Our data add to the growing body of literature assessing risk factors for ICH and lend support to the concept that lobar and nonlobar ICH result from different pathophysiologic factors.
We found hypercholesterolemia less frequently in nonlobar ICH cases compared to controls, but found no difference between lobar ICH cases and controls. The protective association between hypercholesterolemia and all ICH is supported by several studies.8,14–16
Our results suggest that this finding is driven specifically by the association of hypercholesterolemia with nonlobar ICH. The Rotterdam Scan Study found that those with the highest quartile of triglycerides had the lowest rate of deep or infratentorial microbleeds; no such relationship was seen with lobar microbleeds.17
Hypercholesterolemia may therefore play a more relevant pathophysiologic role in maintenance of deep penetrating arterioles than in cortical arterioles. The mechanism of hypercholesterolemia's association with nonlobar ICH is unclear, but low cholesterol has been hypothesized to contribute to vascular fragility.18
We found no association between statin use and ICH, regardless of hemorrhage location or APOE
genotype. This lack of association would seem to be at odds with the results of the Stroke Prevention by Aggressive Reduction in Cholesterol Levels trial, which found an increased risk of ICH among subjects randomized to atorvastatin 80 mg vs placebo.19
Further research on statin dosage and target low-density lipoprotein would help clarify the relationship between statin usage and ICH.
We found self-reported hypertension more frequently in ICH cases compared to controls. Nonlobar ICH cases appear to drive this association, for we found no such association with lobar ICH. Other studies also found hypertension more frequently in nonlobar compared to lobar ICH, but did not collect hypertension frequency in the population as a whole, preventing calculation of attributable risk.9,11
In the current analysis, we determined the attributable risk for hypertension in nonlobar ICH, and found the attributable risk to be significantly higher for black than for white subjects. The higher attributable risk in black subjects is related in part to the higher frequency of hypertension in black controls, but may also be related to blood pressure control, which could not be determined from our data. Given the relatively small sample size of nonlobar ICH among black subjects, variables beyond hypertension may not have reached significance due to limited power. The absence of such variables in a multivariate model may overestimate the influence of hypertension as a risk factor among black subjects.
alleles ε2 and ε4 appear to play a role in the pathogenesis of amyloid angiopathy.20,21
Most studies have found APOE
alleles ε2, ε4, or both more frequently in lobar ICH cases, although other studies have found no association.10,12,22–25
The largest and most detailed analysis of APOE
genotype and ICH was a meta-analysis that included our cases as well as those from 6 other studies. That study found that alleles ε2 and ε4 were both associated with lobar ICH in white subjects, most strongly in subjects with a high probability of cerebral amyloid angiopathy.10
Here we similarly found APOE
alleles ε2 and ε4 to be associated with lobar ICH.
Our analysis confirms that use of warfarin is associated with both lobar and nonlobar ICH.26
Warfarin use increased the odds of ICH in our present analysis by 3- to 4-fold; however, the frequency of warfarin use in the population is relatively low so the attributable risk for anticoagulation was less than 10%. The frequency of anticoagulation has increased over time,27
as studies have consistently demonstrated the superiority of warfarin over antiplatelet agents for stroke prevention in high-risk patients with atrial fibrillation.28,29
The risk of anticoagulant-associated ICH in recent clinical trials has ranged from 0.38 to 0.47% per patient year, and was lower (0.10%–0.24% per patient year) for the newer anticoagulants dabigatran and apixiban.30,31
Our results can serve as a baseline for ongoing and future studies of ICH in the setting of anticoagulation. As the newer anticoagulants are adopted, the per-person risk of anticoagulant-associated ICH would be expected to decline if real-world experience with these agents mirrors that of the clinical trials.
Alcohol intake has been variably reported to be associated with ICH.15,16
Here we found a small protective association of moderate alcohol consumption (≤2 drinks/day) with ICH. This association was no longer significant when analyzed by ICH location, likely reflecting loss of power after stratification. We found no association of heavy alcohol use with ICH. Studies are difficult to compare because of the varying definitions of alcohol use, but alcohol has not been consistently found to play a role in ICH.
Our finding that less than a high school education is associated with approximately double the risk of ICH is consistent with the findings from the National Health and Nutrition Examination Survey (NHANES I) study.32
Conversely, combined analysis of Atherosclerosis Risk in Communities (ARIC) and the Cardiovascular Health Studies (CHS) found only a nonsignificant trend toward higher risk of ICH with lower education levels.15
Low education level is thought to reflect socioeconomic status and associated issues such as health care access,33
thus the strength of its association with disease is likely to vary among populations. Notably the NHANES I study was designed to oversample individuals of lower socioeconomic status, whereas the ARIC and CHS studies were not.
We found no association between smoking and ICH, and this lack of association persisted regardless of hemorrhage location. The ARIC and CHS studies similarly found no association between ICH and smoking, even when stratified by current smoking status and pack-years.15
Conversely, the INTERSTROKE study found a small association between current smoking and ICH.16
Thus, if smoking is a risk factor for ICH, its effect is weak.
Our study is not without limitations. Cohort study design offers advantages over case-control design; however, a cohort study with 597 cases at an incidence rate of 20 per 100,000 per year4
would require 3 million patient years of follow-up. Potential limitations of our case-control study include inadequate capture of cases, survival bias, and recall bias. Comparison with the GCNKSS34,35
suggests that our method captures >90% of primary ICH cases that occur in our region. Despite our aggressive prospective enrollment, the difference in mortality between interviewed and noninterviewed subjects reveals a significant survival bias. Nevertheless, the minimal differences in risk factors between the interviewed and noninterviewed cases support the external validity of our results. In fact, the lower frequency of hypercholesterolemia in noninterviewed cases suggests that the data presented here may be an underestimate of the true effect size of hypercholesterolemia. Similarly, the trend toward more frequent anticoagulation in noninterviewed cases suggests that our results may underestimate the effect size of anticoagulation. Our structured interview assesses variables by self-report, which raises the possibility of recall bias, creating the potential for misclassification of exposure. Furthermore, self-report does not account for the severity of the hypertension or hypercholesterolemia. Sensitivities for self-report of hypertension, hypercholesterolemia, tobacco use, and alcohol consumption in prior studies ranged from 51%–86%, and specificities ranged from 57%–95%.36–38
Cases and controls had risk factor assessment by the same methodology, however, and common atherosclerotic risk factors such as diabetes and cigarette smoking did not show significant differences between cases and controls, which suggests that recall bias was not a significant factor in our analysis.
This is the largest study with population-based ascertainment of ICH cases that analyzes genetic and environmental risk factors by hemorrhage location and provides attributable risk estimates. We confirmed that APOE ε2 or ε4 genotype is associated with lobar ICH, whereas hypertension and high cholesterol are associated with nonlobar ICH, suggesting that lobar and nonlobar hemorrhages result from different pathophysiologic processes. Furthermore, ours is the first study to suggest that the protective association of hypercholesterolemia varies by hemorrhage location. Ongoing studies will further advance our knowledge of risk factors for ICH, and will allow assessment of risk factors by both race and hemorrhage location.