Associations Between Carotid Artery Plaque Burden, Plaque Characteristics, and Cardiovascular Events: The ARIC Carotid Magnetic Resonance Imaging Study

Key Points

Question  In an asymptomatic community-based cohort, are carotid artery plaque characteristics on magnetic resonance imaging associated with cardiovascular events when adjusted for plaque burden?

Findings  In this cohort study of 1256 participants from the Atherosclerosis Risk in Communities Carotid Magnetic Resonance Imaging substudy, magnetic resonance imaging measures of carotid artery plaque burden and characteristics were analyzed. The presence of a lipid core was independently associated with incident cardiovascular disease events when adjusted for traditional cardiovascular disease risk factors and carotid artery wall thickness.

Meaning  In this study, the presence of a lipid core was associated with incident cardiovascular disease events independent of carotid artery wall thickness in asymptomatic individuals and improved risk prediction of incident cardiovascular disease events over traditional cardiovascular risk factors.

Importance  It remains unknown whether in an asymptomatic community-based cohort magnetic resonance imaging (MRI) measures of plaque characteristics are independently associated with incident cardiovascular disease (CVD) events when adjusted for carotid artery (CA) wall thickness, a measure of plaque burden.

Objective  To assess associations of CA MRI plaque characteristics with incident CVD events.

Design, Setting, and Participants  The Atherosclerosis Risk in Communities (ARIC) study is a prospective epidemiologic study of the incidence of CVD in 15 792 adults of which 2066 women and men were enrolled in the ARIC Carotid MRI substudy. ARIC participants were enrolled from 1987 to 1989, and the substudy was conducted between January 2004 and December 2005. Analysis began January 2017 and ended August 2020.

Exposures  Incident CVD events during a median (interquartile range [IQR]) follow-up time of 10.5 (8.1-10.9) years were assessed.

Main Outcomes and Measures  Proportional hazards Cox analyses were performed to ascertain associations between MRI variables of CA plaque burden and plaque characteristics.

Results  Of 15 792 ARIC participants, 2066 were enrolled in the substudy, of whom 1256 (701 women [55.8%]) had complete data and were eligible for incident CVD analyses. Carotid artery plaques in participants with incident CVD events (172 [13.7%]) compared with those without (1084 [86.3%]) had a higher normalized wall index (median [IQR], 0.48 [0.36-0.62] vs 0.43 [0.34-0.55]; P = .001), maximum CA wall thickness (median [IQR], 2.22 [1.37-3.52] mm vs 1.96 [1.29-2.85] mm; P = .01), maximum CA stenosis (median [IQR], 5% [0%-22%] vs 0% [0%-13%]; P < .001), and when present, a larger lipid core volume (median [IQR], 0.05 [0.02-0.11] mL vs 0.03 [0.01-0.07] mL; P = .03), respectively. The presence of a lipid core was independently associated with incident CVD events when adjusted for traditional CVD risk factors and maximum CA wall thickness (hazard ratio, 2.48 [95% CI, 1.36-4.51]; P = .003), whereas the presence of calcification was not. The frequency of intraplaque hemorrhage presence in this population of individuals free of CVD at baseline who were not recruited for carotid stenosis was too small to draw any meaningful conclusions (intraplaque hemorrhage presence: 68 of 1256 participants [5.4%]). Carotid artery lumen area and maximum stenosis, which were overall low, were independently associated with incident CVD events when adjusted for traditional CVD risk factors, as anticipated.

Conclusions and Relevance  The presence of a CA lipid core on MRI is associated with incident CVD events independent of maximum CA wall thickness in asymptomatic participants.

High-risk atherosclerotic lesions have been associated with plaque characteristics including the presence of a lipid-rich necrotic core, intraplaque hemorrhage (IPH), thin-fibrous cap, and inflammation.¹ Magnetic resonance imaging (MRI)–based carotid plaque composition and the remodeling index have been associated with cardiovascular events in participants of the epidemiological Multi-Ethnic Study of Atherosclerosis (MESA).² Magnetic resonance imaging of high-risk plaque characteristics also identified cerebrovascular events prospectively in asymptomatic individuals with carotid stenosis.³ Despite recent progress, it remains unclear whether measures of MRI plaque characteristics are associated with cardiovascular events independent of measures of plaque burden at the population level. Here, we analyze associations between plaque characteristics and plaque burden with cardiovascular disease (CVD) events in participants of the carotid MRI substudy of the Atherosclerosis Risk in Communities (ARIC) Study.⁴

Briefly, ARIC is a prospective epidemiologic study of the incidence of CVD of 15 792 adults, aged 45 to 64 years who were enrolled from 1987 to 1989. A total of 2066 women and men were enrolled between January 2004 and December 2005 in the ARIC Carotid MRI substudy.⁴ Participants had a carotid artery intima media thickness (C-IMT) of approximately the 68th percentile or higher during visit 3 in 1993 to 1995 or visit 4 in 1996 to 1998.⁴ Because of incomplete MRI examinations or measurements, 128 participants were excluded, and another 169 were excluded because of protocol deviations or poor image quality, resulting in 1769 individuals. Further, 279 had prevalent cardiovascular events at the MRI baseline visit and were therefore excluded from the incident analyses. An additional 234 individuals were removed because of missing any baseline characteristics or MRI variables, resulting in 1256 data sets included in the incident CVD analyses. Race/ethnicity was self-reported. The ARIC Carotid MRI substudy was approved by the local institutional review boards, and all participants provided written informed consent.

All participants were imaged on 1.5-T MRI systems (Excite platform, GE, Medical Systems or Symphony Maestro, Siemens Medical Solutions), with a 4-element phased-array carotid coil (Machnet) and standard MRI protocols across all field centers. After initial localizers, a 3-dimensional time-of-flight magnetic resonance angiography was acquired for both carotids including the extracranial carotid bifurcations (resolution: 0.59 × 0.59 × 2 mm³). High-resolution black-blood MRI (BBMRI) was acquired with a 2-dimensional electrocardiographically gated double inversion recovery fast-spin echo pulse sequence with fat suppression and an inversion time set to null the blood signal [16 slices]). BBMRI was done before and 5 minutes after administration of gadodiamide (Omniscan; GE Healthcare) at 0.1 mmol/kg. BBMRI postcontrast scans were the same as precontrast except for a shorter inversion time of approximately 200 milliseconds to account for the paramagnetic property of the gadolinium-based contrast agent.

Images were analyzed using a semiautomated software (VesselMASS; Leiden University Medical Center).^5,6 Interreader and intrareader variability and scan and rescan variations have been reported previously.^4,7 Reliability, as determined with K statistics, for the presence of a lipid core, calcification, and hemorrhage were fair to good (0.4 ≤ κ ≤ 0.75).^4,7 Briefly, readers delineated the outer wall and the lumen contour of the carotid, the lipid core, and calcification on postcontrast BBMRIs.^8,9 Calcification was identified as hypointense signal on all contrast weightings, whereas the lipid core was identified as isointense on 3-dimensional time-of-flight magnetic resonance angiography and depicted by signal enhancement of the surrounding fibrous tissue. The fibrous cap contour was automatically delineated between lumen and lipid core. Intraplaque hemorrhage was characterized as a hyperintense signal on precontrast 3-dimensional time-of-flight magnetic resonance angiography and BBMRI scans. Stenosis was calculated on the maximum intensity projection of the 3-dimensional time-of-flight magnetic resonance angiography scans using established criteria from the North American Symptomatic Carotid Endarterectomy Trial.¹⁰

Measures of MRI carotid artery (CA) plaque burden include total carotid wall volume, maximum carotid wall thickness (maximum segmental wall thickness of 12 segments measured at the slice with the largest lipid core area or the maximum segmental wall thickness if no core was present^4,7), lumen and wall area, normalized wall index (wall area / [lumen area + wall area]), and maximum stenosis. Measures of categorical CA MRI plaque characteristics include presence of a lipid core (present on at least any 1 slice), a lipid core in 2 or more adjacent slices (present on at least 2 adjacent slices), calcification, and IPH. Measures of continuous CA MRI characteristics were limited to those with a lipid core and include total lipid core volume, maximum lipid core area, mean lipid core area, total lipid core percentage of total wall volume, mean fibrous cap thickness, mean fibrous cap thickness at the maximum lipid core, minimum fibrous cap thickness, and mean minimum fibrous cap thickness. Complex CA plaque composition is categorized for 4 plaque phenotypes as presence of (1) lipid core and calcification, (2) lipid core and IPH, (3) calcification and IPH, and (4) lipid core and calcification and IPH. Not every study participant presented with a measurable lipid core, fibrous cap, or calcification resulting in a lower number of individuals included in the analyses of continuous CA MRI plaque characteristics compared with carotid plaque burden variables and categorical carotid plaque characteristics. Continuous MRI CA plaque characteristics were analyzed only for those with a lipid core and, therefore, results for continuous plaque characteristics are reported in separate Tables.

Incident CVD events were recorded through December 2016 and defined as the composite of incident coronary heart disease (CHD) events or incident ischemic strokes. Incident CHD events included definite or probable myocardial infarction, CHD death, and the occurrence of a coronary artery bypass surgery or angioplasty during follow-up. Ischemic stroke events included validated definite or probable hospitalization for thrombotic or embolic stroke using previously defined criteria for ARIC.¹¹

Participants free of CVD at baseline of the ARIC Carotid MRI substudy were included in the incident CVD analyses. Specifically, 172 of 1256 individuals (13.7%) had incident cardiovascular events. In addition, among the excluded 279 individuals who had prevalent cardiovascular events at the MRI baseline visit, an additional 263 CVD events occurred, which were excluded from the present analyses of incident CVD events.

The analyses were weighted by the inverse of the sampling fractions in the 8 sampling strata.⁷ Values are reported as mean (SD) for normally distributed variables, medians (interquartile range) for nonnormally distributed variables, or as number (percentage). We compared baseline and MRI variables between participants with vs without incident CVD events. Continuous variables were analyzed with an independent-sample t test, and a 2-sample Wilcoxon rank sum test was used for nonnormal variables. Group differences for categorical variables were analyzed with the χ² or Fisher exact test. Univariable and multivariable Cox proportional hazards analyses were performed for incident CVD events with continuous and dichotomous CA MRI variables. Multivariable proportional hazards analyses were conducted with 2 separate adjustment models and hazard ratios (HRs) were reported per 1-unit natural log (ln)–transformed increase for continuous MRI variables. The first model was adjusted for clinical factors and included age, sex, race/ethnicity, ARIC study field center, smoking status, body mass index, blood glucose level, diabetic status, systolic blood pressure, diastolic blood pressure, high-density lipoprotein cholesterol, non–high-density lipoprotein cholesterol, use of blood pressure–lowering medication, cholesterol-lowering medication use, aspirin use, diabetes medication use, and high-sensitivity C-reactive protein. In another model, we additionally adjusted for CA wall thickness. Nonnormally distributed variables were ln-transformed prior to performing proportional hazards analyses.¹² Interactions among variables were tested and in the final model, only the interaction between race/ethnicity and field center was included because of a P value less than .15. Individuals with missing values were excluded from the analysis. Participants lost to follow-up were considered as censored without events. To evaluate improvement in risk prediction by including MRI plaque variables in CVD risk prediction, area under the receiver operating characteristic curve (AUROC) was calculated for 10-year follow-up and censoring, with adjustment for optimism.¹³ The basic model was adjusted for age, sex, total cholesterol, high-density lipoprotein cholesterol, systolic blood pressure, antihypertensive medication use, diabetes, and current or former smoking; the extended models added MRI plaque characteristics. Model calibration was assessed using the Gronnesby-Borgan method.¹⁴ Delta AUROC, categorical and continuous net reclassification improvement, and integrated discrimination index were calculated by comparing the basic and extended models for 10-year risk prediction of incident CVD events.¹⁵ Variable combinations were identified that maximized the value of the AUROC for plaque burden (model 1) and plaque characteristics (model 2) and that resulted in the largest increment of the AUROC when added to traditional CVD risk factors (models 6-8). All other combinations are not listed. Cox proportional hazards models were also performed separately for incident CHD events and incident ischemic stroke, respectively (eTables 3-6 in the Supplement). Analyses were conducted with SAS statistical software version 9.4 (SAS Institute Inc) and Stata statistical software version 13 (StataCorp LP). Two-sided P values were significant at less than .05. Analysis began January 2017 and ended August 2020.

Of 1256 participants included in the incident CVD analyses, 701 (55.8%) were women. The mean (SD) follow-up time for incident CVD was 9.1 (2.98) years, median (interquartile range) follow-up time was 10.5 (8.3-11.0) years, and the cumulative event rate for this period was 13.7% (172 of 1256). In total, there were 183 events in 172 participants among 1256 participants free of CVD at baseline. Adjudicated incident CVD included 125 CHD events (10.0%) and 58 ischemic strokes (4.6%). Incident CHD events included 72 myocardial infarctions (5.7%) and the ratio of incident myocardial infarction to incident ischemic stroke was 1.24. Incident CHD revascularization among all 1256 individuals included 80 events (6.4%) (Figure 1).

Participants who developed incident CVD events were typically older, more often male, and were taking cholesterol-lowering therapy compared with individuals without incident cardiovascular events (Table 1).

Measures of CA plaque burden, including total wall volume, wall area, normalized wall index, and maximum wall thickness, were larger and carotid lumen area were smaller in participants with incident CVD events compared with individuals without (Figure 2). Overall vessel narrowing was low, with a median (interquartile range) maximum stenosis of 5% (0%-22%) in the incident CVD events group (Table 2).

A lipid core was present in 403 individuals (32.0%) and calcification was detected in 376 participants (30.0%), whereas IPH occurred in 68 (5.4%). Individuals with CVD events had a higher percentage of lipid core presence and calcification (Table 2). The most common multicomponent plaque phenotype was the presence of both a lipid core and calcification in 37 participants (21.5%) with incident CVD compared with 146 individuals (13.5%) without events.

The continuous CA plaque characteristics analyses included those with a lipid core (n = 403) for whom also all other continuous MRI variables were available, resulting in the inclusion of 400 participants. Among those with a lipid core, those with incident CVD events had a higher total lipid core volume and larger mean and maximum lipid core areas than those without events (eTable 1 in the Supplement).

Univariate proportional hazard Cox analyses showed that the presence of a lipid core and the presence of a lipid core in 2 or more adjacent slices were significantly associated with incident CVD events (HR, 2.39; 95% CI, 1.61-3.54; P < .001 vs HR, 2.43; 95% CI, 1.63-3.64; P < .001; Table 3). As anticipated, measures of plaque burden were independently associated with incident CVD events using univariate proportional hazard models. When adjusting for clinical factors and maximum CA wall thickness, only the presence of a lipid core and the presence of a lipid core in 2 or more adjacent slices remained significantly associated with incident CVD events (Table 3). The presence of calcification but not IPH was significantly associated with incident CVD events in the unadjusted model. Intraplaque hemorrhage was present in 68 of 1256 participants (5.4%), which was too small to draw any meaningful conclusions in this population of individuals free of CVD at baseline. Continuous carotid plaque characteristics showed no associations with incident CVD events (eTable 2 in the Supplement). Maximum calcification area was not associated with incident CVD (available for 390 of 400 participants [97.5%]; data not shown).

Briefly, when adjusted for clinical factors and carotid wall thickness, the presence of a lipid core remained a significant predictor of both incident CHD and ischemic stroke events (eTables 3, 4, 5, and 6 in the Supplement). The presence of calcification was significantly associated with incident CHD but not with incident ischemic stroke in the unadjusted model and when adjusted for clinical factors. Conversely, the presence of IPH was significantly associated with incident ischemic stroke but not with incident CHD in the unadjusted model. However, the sample of IPH was too small to draw any meaningful conclusions.

Among those with presence of CA plaque characteristics (total lipid core volume, minimum fibrous cap thickness, calcification present) and CA plaque burden (total wall volume, normalized wall index, maximum wall thickness), the combination with traditional CVD risk factors significantly increased the AUROC compared with the model with risk factors alone (AUROC, 0.691; 95% CI, 0.630-0.753 vs AUROC, 0.625; 95% CI, 0.556-0.695; P = .02; n = 400; eTable 7 in the Supplement). Adding the presence of the lipid core to traditional cardiovascular risk factors improves the AUROC, continuous net reclassification improvement, and integrated discrimination index, respectively (n = 1256; eTable 8 in the Supplement). eTable 9 in the Supplement shows details on the conversion between untransformed and ln-transformed values for the MRI variables.

The results of the ARIC Carotid MRI substudy add to the growing body of evidence that noninvasive MRI CA measures of plaque burden and plaque characteristics aid in the prediction of incident CVD events. The results of the ARIC Carotid MRI substudy on MRI plaque characteristics are the largest to date in terms of number of participants, follow-up duration, and number of CVD events. The primary finding of this study is that the presence of a lipid core is independently associated with incident CVD events after adjusting for traditional cardiovascular risk factors and CA wall thickness, indicating that the lipid core, an MRI plaque characteristic, is valuable for CVD risk assessment independent of measures of plaque burden.

Atherosclerotic plaque characteristics have been associated with plaque rupture risk.¹⁶ Associations between MRI carotid wall thickness and carotid plaque burden with cardiovascular risk factors have been reported previously for the ARIC Carotid MRI substudy.^7,17,18 A previous analysis of the ARIC Carotid MRI substudy showed that traditional CVD risk factors predict carotid wall thickness and volume 18 years later.⁷ MRI-based carotid plaque characteristics and the remodeling index have been associated previously with CVD events in 946 participants in MESA.² In MESA, C-IMT was assessed by ultrasonography and individuals were followed up for an average of 5.5 years. Plaque characteristics including lipid core, internal CA calcification, MRI carotid remodeling index, and ultrasonography C-IMT were significant predictors of new CVD events in individuals without a history of CVD. The combination of C-IMT and the presence of a lipid core on MRI significantly improved the identification of CVD events compared with traditional risk factors. In our study, all measures of plaque characteristics were obtained by MRI, and we also identified the presence of a lipid core, maximum stenosis as predictors of CVD events.

A recent study from MESA showed that carotid wall thickness, as measured with MRI, was more consistently associated with incident CVD, particularly stroke, than was C-IMT measured with ultrasonography.^19,20 In this study, we explored the utility of a lipid core, which can be measured reliably with MRI, in addition to measures of plaque burden, to analyze associations with CVD events in participants of the ARIC Carotid MRI substudy.

Carotid artery MRI plaque characteristics have been studied prospectively in 154 asymptomatic individuals with moderate carotid stenosis.³ Cerebrovascular events were associated with lipid core size, maximum wall thickness, thin/ruptured plaque, and IPH size. In our study, we identified the presence of a lipid core and maximum stenosis as predictors of CVD events.

The association of plaque characteristics and CVD events has been studied with carotid ultrasonography in 581 patients with type 2 diabetes.²¹ A total of 581 patients with diabetes were followed up for 9 years, and plaques were categorized as echolucent (lipid rich), heterogenous, and echogenic (calcific). The results showed that an echogenic (calcific) plaque type is predictive of CVD events after adjusting for C-IMT.

Previous reports have shown that CA MRI wall thickness is associated with improved discrimination of CVD events when added to traditional risk factor models.²² In this study, the combination of MRI measures of carotid plaque burden and plaque characteristics including among others a lipid core volume, when present, resulted in a significant improvement in discriminating individuals with incident CVD events when added to traditional risk factors. The presence of a lipid core is associated with future incident CVD events independent of plaque burden. This is important because, as we showed earlier, some plaques characteristics are difficult to measure or detect in small plaques because of limited resolution of current MRI systems.⁴ Magnetic resonance plaque characteristics could be useful surrogates when conducting outcomes trials for novel CVD therapies or to enrich clinical trials with participants who are most likely to have incident CVD events, likely reducing the number of participants needed.

Because of the epidemiologic nature of the ARIC study, only a limited number of participants have developed advanced atherosclerotic CA lesions with IPH, a marker of high-risk plaques.^4,23-25 Our sample of participants with IPH present was too small to draw any meaningful conclusions. Schindler et al²⁶ had a high enough prevalence of IPH in an asymptomatic population to report on its association with CVD events; however, their sample was selected based on having carotid stenosis 50% or higher unlike our study (ie, approximately 4% of participants in our community-based cohort had carotid stenosis ≥50%).⁴ Imaging and measures of plaque characteristics were highly reproducible in the ARIC cohort, except for fibrous cap thickness and small structures, which were measured much less reliably.⁴ It has to be noted that those with prevalent CVD events at the MRI visit have been excluded from this study. We did not match the side of stroke with the side of carotid studied. However, these MRI features are considered systemic risk features, especially in the light of the concept of vulnerable patient rather than vulnerable plaque.

The presence of a lipid core is associated with incident CVD events independent of CA wall thickness, a measure of plaque burden, in asymptomatic ARIC participants. Noninvasive MRI carotid plaque characteristics, when present, improved risk prediction of incident CVD events and may be useful imaging surrogates to identify those at higher risk of future CVD events.

Corresponding Author: Bruce A. Wasserman, MD, Johns Hopkins School of Medicine, 600 N Wolfe St, 367 E Park Bldg, Baltimore, MD 21287 (bwasser@jhmi.edu).

Accepted for Publication: September 17, 2020.

Published Online: November 18, 2020. doi:10.1001/jamacardio.2020.5573

Author Contributions: Dr Wasserman had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Brunner and Virani contributed equally.

Concept and design: Virani, Nambi, Coresh, Boerwinkle, Wasserman.

Acquisition, analysis, or interpretation of data: Brunner, Virani, Sun, Liu, Dodge, Nambi, Coresh, Mosley, Sharrett, Ballantyne, Wasserman.

Drafting of the manuscript: Brunner, Sun, Dodge.

Critical revision of the manuscript for important intellectual content: Brunner, Virani, Liu, Nambi, Coresh, Mosley, Sharrett, Boerwinkle, Ballantyne, Wasserman.

Statistical analysis: Brunner, Sun, Liu, Dodge.

Obtained funding: Coresh, Boerwinkle, Wasserman.

Administrative, technical, or material support: Coresh, Boerwinkle, Wasserman.

Supervision: Brunner, Coresh, Mosley, Wasserman.

Conflict of Interest Disclosures: Dr Brunner reports grants from the National Institutes of Health (NIH) and the American Heart Association during the conduct of the study. Dr Nambi reports grants from NIH during the conduct of the study; other support from Merck, Amgen, Siemens, and Roche; grants from Veterans Health Administration; and personal fees from DynaMed outside the submitted work. Dr Virani reports grants from US Department of Veterans Affairs, World Heart Federation, and Tahir and Jooma Family and other support from American College of Cardiology outside the submitted work. Dr Coresh reports grants from NIH during the conduct of the study. Dr Mosley reports grants from NIH during the conduct of the study. Dr Ballantyne reports grants from NIH during the conduct of the study; grants and personal fees from Abbott Diagnostics, Amgen, Esperion, Novartis, Regeneron, Roche Diagnostic, and Akcea; grants from NIH, American Heart Association, and American Diabetes Association; personal fees from AstraZeneca, Amarin, Matinas BioPharma, Pfizer, Sanofi-Synthelabo, Althera, Novo Nordisk, Denka Seiken, Gilead Sciences, Janssen, Corvidia, Arrowhead, and New Amsterdam, outside the submitted work. Dr Liu reports grants from NIH during the conduct of the study. Dr Boerwinkle reports grants from NIH during the conduct of the study. Dr Wasserman reports grants from NIH during the conduct of the study.

Funding/Support: The Atherosclerosis Risk in Communities (ARIC) study has been funded in whole or in part with federal funds from the National Heart, Lung, and Blood Institute, NIH, US Department of Health and Human Services (grants HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700005I, and HHSN268201700004I) with the ARIC Carotid Magnetic Resonance Imaging substudy examination funded by grant U01HL075572-04.

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank the staff and participants of the ARIC study for their important contributions.