Coronary Artery Calcification Is an Independent Stroke Predictor in the General Population
Background and Purpose—Coronary artery calcification (CAC) is a noninvasive marker of plaque load that predicts myocardial infarcts in the general population. Herein, we investigated whether CAC predicts stroke events in addition to established risk factors that are part of the Framingham risk score.
Methods—A total of 4180 subjects from the population-based Heinz Nixdorf Recall study (45–75 years of age; 47.1% men) without previous stroke, coronary heart disease, or myocardial infarction were evaluated for stroke events over 94.9±19.4 months. Cox proportional hazards regressions were used to examine CAC as stroke predictor in addition to established vascular risk factors (age, sex, systolic blood pressure, low-density lipoprotein, high-density lipoprotein, diabetes mellitus, smoking, and atrial fibrillation).
Results—Ninety-two incident strokes occurred (82 ischemic, 10 hemorrhagic). Subjects suffering a stroke had significantly higher CAC values at baseline than the remaining subjects (median, 104.8[Q1;Q3, 14.0;482.2] vs 11.2[0;106.2]; P<0.001). In a multivariable Cox regression, log10(CAC+1) was an independent stroke predictor (hazards ratio, 1.52 [95% confidence interval, 1.19–1.92]; P=0.001) in addition to age (1.35 per 5 years [1.15–1.59]; P<0.001), systolic blood pressure (1.25 per 10 mm Hg [1.14–1.37]; P<0.001), and smoking (1.75 [1.07–2.87]; P=0.025). CAC predicted stroke in men and women, particularly in subjects <65 years of age and independent of atrial fibrillation. CAC discriminated stroke risk specifically in participants belonging to the low (<10%) and intermediate (10%–20%) Framingham risk score categories.
Conclusions—CAC is an independent stroke predictor in addition to classical risk factors in subjects at low or intermediate vascular risk.
- electron beam-computed tomography
- plaque load
- risk stratification
- stroke prediction
- subclinical atherosclerosis
Vascular risk factors predict the 10-year incidence of stroke in a relatively reliable way.1 Unfortunately, risk factors do not offer any information about the severity of actual vascular disease. For this reason, their prognostic value in individual subjects is limited. In the past, subclinical atherosclerosis markers, such as the carotid intima-media thickness2–4 or ankle-brachial index,5,6 have been used for further risk stratification. When used in addition to classical risk factors, these markers enhanced the prediction of future vascular events.
With respect to cardiovascular diseases, coronary artery calcification (CAC) determined by electron beam-computed tomography (EBCT) was recently identified as potent predictor of incident myocardial infarcts.7–11 CAC provides a measure of existing plaque load, which allowed to enhance the discrimination of future cardiovascular events, particularly in subjects belonging to an intermediate-risk cohort, in which CAC identified subjects at high vascular risk.9 Representing a measure of global atherosclerosis, the question arose previously whether CAC also predicts incident stroke.
Supported by observations from cohort studies, in which an association between CAC and stroke was found in cross-sectional analyses,12,13 the effect of CAC on future stroke events was previously evaluated in 3 population-based studies: the Cardiovascular Health Study (CHS),7 Rotterdam study,11 and Multi-Ethnic Study of Atherosclerosis (MESA).8 Notably, in none of these studies, CAC was found to represent a stroke predictor, when adjustments for classical risk factors were made. The CHS observed a tendency toward an elevated stroke hazard in subjects with CAC scores belonging to the fourth quartile.7 Yet, because of the small number of subjects enrolled into EBCT measurements (559 subjects), resulting in a small number of incident strokes (n=25) over an observation period of 5 years, these data failed to reach significance.
To elucidate the possible link between CAC and stroke, we herein evaluated the effect of CAC on stroke incidence in the Heinz Nixdorf Recall (HNR) cohort. The HNR is a population-based cohort of 4814 subjects 45 to 75 years of age that examines the role of risk factors and subclinical atherosclerosis in the development of overt vascular diseases.9 On the basis of its age profile and focus on vascular pathologies, as well as state-of-the-art population enrollment, assessment, and end point evaluation enabling exceptional follow-up rates of as much as 90% of subjects at 5 years, the HNR study is well suited for analyzing stroke predictors.4
The HNR cohort is a random sample of men and women 45 to 75 years of age, who were prospectively enrolled via mandatory citizen registries in Essen, Bochum, and Mülheim/Ruhr, 3 cities of the industrialized Ruhr area, between December 2000 and August 2003. All subjects gave informed consent. Exclusion criteria were inability or unwillingness to give informed consent, conditions (medical or other) precluding follow-up over 5 years, severe psychiatric disorders or illegal substance abuse, and pregnancy in women. Computer-assisted interviews and questionnaires were used to document medical histories. The study was approved by the ethical committee at the University Duisburg-Essen. Of 4814 subjects included, 4356 subjects had a negative history for previous stroke, coronary heart disease, and myocardial infarcts. Of these, CAC measurements were obtained in 4180 subjects. Subjects were followed up over 94.9±19.4 months. During that time, stroke events (both ischemic and hemorrhagic), defined as focal neurological deficits of presumed cerebrovascular origin that persisted over a period of ≥24 hours, were assessed in annual questionnaires and a follow-up visit after 5 years. Stroke events were validated by an independent internal and external end point committee that provided consensus decisions in case of disagreements (formed by K. Berger, Münster; M. Dichgans, Munich; C. Weimar, Essen). Stroke events were allocated to the date of stroke diagnosis, and nonstroke cases were censored at the date of last contact when the person was still alive or date of death. Trial of Org 10172 in Acute Stroke Treatment (TOAST) classifications were also performed.14
Classical Risk Factors
Systemic blood pressure was measured with an automated oscillometric blood pressure device (Omron 705-CP, Omron, Mannheim, Germany), taking the mean of the second and third of 3 measurements with a minimum of 3 minutes between the measurements. In some cases, automated blood pressure values were missing. Then values from a random-zero blood pressure device measurement (Mark II, HawksleyTechn, Lancing, United Kingdom) were used. Hypertension was classified according to Joint National Committee-7 thresholds.15 Participants were considered diabetic in cases of physician-diagnosed diabetes mellitus, having a blood glucose of >200 mg/dL or fasting glucose of >126 mg/dL or taking antidiabetic medication. For evaluating consequences of nicotine abuse, only current smoking was considered. Total, low-density lipoprotein (LDL), and high-density lipoprotein cholesterol and triglycerides were measured with standardized enzymatic methods. Antihypertensive, lipid-lowering, antidiabetic, and antiplatelet medications were noted. With the data obtained, the Framingham risk score (FRS) was determined.16 Standardized height and weight measurements were used for calculating the body mass index. Atrial fibrillation was classified in an ECG that was routinely recorded in all subjects on occasion of the baseline examination (prevalent atrial fibrillation at baseline) and the follow-up visit after 5 years (incident atrial fibrillation at 5 years). Level of education was evaluated according to International Standard Classification of Education in years.17
Coronary Artery Calcification
Non-enhanced EBCT scans were performed with a C-150 scanner (GE Imatron, South San Francisco, CA). Prospective ECG triggering was done at 80% of the RR interval. Contiguous 3-mm-thick slices were obtained at an image acquisition time of 100 ms. CAC was defined as a focus of ≥ 4 contiguous pixels with a CT density ≥130 Hounsfield units. The CAC Agatston score was computed by summing weighted CAC scores of all foci in the epicardial coronary system.18 The CAC score was communicated neither to the participants nor to their physicians.
Continuous data are presented as mean±SD (normally distributed data) or median (Q1;Q3; non-normally distributed data), categorical data as counts (%). Cox proportional hazards models for continuous and categorical regressors were used to evaluate predictors of stroke risk. C-statistics for time-to-event data (Harrell's c), category-free net reclassification improvement, and integrated discrimination improvement for time-to-event data were also analyzed. Incidence rates and hazards ratios (HR) were computed with their 95% confidence intervals. P values <0.05 indicate statistical significance. Details are given in the Statistics of the online-only Data Supplement.
The baseline characteristics of HNR subjects receiving CAC measurements are summarized in Table 1. A total of 92 subjects (55 men, 37 women) developed a stroke during the follow-up period [82 ischemic, 10 hemorrhagic]. Of the ischemic strokes, 11 were macroangiopathic, 14 microangiopathic, 22 cardioembolic, and 35 unknown/other pathogenesis. Subjects experiencing a stroke were older, had a higher body mass index, more often revealed arterial hypertension and diabetes mellitus, and had higher triglycerides and a higher FRS than subjects without stroke. Subjects with a stroke more frequently exhibited a clinical history of peripheral artery disease and more frequently received antihypertensive and antidiabetic medications.
The median age of patients experiencing a stroke event during the follow-up period was 65 years. All acquired risk factors were more prevalent in subjects >65 years rather than ≤65 years of age. The percentage of current smokers was lower in subjects >65 years rather than ≤65 years of age, whereas the CAC score was higher (Table I in the online-only Data Supplement). Antihypertensives, lipid-lowering drugs, antidiabetics, and platelet inhibitors were more frequently prescribed in old than in young subjects (Table I in the online-only Data Supplement).
When stratified by sex, all vascular risk factors except total and LDL cholesterol were more often noted in men than in women (Table II in the online-only Data Supplement). In line with this, stroke incidence was higher in men than in women, and men more often received antidiabetics and platelet inhibitors than women (Table II in the online-only Data Supplement).
When stratified by CAC, the prevalence of all risk factors increased with higher CAC categories, as did stroke incidence (Table III in the online-only Data Supplement). Antihypertensive, lipid-lowering, antidiabetic, and antiplatelet drugs were more frequently documented in higher than in lower CAC categories (Table III in the online-only Data Supplement).
CAC as Independent Stroke Predictor
CAC revealed a skew distribution with a high shoulder at the 0 value and decreasing frequencies toward higher values (median, 11.6 [Q1;Q3, 0;111.3]). Of 4180 participants, 1361 revealed no CAC. To understand how CAC influences stroke risk, we first examined stroke incidence rates for various CAC intervals. In participants without CAC, stroke incidence was low (1.08 per 1000 person-years; Figure 1). Stroke incidence increased with detection of CAC (2.52 and 3.33 per 1000 person-years for 1≤CAC≤99 and 100≤CAC≤399, respectively), reaching an incidence rate of 9.23 per 1000 person-years for CAC values ≥400 (Figure 1).
To elucidate how CAC affects stroke risk in addition to established risk factors, we next performed Cox proportional hazards regression analyses, including age, sex, systolic blood pressure, LDL, high-density lipoprotein, diabetes mellitus, smoking status, and log10(CAC+1). In an unadjusted regression analysis and a regression analysis adjusted for age and sex (model 1 in Table 2), log10(CAC+1) predicted stroke risk. In an analysis, including age, sex, and all other Framingham risk factors, the factors age (HR, 1.35 per 5 years [1.15–1.59]; P<0.001), systolic blood pressure (HR, 1.25 per 10 mm Hg [1.14–1.37]; P<0.001), smoking (HR, 1.75 [1.07–2.87]; P=0.025), and log10(CAC+1); (HR, 1.52 [1.19–1.92]; P=0.001) were independent stroke predictors (model 2 in Table 2). When patients developing hemorrhagic stroke during the follow-up period were excluded from analyses, log10(CAC+1) remained an independent stroke predictor (HR, 1.39 [1.09–1.79]; P=0.009). Results did not change relevantly, when analyses were restricted to nondiabetics (not shown).
In the next step, we calculated Harrell's c-statistics (AUC[t]), category-free net reclassification improvement, and integrated discrimination improvement for time-to-event data to analyze the benefit of adding CAC to the model based on Framingham risk factors (age, sex, and the risk factors systolic blood pressure, LDL, high-density lipoprotein, diabetes mellitus, smoking) in stroke discrimination. This analysis revealed that the full model, including age, sex, risk factors, and log10(CAC+1); AUC[t]=0.765), predicted strokes better than the model including age, sex, and risk factors only (0.743). This increase by 0.021 showed a trend toward statistical significance (95% confidence interval, −0.004 to 0.047; P=0.098). Category-free net reclassification improvement was 32.90% (P=0.002), and integrated discrimination improvement was 0.35% (P=0.289).
For myocardial infarcts, the risk categories CAC=0, 1≤CAC≤99, 100≤CAC≤399, and CAC≥400 have previously been used for risk stratification.9,10 To evaluate whether these CAC categories also predicted stroke events, we next performed Cox proportional hazards regression analyses with categorical CAC data, showing that compared with CAC=0, CAC values ≥400 predicted an elevated stroke risk (HR, 3.34 [1.55–7.21]; P=0.02) in a multivariable analysis that included age, sex, and all other Framingham risk factors.
CAC Predicts Stroke Events in Men and Women
To evaluate the impact of sex, we next computed separate Cox proportional hazards regressions for men and women. Both in unadjusted regressions and regressions adjusted for age (model 1 in Table IV in the online-only Data Supplement), log10(CAC+1) predicted stroke risk in men and women (age-adjusted HR, 1.83 [1.32–2.54] and 1.52 [1.09–2.12], respectively). In a regression adjusted for age and Framingham risk factors, which in view of the limited number of stroke events was calculated for men but not women, log10(CAC+1) remained an independent stroke predictor (model 2 in Table IV in the online-only Data Supplement).
CAC Predicts Stroke Events Specifically in Young Subjects
To assess whether the stroke predictor log10(CAC+1) is modified by age, we calculated Cox proportional hazards regressions for subjects ≤65 years and >65 years of age. In unadjusted regressions, regressions adjusted for sex (model 1 in Table V in the online-only Data Supplement), and regressions adjusted for sex and Framingham risk factors (model 2 in Table V in the online-only Data Supplement), log10(CAC+1) predicted stroke events in younger but not in older subjects (fully adjusted HR, 2.21 [1.59–3.06] and 1.11 [0.80–1.54], respectively).
CAC Discriminates Stroke Risk Independent of Atrial Fibrillation
Atrial fibrillation predisposes to stroke events via cardiac thromboembolism, which is responsible for ≈1 of 5 ischemic strokes. Resulting in the majority of cases from coronary heart disease, we wondered whether the predictive role of CAC depends on the presence of atrial fibrillation. Of 4180, 52 participants in this study exhibited atrial fibrillation. In a multivariable Cox proportional hazards regression that involved the factors age, sex, Framingham risk factors, atrial fibrillation at baseline, and log10(CAC+1), log10(CAC+1) remained an independent stroke predictor (HR, 1.45 [1.14–1.84]; P=0.003; model 3 in Table 2).
Besides its association with atrial fibrillation at baseline, CAC may have predisposed to the development of atrial fibrillation during the follow-up period, which may have been responsible for subsequent stroke events. Fifty-three participants revealed hitherto unknown atrial fibrillation on the follow-up visit after 5 years that was not detectable at baseline. In a multivariable Cox proportional hazards regression that involved the factors age, sex, vascular risk factors, atrial fibrillation at baseline and on occasion of the follow-up, and log10(CAC+1), log10(CAC+1) was still an independent stroke predictor (fully adjusted HR, 1.31 [1.00–1.71]; P=0.049; model 4 in Table 2).
CAC Predisposes to Stroke Risk Independent of Preceding Myocardial Infarcts or Sudden Cardiac Death
Five participants in this study experienced myocardial infarcts (n=4 participants) or cardiac arrest followed by successful resuscitation (n=1) before their stroke. When these subjects were excluded from the analyses, log10(CAC+1) remained a significant stroke predictor when combined with Framingham risk factors (age, sex, systolic blood pressure, LDL cholesterol, high-density lipoprotein cholesterol, diabetes mellitus, and smoking) and atrial fibrillation at baseline (HR, 1.54 [1.21–1.97]; P=0.001).
CAC Discriminates Stroke Incidence in Subjects With Low and Intermediate Vascular Risk
To elucidate how CAC modifies stroke risk in the presence of vascular risk factors, we formed compound risk groups based on CAC and FRS categories. Inserting these groups in Cox regression revealed that compared with subjects belonging to the low FRS (<10%) and lowest CAC (0) category, subjects of the high FRS (>20%) and highest CAC (≥400) category carried an 11.73-fold stroke risk (Figure 2). Log-rank tests of trend showed that in subjects belonging to the low (<10%) and intermediate (10% to 20%) FRS categories, stroke risk increased with increasing CAC category (P for trend <0.001 and =0.022, respectively). Thus, CAC detected subjects with high stroke incidence in these low- and intermediate-risk groups. In the high FRS category, CAC did not discriminate stroke risk (P=0.360).
Using a sample of 4180 subjects 45 to 75 years of age, we showed that CAC is an independent predictor of future stroke events in the general population. CAC predicted stroke in men and women, more potently in subjects ≤65 years than >65 years of age. CAC predicted stroke independent of the presence of atrial fibrillation, an established cause of stroke. CAC discriminated stroke risk specifically in subjects belonging to the low (<10%) and intermediate (10% to 20%) FRS categories.
CAC, evaluated by EBCT, has previously been shown to predict incident myocardial infarcts in the general population.7–11 In the HNR cohort, CAC discriminated incident cardiovascular events most potently in subjects belonging to an intermediate-risk cohort, in which CAC identified subjects at high vascular risk.9 Reflecting a measure of coronary atherosclerotic plaque burden, CAC images subclinical atherosclerosis in vascular territories that are directly affected by atherothrombotic events in case myocardial infarcts take place. The question arose whether CAC is similarly suitable as marker of systemic atherosclerosis and whether CAC is able to predict vascular events outside the coronary arteries.
In contrast to 3 previous studies, the CHS,7 the Rotterdam study,11 and MESA,8 which were unable to identify CAC as stroke predictor, most likely because of lack of power related to lower stroke numbers observed in the population sample (28, 52, and 59 incident strokes, respectively), we herein demonstrated that CAC predicts stroke events in the general population, for both men and women. CAC predicted stroke independent of the presence of atrial fibrillation, which results in the majority of patients from coronary heart disease and represents an established cause of stroke events. These data revealed that the predictive value of CAC was not limited to atrial fibrillation, resulting as a consequence of coronary heart disease. Instead, CAC seems to reflect a marker of generalized plaque burden, indicating the presence of systemic atherosclerotic disease.
In this study, CAC more potently predicted stroke events in subjects ≤65 than >65 years of age. Moreover, CAC discriminated stroke risk in subjects belonging to low (<10%) and intermediate (10% to 20%) FRS categories, indicating that CAC is able to identify individuals at risk among subjects with low or moderate classical risk profile. These observations indicate that among cohorts without apparent risk, subjects exist that nonetheless exhibit a high stroke incidence. On the basis of our data, CAC is suitable to identify those subjects. The obvious strength of this study is its large size and observation period, which provided an adequate power for analyzing CAC as stroke predictor, and the validation of stroke events by an independent committee not involved in the further data analysis. That CAC, as we now have shown, is able to predict stroke events independent of established risk factors, makes this marker promising for risk stratification not only in the hands of cardiologists but also in the hands of neurologists. Evaluation of CAC by EBCT involves radiation exposure, which needs to be considered when taking into account CAC for risk stratification. The radiation exposure in EBCT so far was 1 to 1.3 mSv.19 Newer multislice computed tomography could demonstrate a reduction of radiation exposure to <0.3 mSv,20 so that this problem will substantially be reduced in the future.
We thank the Investigative Group, the staff, and all participants for assistance.
Sources of Funding
This work was supported by the Heinz Nixdorf Foundation (Chairman: Martin Nixdorf, past chairman: Dr Gerhard Schmidt), the Bundesministerium für Bildung und Forschung (BMBF), and the Deutsche Forschungsgemeinschaft (DFG; Projects SI236/8-1, SI236/9-1, and ER155/6-2). Sarstedt AG (Nümbrecht, Germany) supplied laboratory equipment.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.678078/-/DC1.
- Received September 21, 2012.
- Accepted January 7, 2013.
- © 2013 American Heart Association, Inc.
- Wolf PA,
- Abbott RD,
- Kannel WB
- del Sol AI,
- Moons KG,
- Hollander M,
- Hofman A,
- Koudstaal PJ,
- Grobbee DE,
- et al
- Ohira T,
- Shahar E,
- Iso H,
- Chambless LE,
- Rosamond WD,
- Sharrett AR,
- et al
- Folsom AR,
- Kronmal RA,
- Detrano RC,
- O’Leary DH,
- Bild DE,
- Bluemke DA,
- et al
- Erbel R,
- Möhlenkamp S,
- Moebus S,
- Schmermund A,
- Lehmann N,
- Stang A,
- et al
- Elias-Smale SE,
- Wieberdink RG,
- Odink AE,
- Hofman A,
- Hunink MG,
- Koudstaal PJ,
- et al
- Adams HP Jr.,
- Bendixen BH,
- Kappelle LJ,
- Biller J,
- Love BB,
- Gordon DL,
- et al
- Chobanian AV,
- Bakris GL,
- Black HR,
- Cushman WC,
- Green LA,
- Izzo JL Jr.,
- et al
- Wilson PW,
- D’Agostino RB,
- Levy D,
- Belanger AM,
- Silbershatz H,
- Kannel WB
- 17.↵United Nations Educational, Scientific and Cultural Organization. International Standard Classification of Education. Paris, 1997.
- Schuhbaeck A,
- Achenbach S,
- Layritz C,
- Eisentopf J,
- Hecker F,
- Pflederer T,
- et al