetic risk may vary or could be concealed by differing nongenetic coexposures in nonwhite populations. Transethnic analysis of risk may clarify the role of genetics in ICH risk across populations.

S pontaneous intracerebral hemorrhage (ICH) is the most severe form of stroke. In the United States, 160 000 people experience an ICH each year with a case fatality rate of 54% at 1 year. 1 The prevalence of ICH has increased 47% between 1990 and 2010, 2 and ICH risk appears to vary among white, black and Hispanic populations. [3][4][5][6] Compared with white individuals, young and middle-aged black individuals have an almost 2-fold increased risk for ICH. 3,4 Similarly, Hispanic individuals have a relative risk increase that ranges from 1.4 for lobar ICH to 3.7 for nonlobar ICH. 5 Moreover, not only is hypertension prevalence among older adults lower among non-Hispanic white groups (76.3%) than among non-Hispanic black (82.5%) and Hispanic (79.2%) groups, but the risk of ICH in the presence of hypertension increases more than 50% from white to Hispanic groups. [7][8][9] The associations of genetic and acquired ICH risk factors with these observed risk differences are poorly understood.
Previous studies conducted in predominantly Europeanancestry populations have demonstrated that apolipoprotein E (APOE [OMIM 107741]) ε2 and ε4 alleles potently increase risk of lobar ICH. 10 In Alzheimer disease, another disorder associated with APOE ε allele status, the degree of risk contributed by APOE genotype varies substantially by the ancestry of the population studied. Among non-Hispanic white people, homozygous carriers of APOE ε4 exhibit up to a 12-fold higher risk of Alzheimer disease, but this same haplotype exerts little or no risk for black or Hispanic people. [11][12][13] Understanding how genetic risk factors vary across race/ ethnicity may highlight novel underlying disease mechanisms and identify populations who may be particularly responsive to specific prevention strategies, as has previously been shown in treatment response for heart failure by race/ethnicity. 14 Unfortunately, with individuals of African American and Hispanic ancestry representing less than 4% of all samples in genome-wide association studies, only recently has it become possible to study genetic risk of common disease across representative US populations. 15 We tested the associations of APOE ε alleles with risk of lobar and nonlobar ICH among white, black, and Hispanic individuals, using direct genotyping data supplemented by genome-wide genotyping as available in cases and controls from the International Stroke Genetics Consortium. Because these analyses revealed substantial heterogeneity by race/ethnicity, we further explored the degree to which the differential burden of hypertension across populations is associated with the variability in observed APOE risks.

Participating Studies and Data Collection
Case and control participants included in the study were gathered from 3 multicenter studies in the United States and from 8 European sites participating in the International Stroke Genetics Consortium, according to availability of directly ascertained APOE ε genotypes and a harmonized local acute case recruitment scheme. These participants were enrolled in the aforementioned studies from January 1, 1999, to December 31, 2017. Institutional review board approval was obtained from all participating centers, and informed consent was obtained from all participants or their legally authorized representative.
The ICH cases from population-based cohorts were not included because of potential imbalances in lethal case recruitment between the 2 sampling approaches. 16 Studies included the Genetics of Cerebral Hemorrhage with Anticoagulation (GOCHA) study, 17 25 Because of variable sample sizes from contributing centers, data from European studies were analyzed together for association testing in a meta-analysis (International Stroke Genetics Consortium Europe), as done previously. 26,27 Participants with secondary causes of ICH were excluded from enrollment. More specific inclusion and exclusion criteria for each of the included studies are reported in eTable 1 in the Supplement. Demographic variables, including selfidentified race/ethnicity, 8 were systematically obtained from structured patient and family member interviews within each site, 19,28 along with additional covariates. 29 Computed tomographic images on admission were analyzed at each participating site for classification as lobar (involving predominantly the cortex and underlying white matter) and nonlobar (involving predominately the basal ganglia, periventricular white matter, or internal capsule) according to prespecified criteria. 26,27 The APOE genotype was centrally determined according to standard procedures. 30 Genomewide data were available for a subgroup of participants. Genetic and bioinformatic analysis followed standardized prespecified qualitycontrol procedures 31 (eMethods in the Supplement).

Population Stratification
Fifteen ancestry informative markers were selected from participants with direct or genomewide genotyping and subjected to principal component (PC) analysis in accordance with

Key Points
Question Is history of hypertension and apolipoprotein E (APOE) associated with intracerebral hemorrhage risk in participants stratified by self-reported race/ethnicity? Findings In this case-control study of 13 124 adults, having a copy of APOE ε4 alleles increased the risk for lobar intracerebral hemorrhage only in white individuals, but after propensity score matching for hypertension burden, Hispanic individuals showed the same risk of APOE ε4.
Meaning APOE ε4 appears to be confirmed as a risk factor for lobar intracerebral hemorrhage in nonwhite populations but is masked by differential hypertension burden in Hispanic individuals; further studies are needed to explore the interactions between APOE alleles and environmental exposures. previously published methods. [32][33][34][35] The first 4 PCs were included in regression models to adjust for population stratification in this subgroup. This PC analysis was not used to reclassify participants, as self-identified race/ethnicity may capture exposures that transcend genetic ancestry and could help explain the stratification among different populations. A complete description of the genetic analysis, the participants genotyped, and the markers selected is available in eTable 2 in the Supplement.

Statistical Analysis
Categorical variables were expressed as count (%), and continuous variables were expressed as median (interquartile range [IQR]) or mean (SD), as appropriate. Categorical variables were compared using the 2-tailed χ 2 test, whereas continuous variables were compared with unpaired Mann-Whitney tests. The threshold for statistical significance was set at P = .05. We tested APOE allele association with ICH risk using 3 logistic regression models. Model 1 was adjusted for age, sex, and history of hypertension. Model 2 included variables from model 1 in addition to history of hypercholesterolemia; history of ischemic stroke; warfarin, statin and antiplatelet use; smoking; and alcohol use. Model 3 also included variables from model 1 with the addition of the first 4 principal components derived from ancestry-informative genotypes. APOE risk allele status was modeled as 2 variables, ε2 and ε4, coded for allele counts (0, 1, or 2 for each) in an additive model referent to the wildtype ε3 allele. 17 Analyses were performed in lobar and nonlobar ICH, given the known differences in underlying biology between the 2 ICH locations. 36 All statistical analyses were performed using Stata, version 13.0 (Stata Corp LLC), and R statistical software (R Foundation for Statistical Computing).

Transethnic Meta-analysis
We applied a 2-stage clustering approach for meta-analysis, based on race/ethnicity and stratified by study. 37 Cases and controls in each study were divided into black, white, and Hispanic groups, based on self-identified race/ethnicity. Each race/ethnicity group within each study was allocated to the same cluster and tested using the regression models described above. Individual cluster results were presented graphically by plotting OR estimates on a forest plot to visually assess heterogeneity. The effect sizes obtained were then used for a DerSimonian-Laird random-effects, inverseweighted nonparametric meta-analysis. 38 Cochran Q and I 2 tests were used to quantify heterogeneity.

Propensity Score Modeling of APOE and Hypertension
To address imbalances in the burden of hypertension across ICH populations, and associated imbalances of baseline characteristics among participants with and without hypertension, 2 propensity score (PS) analyses were performed using the nearest neighbor matching method to compare participants of similar underlying hypertensive pathophysiologic burden. 39,40 The first PS analysis was constructed using history of hypertension and included variables of age, sex, and self-identified race/ethnicity. The second PS analysis, leveraging data only available in the ERICH study, contained the same variables as the first PS analysis, in addition to the number of medications prescribed to treat hypertension as well as systolic and diastolic blood pressure readings at ICH presentation. Propensity score results were used in a logistic regression model for ICH risk identical to model 1, as described. In a sensitivity analysis, the same PS procedure was tested against age (older or younger than 65 years), sex, and hypercholesterolemia to increase the confidence that the PS findings were specific for hypertension.

Power Calculation
Using empirical data from our analyses, we performed a post hoc calculation of the statistical power to detect an association of APOE ε alleles with lobar ICH risk in black and Hispanic participants, commensurate with the effect size detected in white participants. Type I error rate of 0.05, log additive inheritance mode, and 0.01 of population risk were assumed, with analyses performed using Quanto software, version 1.

Lobar ICH
We analyzed 2305 lobar ICH cases from all studies. Model 1 confirmed the previously reported association of APOE ε2 (pooled OR, 1.49; 95% CI, 1.24-1.80; P < .001) and APOE ε4 (pooled OR, 1.51; 95% CI, 1.23-1.85; P < .001) with ICH risk; however, within self-identified Hispanic and black groups, no associations were found ( Figure 1). Model 2 was used to interrogate the independent association of APOE alleles with ICH, controlled for established ICH factors ( Figure 2). Here, APOE ε2 and ε4 alleles retained an association with lobar ICH. As with model 1, this association was observed in the white group, but not in black or Hispanic groups (for APOE ε2 OR, 1.45 [95% CI, 1.04-2.03; P = .03]; for APOE ε4 OR, 1.51 [95% CI, 1.14-1.99; P = .004]). Model 3 considered population stratification  Figure 3). In the white group, both APOE ε2 (OR, 1.81; 95% CI, 1.33-2.45; P < .001) and APOE ε4 (OR, 1.80; 95% CI, 1.33-2.44; P < .001) conferred higher risk for lobar ICH. For APOE ε4 alone, we found a similar association in the Hispanic group, suggesting that population stratification may have played some role in the lack of ε4 association found in models 1 and 2, particularly for the large and ethnically diverse Hispanic population recruited through the ERICH study. In contrast, for the black population, neither APOE ε2 nor APOE ε4 conferred a statistically significant risk for lobar ICH after controlling for population structure.

Nonlobar ICH
We analyzed 3897 nonlobar ICH cases ( Figure 1). In model 1, APOE ε2 and ε4 did not show an association with nonlobar ICH risk, across any of the self-identified race/ethnicity groups. When comparing nonlobar ICH cases with controls, we found that the APOE ε4 P values were all P > .10. For model 2 and model 3 in nonlobar ICH, again neither APOE ε2 nor APOE ε4 showed an association with disease risk across all the studies and racial/ethnic groups ( Figure 2 and Figure 3).

Power Calculation (Lobar ICH)
Given the differences in sample sizes between white, black, and Hispanic groups, we performed post hoc power calculations to determine whether the study was powered to detect a comparable APOE association in the smaller populations of blacks and Hispanics. Given the frequency of APOE ε4 in black participants (ε4 frequency 37.7%), the sample size (assuming an unmatched case-control ratio of 1:1) would provide 99% power to detect an ε4 association similar to the lower bound of the 95% CI seen in white participants (OR, 1.43). Our analyses of APOE ε4 association in Hispanic participants were similarly powered at 90%. Furthermore, the APOE ε2 frequency in black participants (19.9%) at the reported sample sizes would provide 93.8% power to detect the lower bound of the association seen in white participants (OR, 1.38). For Hispanic participants, given the lower APOE ε2 frequency (0.8%), 80% power would be achieved at a slightly higher effect size (OR, 1.60) but still would be below that found in white participants.

Propensity Score Modeling for Hypertension
We used a PS analysis to attempt to isolate the association of APOE against the imbalanced burden of hypertension across race/ethnicity. In the first PS, we selected case and control participants with a balanced hypertension burden to form a group comprising individuals of white, black, or Hispanic ancestry. In this matched and homogeneous group, we were able to detect an association of APOE ε4 with lobar ICH risk among Hispanic participants (OR, 1.14; 95% CI, 1.03-1.28; P = .01) but not in black participants (OR, 1.02; 95% CI, 0.98-1.07; P = .25). Results were confirmed in the second PS analysis performed only in the ERICH data set, which included hypertension diagnosis as well as additional hypertension severity variables, including number of medications used to treat hypertension and systolic and diastolic blood pressure readings ( Figure 4).

Discussion
Although APOE associations with ICH risk have been characterized in multiple previous studies and meta-analyses for race/ethnicity, APOE ε4 emerges as a risk factor for lobar ICH among self-identified Hispanic individuals. These results highlight the challenges of generalizing genetic risk factors across ancestries, where nongenetic exposures are known to vary by race/ethnicity. In Alzheimer disease, the relative risks for Hispanic or black individuals associated with an APOE ε 4 allele become progressively weaker or disappear entirely in comparison with white individuals. [42][43][44][45][46] In ICH, APOE ε alleles have already been shown to exert higher risks in East Asian individuals when compared with those of European ancestry. 47 Although recent analyses by Sawyer et al 48 demonstrate the association of hypertension and APOE ε allele status with ICH risk across race/ethnicity specific to the ERICH study, the present analysis has the advantage of a larger sample size via formal transethnic meta-analysis as well as a PS matching approach that helps to illustrate the potential mechanisms underlying the observed variability of APOE ε alleles on lobar ICH risk across populations.
The observed differences in association between APOE ε alleles and lobar ICH risk do not provide direct evidence that the biological risks of the APOE gene or associations with underlying cerebral amyloid angiopathy (CAA, a major cause of lobar ICH) are necessarily different across racial/ethnic boundaries. It would seem more likely that genetic and/or environmental risk exposures covarying with race/ethnicity exert a role in modifying or mitigating the underlying APOE genetic risks. The PS analysis supports this conjecture, demonstrating that hypertension, the most important known risk factor for ICH, may simply obscure the APOE risks that may be common across ancestries. Aside from variation in environmental risk exposures, variants in a modifier gene (or genes) that differ across populations may alter the biological risk of APOE and consequently vary ICH risk, as has been hypothesized for Alzheimer disease. 49,50 Furthermore, genetic variants that are racially stratified and not associated with APOE may directly modify the risk of ICH. This hypothesis represents an alternative explanation for why the PS matching for hypertension only partially remediated the association of APOE ε4 with ICH risk in Hispanic participants and had little to no association in black participants. APOE interaction studies and transethnic genome-wide association studies for ICH will likely provide insights on these hypotheses. Similarly, analyses of CAA in nonwhite populations will also clarify the association of race/ethnicity with this pathologic pathway.

Figure 2. Forest Plots of Meta-analysis of Apolipoprotein E (APOE) in Lobar and Nonlobar Intracerebral Hemorrhage Cases and Controls in Model 2, Stratified Across Participating Studies and Race/Ethnicity
In this study, we have not attempted to stratify subsets of participants by probable CAA status using magnetic resonance imaging data as has been done in previous metaanalyses. Most studies linking lobar hemorrhage locations to the pathologic diagnosis of CAA were performed in largely white populations. 51 As such, widely accepted criteria for classifying probable and possible CAA using hemorrhage location and microbleed counts have not been validated in  Previously demonstrated associations between APOE ε4 and nonlobar ICH risk, also seen in nonlobar ICH recurrence, 17,53 were not replicated in this study. Potential explanations include a higher rate of participants affected by hypertension and an overall younger age of participants in this study. These factors may reflect the associations of environmental or non-APOE genetic exposures in younger populations with nonlobar ICH in particular. Demographic heterogeneity is also higher in this study, and the reduced availability of covariates such as steady state lipid levels 53 for risk modeling may have affected this finding. In addition, a previous metaanalysis of APOE risks in ICH also did not show the association between APOE ε4 and nonlobar ICH in black individuals, a finding supported by the present analyses. 17 Future studies in larger data sets with well-phenotyped cases are needed to further elucidate the potential role of APOE ε4 in nonlobar ICH.

Strengths and Limitations
The targeted enrollment of Hispanic and black individuals through the ERICH study, lacking in previous reports, 17,54 is an important strength of the present study. The high number of nonwhite individuals enrolled permits well-powered analyses in these populations and promotes confidence that the lack of observed associations is not a false acceptance of the null hypothesis, as supported by our post hoc power calculations.
Some limitations of the study should be acknowledged. Diagnoses of comorbidities were based on self-identified attestation and therefore affected by patient or caregiver awareness. This concern is present in both cases and controls, however, and internal consistency between diagnoses and prescribed medications helps to limit this potential source of bias. Furthermore, the propensity score was based on variables that only partially captured the complex phenotype represented by hypertension. However, this lack of information content is likely to bias our score results toward the null; we expect that a more precise index of hypertension burden would have increased our ability to normalize this phenotype across race/ethnicity and to demonstrate even more homogeneous APOE ε4 risks. Finally, genomewide genotypes for the ERICH study participants are not currently available, preventing us from determining whether additional genetic exposures modify the association of APOE with ICH risk across race/ethnicity.

Conclusions
In this meta-analysis, APOE ε2 and ε4 alleles remain genetic risk factors for lobar ICH, but these results are largely driven by the associations in white individuals. However, the results support a biological risk of APOE ε4 alleles that seems to transcend ancestral backgrounds, 55 albeit with varying effect because of the presence of racial/ethnic disparities across associated risk factors. As availability of genetic data on US minority populations continues to increase, it is hoped that improved modeling of covarying genetic and nongenetic exposures in these populations will provide new insights into treatment and prevention strategies in ICH that maximize the potential advantages for all.