Racial Differences in the Association of Insulin Resistance with Stroke Risk: The REasons for Geographic And Racial Differences in Stroke (REGARDS) Study

Abstract

Background and Purpose

Insulin resistance is associated with increased stroke risk, but the effect has not been adequately examined separately in white and black populations.

Methods

The association of baseline insulin resistance with risk of cerebral infarction (CI) and intracerebral hemorrhage (ICH) was assessed in 12,366 white and 6,782 black participants from the REGARDS cohort, recruited between 2003 and 2007 and followed an average of 5.7 years. Insulin resistance was measured with the homeostasis model assessment (HOMA-IR).

Results

There were 364 incident CI and 41 incident ICH events. The risk for CI increased with the log of insulin resistance in whites (HR_ln(IR) = 1.17; 95% CI: 1.00 – 1.38), but was largely attenuated by adjustment for stroke risk factors (HR_ln(IR) = 1.05; 95% CI: 0.88 – 1.26). There was no association in blacks (HR_ln(IR) = 1.01; 95% CI: 0.81 – 1.25). After adjustment for demographic factors and risk factors, there was a significant difference by race in the association of insulin resistance with risk of ICH (p = 0.07), with a decrease in the risk of ICH in whites (HR_ln(IR) = 0.61; 95% CI: 0.35 – 1.04) but a non-significant increase for blacks (HR_ln(IR) = 1.20; 95% CI: 0.60 – 2.39).

Conclusion

These data support the growing evidence that insulin resistance may play a more important role in stroke risk among white than black individuals, and suggest a potentially discordant relationship of insulin resistance on CI and ICH among whites.

Introduction

There is growing evidence that the impact of insulin resistance on cardiovascular health may differ by race, with a lesser impact in the black population. In the Insulin Resistance Atherosclerosis Study (IRAS) study, insulin resistance (indexed by Bergman’s S_I¹) was related to carotid intimal medial thickness (CIMT) in white and Hispanic participants, but not in blacks.² Likewise in the Multi-Ethnic Study of Atherosclerosis (MESA), the difference in CIMT between 1^st and 5^th quintiles of insulin resistance (by the HOMA-IR model³) was twice as large for whites (0.08 mm) as for blacks (0.04 mm), with tests for trends stronger in whites (p < 0.0001) than for blacks (p = 0.01).⁴ In addition, there was a consistent increased risk for coronary calcium with increased insulin resistance in whites (p < 0.001), but not blacks (p = 0.1),⁴ and insulin resistance remained associated with both coronary calcium and CIMT after adjustment for the “metabolic syndrome” in whites but not blacks (p > 0.1).⁴ Hence, insulin resistance may be a more important risk factor contributing to vascular disease in whites than blacks.

An association between insulin resistance and stroke risk was demonstrated in the Atherosclerosis Risk in Communities (ARIC) study (using fasting insulin levels),^{5, 6} the Cardiovascular Health Study (CHS; using the Gutt insulin sensitivity index⁷),⁸ National Health and Nutrition Examination Survey (NHANES; using the HOMA-IR model),⁹ the Northern Manhattan Study (NOMAS; using HOMA-IR, but only a marginal effect with a threshold in the highest quartile),¹⁰ the Uppsala Longitudinal Study of Adult Men (using serum insulin, fasting pro-insulin, and insulin sensitivity by the euglycemic clamp),¹¹ and a study in the general Japanese population (using the HOMA-IR model).¹² In contrast, no statistically significant association was observed in the Rotterdam Study (using the HOMA-IR model)¹³ or the Bezafibrate Infarction Prevention Study (BIP) study of patients in stable coronary heart disease.¹⁴

Of these studies, only the ARIC and NOMAS reports explicitly discussed a potential differential association by race/ethnicity. In ARIC, higher insulin levels were associated with stroke risk among whites but not blacks (p_interaction = 0.036).⁶ In NOMAS there was not a differential association by race; however, the analysis diluted the opportunity for detection by the use of an undirected alternative hypothesis across three ethnic groups (i.e., only testing racial differences rather than specifically testing if the relationship is less in blacks compared to whites plus Hispanics) and was limited by a relatively small number of white and black participants (only approximately 317 of each).¹⁰ Only ARIC examined racial differences in the association of insulin resistance with heart disease, failing to find a difference in the magnitude of the association between whites and blacks.⁶

The currently active IRIS (Insulin Resistance Intervention after Stroke) Trial is assessing the potential benefit of insulin sensitization using pioglitazone, and has a stated secondary aim of assessing racial differences in treatment efficacy.¹⁵ Currently 11% of the IRIS patients are black (personal communication, Walter Kernan). The growing body of literature (including this report) of a weaker association of insulin resistance and stroke risk in the black population will make assessing effect modification by race important; however, the number of blacks in the trial imply that IRIS will have marginal statistical power to assess interaction.

Some risk factors have a differential association with cerebral infaction (CI) and intracerebral hemorrhage (ICH). For example, total cholesterol is positively associated with CI,^16-18 but negatively associated with ICH.^{16, 19, 20} While it is likely that the positive association of lipid levels with CI is based on atherosclerosis, the mechanism for the protective association for hemorrhagic stroke is not well understood. Most of the studies of the association of insulin resistance and stroke risk focused on CI outcomes, and to our knowledge only two studies have assessed the association of insulin resistance and ICH risk, both failing to find an association.^11,13

Herein, we assess the relationship of insulin resistance with CI and ICH risk in a longitudinal cohort study of black and white participants.

Methods

The overall goal of the REasons for Geographic And Racial Differences in Stroke (REGARDS) Study is to advance the understanding of racial and geographic differences in stroke mortality, including differences in the impact of risk factors. REGARDS recruited 30,239 community-dwelling black and white participants between 2003 and 2007, oversampling black participants (42%). A cardiovascular risk survey was completed by telephone and an in-home physical assessment (including fasting sample collection and electrocardiogram) conducted approximately 2 to 3 weeks later. Participants were followed at 6-month intervals by telephone, and medical records were retrieved for adjudication of outcome events. Details of the study methods²¹ are provided elsewhere. Of the eligible participants contacted, the cooperation rate was 49%.

Suspected strokes were identified by hospitalization for stroke or stroke-like symptoms solicited during telephone interviews conducted at 6-month intervals. Medical records were retrieved for suspected strokes, and stroke endpoints (including the determination of stroke subtype) were determined by physician review.²²

Fasting insulin was measured for all participants not self-reporting diabetes using a electrochemiluminescence immunoassay using the Roche Elecsys 2010 system (Roche Diagnostics, Indianapolis, IN). The primary exposure variable, insulin resistance, was assessed using the homeostasis model (HOMA-IR = (Insulin (mg/dL) × Glucose (mg/dL)) / 405).³

Risk factors included in the Framingham Stroke Risk Profile²³ were considered as potential confounders of the relationships of insulin resistance with CI and ICH. During the in-home assessment, two blood pressure measures were taken and average systolic blood pressure (SBP) was used in analyses. Use of antihypertensive medications was defined by self-report. Because of the challenges in assessing insulin resistance among diabetics, we excluded all participants who self-reported having diabetes or who were on treatment for diabetes. However, there were a small number of participants with a fasting glucose ≥126 mg/dL who reported being non-diabetic, and these participants were retained in the study and are referred to as “undiagnosed diabetic”. Smoking was defined by self-report. Atrial fibrillation was defined as self-reported physician diagnosis or ECG evidence. Left ventricular hypertrophy (LVH) was defined from the ECG.²⁴ History of heart disease was defined as self-reported myocardial infarction, electrocardiogram evidence of myocardial infarction, or self-reported coronary artery bypass, angioplasty or stent.

Proportional hazards analysis was used to assess associations between risk factors and incident CI and ICH events though April 1, 2012. A priori, main effects were assessed at α = 0.05 and interactions at α = 0.10. Models were fit to assess the relationship between HOMA-IR and stroke risk both after adjustment for demographic factors (age and sex) and after further adjustment for risk factors. The risk factor adjustment included all Framingham Stroke Risk Profile variables except diabetes. Because the HOMA-IR includes the glucose level used to define undiagnosed diabetics, a final separate adjustment was done for this factor. For ICH, there were 41 events (28 in whites + 13 in blacks), and subsequent to the adjustment for demographic factors (age, race and sex), because of the small number of events the risk factor adjustment included only SBP and use of antihypertensive medications (factors previously shown predictive of ICH events²⁵). Because some medical records could not be retrieved and other records remained in the adjudication process at the time of analysis (approximately 10% each), multiple imputation²⁶ techniques were employed in the analysis to reduce the potential bias arising from unconfirmed stroke events. Details of the multiple imputation approach employed are provided elsewhere.²⁷

Results

Of the 30,239 REGARDS participants, 56 (0.2%) participants had data anomalies requiring exclusion, 6,527 (22%) self-reported being diabetic, 402 (1%) did not have follow-up data, 2,873 (10%) were not fasting at the baseline visit, 986 (3%) reported stroke at baseline, and 247 (0.8%) did not have glucose data for calculation of the HOMA-IR model; collectively reducing the cohort to 19,148 participants. Among these, 12,366 were white with 71,683 per-years of exposure, during which 240 CI events and 28 ICH events occurred. The remaining 6,782 were black with 36,986 person-years of exposure, during which 124 CI events and 13 ICH events occurred.

Those with higher levels of HOMA-IR had higher blood pressure, were more likely to be on antihypertensive treatment and have undiagnosed diabetes, left ventricular hypertrophy, and history of heart disease (Table 1). A total of 808 participants reported no diabetes but had a fasting glucose greater than 126 mg/dL.

Table 1.

Characteristics of the study population shown by quartile of HOMA-IR level

		Insulin Resistance (HOMA-IR)
		Quartile 1 (< 1.32) (n = 4,786)	Quartile 2 (1.32 – 2.16) (n = 4,788)	Quartile 3 (2.16 – 3.65) (n = 4,787)	Quartile 4 (3.65+) (n = 4,787)
Demographic Factors	Age (mean ± SD)	64.8 ± 10.0	65.0 ± 9.6	64.3 ± 9.3	63.0 ± 8.7
	Male (%)	42.8	43.6	44.1	45.5
	Black (%)	26.8	31.1	38.4	45.4
Risk factors	Systolic Blood Pressure mmHg (mean ± SD)	122.6 ± 16.6	125.6 ± 16.1	127.6 ± 15.7	129.6 ± 15.9
	Antihypertensive medication (%)	29.7	39.7	49.2	58.1
	Diabetes (undiagnosed)	0.3%	0.5%	1.6%	14.5%
	Current Smoking	16.1	13.9	12.4	13.8
	Atrial Fibrillation (%)	7.2	7.7	7.4	8.0
	Left Ventricular Hypertrophy (%)	5.4	7.8	9.6	11.5
	History of Heart Disease	12.2	13.5	14.7	17.3

For whites, there was an increasing risk (HR_ln(IR) = 1.17; 95% CI: 1.00 – 1.68) of CI with increasing levels of insulin resistance (see Table 2), however, much of this association was attenuated by adjustment for risk factors (HR_ln(IR) = 1.09; 95% CI: 0.92 – 1.29) and subsequent adjustment for diabetes (HR_ln(IR) = 1.05; 95% CI: 0.88 – 1.26). Much of the association present in the demographic model appeared to be associated with higher risk in the 4^th quartile. In contrast, there was no evidence of an increasing stroke risk at higher levels of HOMA-IR for blacks in the demographic model (HR_ln(IR) = 1.01; 95% CI: 0.81 – 1.25), with adjustment for risk factors resulting in a non-significant protective association. These differences between races should be interpreted with caution as a formal test of interaction failed in both either the demographic model (p = 0.19), or after adjustment for risk factors or diabetes (p = 0.26 for both).

Table 2.

Hazard ratio (and 95% confidence intervals) for cerebral infarction (infarction) or intracerebral hemorrhage (ICH) by quartile of insulin resistance and as a function of the log of the insulin resistance value. Demographic models included adjustment for age and sex. For analysis of infarction, adjustment for risk factors included systolic blood pressure, use of antihypertensive medications, diabetes, cigarette smoking, atrial fibrillation, left ventricular hypertrophy and history of heart disease. Because of a smaller number of events for ICH, adjustment for risk factors included only systolic blood pressure and use of antihypertensive medications.

			White (n = 14,085)			Black (n = 7,984)
			Demo	Demo + RF	Demo + RF + Diabetes	Demo	Demo + RF	Demo + RF + Diabetes
Infarction ewhite = 240 eAA = 124	Insulin Quartile	Q1 (0.0 – 1.3 uU/mL) Q2 (1.3 – 2.2 uU/mL) Q3 (2.2 – 3.7 uU/mL Q4 (3.7 – 180.1 uU/mL)	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)	1.0 (ref)
			0.94 (0.68 – 1.30)	0.87 (0.62 – 1.22)	0.87 (0.62 – 1.21)	1.03 (0.63 – 1.71)	1.07 (0.64 – 1.79)	1.07 (0.64 – 1.79)
			0.94 (0.67 – 1.32)	0.86 (0.60 – 1.21)	0.85 (0.60 – 1.21)	1.03 (0.64 – 1.68)	0.93 (0.56 – 1.54)	0.93 (0.56 – 1.54)
			1.19 (0.84 – 1.68)	0.97 (0.68 – 1.39)	0.92 (0.63 – 1.34)	1.09 (0.66 – 1.78)	0.94 (0.56 – 1.58)	0.96 (0.56 – 1.63)
	1 unit Ln(IR)		1.17 (1.00 – 1.38)	1.09 (0.92 – 1.29)	1.05 (0.88 – 1.26)	1.01 (0.81 – 1.25)	0.92 (0.74 – 1.16)	0.93 (0.73 – 1.18)
ICH ewhite = 28 eAA = 13	Insulin Quartile	Q1 (0.0 – 1.3 uU/mL) Q2 (1.3 – 2.2 uU/mL)	1.0 (ref)	1.0 (ref)		1.0 (ref)	1.0 (ref)
			1.10 (0.46 – 2.61)	1.05 (0.44 – 2.49)		0.92 (0.13 – 6.80)	0.85 (0.11 – 6.29)
		Q3 (2.2 – 3.7 uU/mL) Q4 (3.7 – 180.1 uU/mL)	0.51 (0.15 – 1.74)	0.48 (0.14 – 1.61)		1.33 (0.24 – 7.34)	1.09 (0.11 – 6.29)
			0.58 (0.17 – 1.97)	0.52 (0.16 – 1.77)		1.19 (0.21 – 6.80)	0.88 (0.15 – 5.23)
	1 unit Ln(IR)		0.64 (0.37 – 1.11)	0.61 (0.35 – 1.04)		1.35 (0.69 – 2.63)	1.20 (0.60 – 2.39)

For ICH among whites, there was a trend for a decreased risk for ICH at higher levels of HOMA-IR (HR_ln(IR) = 0.64; 95% CI: 0.37 – 1.11), with little effect for after adjustment for SBP and use of antihypertensive medications (HR_ln(IR) = 0.61; 95% CI: 0.35 – 1.04). In contrast, for blacks there was no association, whether adjusted only for demographics (HR_ln(IR) = 1.35; 95% CI: 0.69 – 2.63) or additionally adjusted for SBP and use of antihypertensive medication (HR_ln(IR) = 1.20; 95% CI: 0.60 – 2.39). While the decrease in ICH risk at higher levels of insulin resistance for whites and the increased ICH risk with insulin resistance in blacks were both non-significant, that they were in different directions contributed to evidence of interaction in the risk factor model (p = 0.07), and borderline evidence in the demographic model (p = 0.11).

The sensitivity analysis excluding undiagnosed diabetics from the analysis is provided in supplemental table I, and showed no substantial difference in interpretation.

Discussion

These data support a potentially larger impact of insulin resistance on the risk of CI in the white population than the black population; however, this finding should be interpreted with caution. First, much of the association present in the demographic model was attenuated by adjustment for cerebrovascular risk factors. This is to be expected as it is well-known that individuals with high levels of insulin resistance are more likely to have the “metabolic syndrome,”²⁸ and many of these factors are likely in the causal pathway of the action of insulin resistance. That is, many of components of the metabolic syndrome (i.e., obesity, insulin resistance or hyperglycemia, hypertension, dyslipidemia) have a common etiology ^{29, 30}, which primarily involves abnormal energy balance and inflammation. There are associations between elements of the metabolic syndrome, however, which are complex and important.³¹ Insulin resistance, for example, is causally linked to abnormal lipid metabolism and hyperglycemia. In addition, it has been suggested that insulin resistance is causally related to hypertension,²⁹ and this is the risk factor with the largest population attributable risk for stroke.^{32, 33} If this is the case, since so many of the stroke risk factors are part of the metabolic syndrome, the attenuation of the association between insulin resistance and stroke risk with the adjustment for these risk factors underscores the truth of the observed association between insulin resistance and stroke risk.

Second, while the association between HOMA-IR and CI risk was significant for whites but not blacks, the formal assessment of whether the association was different for whites and blacks (i.e., the interaction test between HOMA-IR and race) failed to reach a level of statistical significance for cerebral infarction (although it was marginally significant for hemorrhages in the demographic model, and significant in the risk factor adjusted model). Hence, while we saw precisely what we hypothesized (a significant association of insulin resistance with stroke risk in whites, and no association in blacks) there is not clear evidence that the significant association in whites and the non-significant association in blacks differ in their magnitude.

Despite a relatively small number of ICH events, there was an inverse association of insulin resistance with risk of ICH events in whites, suggesting the impact of insulin resistance on ICH risk differed in blacks and whites (p = 0.07 in the risk factor model and p = 0.11 in the demographic model). To our knowledge, only the Rotterdam study¹³ and the Uppsala Longitudinal Study of Adult Men¹¹ have previously examined associations between insulin resistance and risk of ICH, finding virtually no evidence of an association (HR = 1.03; 95% CI: 0.76 – 1.39 for the Rotterdam study, HR = 0.95; 95% CI: 0.60 – 1.50 for Uppsala). The relationship of insulin resistance and stroke risk is similar to patterns observed between elevated lipids and stroke risk, where in other studies showed increased risk of CI, but lower risk of ICH with elevated lipids.^16-20 Like the observed protective association for lipids on ICH risk, we are also uncertain regarding the potential protective pathway of action for insulin resistance on ICH risk. Also, like our observation of an association between insulin resistance and CI in white participants but not black participants, we do not have a proposed mechanism through which insulin resistance is protective of ICH in white participants but not black participants.

An additional challenge to understanding the association between insulin resistance and stroke risk is the quantification of insulin resistance. While there are several approaches to measuring insulin resistance, the invasive and time-consuming euglycemic clamp³⁴ is the generally-recognized “gold standard,” with the marginally less invasive and time consuming Frequently Sampled Insulin/Glucose Test (FSIGT)¹ and the Insulin Suppression Test (IST)²⁹ being considered the best “second line” measures. However, all of these approaches require infusion of glucose and/or insulin, and require several hours for assessment; hence, these approaches may be impractical clinically and for epidemiological studies. Other diagnostic strategies involve measurement of insulin and glucose during an oral glucose tolerance test⁷, which has seen limited use in epidemiologic studies due to the two-hour time requirement. We measured insulin resistance with the HOMA-IR, which is based on fasting glucose and fasting insulin. This method is commonly used in clinical trials and epidemiological research because it is safer and less invasive, and provides a reliable index of these more complex measures.³⁵ Unlike more complex tests, HOMA-IR measures primarily the hepatic component of insulin resistance and not the peripheral component. Persons with peripheral insulin resistance may not be identified with the HOMA, which is important as peripheral insulin resistance may correlate more closely with adverse metabolic consequences of insulin resistance such as inflammation, hypertension, and dyslipidemia. There are other alternative simple measures of insulin resistance (including fasting insulin level),³⁵ and part of the inconsistent association of insulin resistance and stroke risk is potentially attributable to limitations of fasting measures characterizing peripheral insulin resistance.

The strengths of this study include the large sample size of the cohort (particularly a large number of African Americans) and a reasonable number of CI events (n = 364). In addition, physician adjudication of medical records gives confidence in the stroke diagnosis and stroke subtype distinction between CI and ICH. Perhaps the greatest weakness is a relatively small number of ICH events (n = 41); however, this number was sufficient to provide precision to detect a significant racial difference in the relationship between insulin resistance and ICH risk.

In conclusion, while there was an association between insulin resistance and increased risk for CI in white participants, but not in black participants, although the race-by-HOMA-IR interaction was non-significant. While the association observed in white participants was attenuated by adjustment for cerebrovascular risk factors, it could be argued that many of these risk factors are in the pathway of action for insulin resistance (specifically hypertension and diabetes), and as such the model may be over adjusting for risk factors. We also provide the first evidence of a racial difference in the relationship between insulin resistance and risk of ICH. As such, these data suggest that insulin resistance, as measured by the HOMA, may be playing a larger role in white than black populations. The reasons for these racial differences, and the pathway for a protective effect of insulin resistance on ICH, certainly require additional research.