Coronary Artery Calcium and Risk of Dementia in the Multi-Ethnic Study of Atherosclerosis

Abstract

Background

Studies suggest a link between vascular injuries and dementia. Only a few studies, however, examined a longitudinal relation of subclinical vascular disease with dementia. We tested whether baseline coronary artery calcium (CAC), a biomarker of subclinical vascular disease, is associated with incident dementia independent of vascular risk factors (VRF) and APOE-ε4 genotype in a community-based sample.

Methods and Results

We analyzed 6,293 participants of the Multi-Ethnic Study of Atherosclerosis (MESA), aged 45–84 years at baseline (2000–2002), initially free of cardiovascular disease (CVD) and noticeable cognitive deficit. Dementia cases were identified using hospital and death certificate ICD codes. Cox models were used to obtain hazard ratios according to CAC category, or per 1 standard deviation (SD) log2[CAC+1], adjusted for VRF, APOE-ε4, with or without exclusion of interim stroke or CVD. We observed 271 dementia cases in a median follow-up of 12.2 years. Baseline CAC had a graded positive association with dementia risk. Compared to no CAC, CAC score of 1–400, 401–1000, and ≥1001 had increased risk of dementia by 23%, 35%, and 71%, respectively (P_trend=0.026) after adjustment. 1SD higher log2[CAC+1] was associated with 24% (95%CI: 8–41%, P=0.002) increase in dementia risk. Although the association was partially explained by interim stroke/CVD, it remained significant even after excluding the interim events, or regardless of baseline age.

Conclusions

Higher baseline CAC was significantly associated with increased risk of dementia independent of VRF, APOE-ε4, and incident stroke. This is consistent with a hypothesis that vascular injuries play a role in development of dementia.

Vascular risk factors predict cognitive decline and dementia.^1–3 Coronary artery calcium (CAC) is subclinical vascular disease and can be viewed as marking biological response to cumulative vascular injuries and may predict dementia independently of baseline vascular risk factors. Only a few longitudinal studies, however, reported on dementia and subclinical vascular disease.^4–6

Our primary aim is to examine the association of CAC with dementia risk in a community-based sample. We hypothesize that baseline CAC predicts future dementia independently of conventional vascular risk factors and Apolipoprotein E -ε4 genotype (APOE4). We also examine whether the association is attenuated by accounting for interim stroke or cardiovascular disease (CVD) since such clinical vascular manifestations increased future risk of dementia.⁷

METHODS

Study Population

Participants were from the Multi-Ethnic Study of Atherosclerosis (MESA), a study of the prevalence, risk factors, and progression of subclinical CVD in a multiethnic cohort in the United States.⁸ In brief, 6,814 participants aged 45–84 years who identified themselves as white, black, Hispanic, or Chinese, free of clinically apparent CVD and other serious illness were recruited from 6 US communities in 2000–2002. Cognitive inability in the screening interview was an exclusion criterion (an interviewer judged whether the respondent was cognitively able to participate in the study). For the present study, we utilized follow-up and events data through 2013. All participants in the analyses gave informed consent, and the study protocol was approved by the Institutional Review Board at each site.

Coronary artery calcium score

Detailed methods for computed tomography (CT) scan technique and interpretation were previously described.⁹ CT scans were performed twice per participant at baseline, and read centrally at the Harbor-University of California, Los Angeles Research and Education Institute to identify and quantify coronary calcium. We analyzed the baseline total Agatston score (CAC score)¹⁰ averaged across the two CT scans. Agreement for presence of coronary calcium was high (kappa-statistic 0.90 to 0.93 between and within readers), and the intraclass correlation coefficient for CAC score between readers was 0.99.⁹

Other measures at baseline

Standardized questionnaires were used to obtain information at baseline including highest level of education attained, possession of health insurance, physical activity, smoking history, and medication usage. Height and weight were measured in light clothing and no shoes. Physical activity level was quantified as average metabolic equivalent (MET) -minutes per week of moderate or vigorous activity using the MESA Typical Week Physical Activity Survey.¹¹ Body mass index (BMI) was calculated as weight divided by height squared (kg/m²). Obesity was defined as BMI≥ 30 kg/m². Seated resting blood pressure was measured 3 times with a Dinamap model Pro 100 automated oscillometric sphygmomanometer (Critikon). The average of the last 2 measurements was used in analysis.¹² Total and high-density lipoprotein (HDL) cholesterol, triglycerides, and glucose concentrations were measured from blood samples obtained after a 12-hour fast. Non high-density lipoprotein cholesterol (non-HDL-C) was calculated by subtracting HDL cholesterol from total cholesterol. Diabetes was defined as fasting glucose >6.99 mmol/L (126 mg/dL) or use of hypoglycemic medication. APOE4 isoforms were estimated from single nucleotide polymorphisms rs429358 and rs741 using the algorithm reported previously.¹³ It was categorized here according to carriage of APOE4 as homozygous, heterozygous or nullizygous (non-carrier).

Ascertainment of clinical events

In MESA, telephone interviews inquired about interim hospital admissions and deaths every 9 to 12 months. Copies of death certificates and corresponding ICD10 (the International Statistical Classification of Diseases and Related Health Problems, 10^th version) codes, face sheets and ICD9 codes from hospital records and some outpatient diagnoses were assembled at each center. Medical records were purposed to adjudicate cardiovascular events. An attempt was made to obtain at least ICD coding for each hospitalization or death. If the events were related to MESA outcomes, the full medical records were requested and the clinic coordinator submitted copies of parts deemed relevant to the MESA adjudication committee for review. CVD-related endpoints were classified by two physicians from the MESA mortality and morbidity review committee.¹⁴ Methodology of identification of incident dementia and its validity was reported elsewhere.¹⁵ In brief, we identified candidate dementia cases using the following diagnosis codes: ICD9: 290, 294, 331.0, 331.1, 331.2, 331.82, 331.83, 331.9, 438.0, and 780.93; ICD10: F00, F01, F03, F04, G30, G31 [excluding G31.2], I69.91, and R41. A physician (AF), blinded to ICD codes and CAC-status, reviewed available medical records for the potential cases identified with the ICD codes. We excluded those who were judged as “likely non-dementia” from our cases.

We only included those with no missing variables pertinent to analyses (N=6293).

Statistical Analyses

The participants were categorized according to baseline CAC score of 0, 1 to 400, 401 to 1000, and ≥1001. To test a graded relation of baseline characteristics across CAC categories, p-values were obtained with linear regression for continuous, and Mantel-Haenszel Chi-Square test for categorical variables.

We first calculated crude rates of dementia across CAC categories. Adjusted hazard ratio (HRs) and 95% confidence intervals (95%CIs) were obtained by Cox regression models. We treated time of first qualifying dementia ICD-code as time of event. Upon visual inspection of adjusted survival curves, we observed an overall gradient relationship between CAC category and dementia risk. Then, we treated CAC score as continuous using base-2 logarithm of CAC score plus 1 (log2 [CAC+1]) in accordance with prior studies.^{16, 17} For “minimal adjustment”, we adjusted for age (years), gender, and race/ethnicity. For “full adjustment”, we further adjusted for highest education attained (up to high school/up to associate degree/bachelor’s degree or higher), having health insurance (yes/no), physical activity (log-transformed MET-minutes/week), smoking (current/former/never), obesity (yes/no), systolic blood pressure (mmHg), non-HDL-C (mg/dL), medication(s) for hypertension or dyslipidemia (yes/no), diabetes mellitus (yes/no), and APOE4 (homo/hetero/none). There was no evidence against the proportional hazard assumption for the CAC categories and dementia risk based on the Schoenfeld residuals and visual inspection of the log-log curve in the full adjustment model.

In secondary analyses to study mediation of the association, we additionally adjusted for time-dependent stroke or CVD (myocardial infarction, resuscitated cardiac arrest, or stroke) occurring before dementia diagnosis. We also ran models excluding such interim events. In sensitivity analyses, we used a competing risk model using the proportional subdistribution hazards model¹⁸ with otherwise the same set of adjusting covariates. In this model, we treated death without dementia diagnosis as the risk competing with dementia. To address possible residual confounding by age and birth cohort effect, we conducted a model using age as the time scale with stratification by 3 birth cohorts: <1925 (n=626, [age range at baseline, 77 to 84 years], 1925 to <1935 (n=1688, [67 to 76 years]), 1935 or after (n=3979, [45 to 66 years]). This model may be preferred when age at event may have a larger effect on the hazard (i.e. dementia) than follow-up time period,¹⁹ since age is a strong risk factor for dementia and CAC. Finally, we ran the full adjustment model according to the baseline age (<75 years vs ≥75 years). The age cutoff was chosen so that number of dementia cases was approximately halved for each age group. Interactions by sex, race/ethnicity and baseline age (<75 or ≥75 years) on the relation between CAC and dementia were tested by inserting a product term in the full adjustment model. All adjusting covariates except interim stroke and CVD were assessed at baseline.

Because an interaction between atherosclerosis and APOE4 genotype was suggested in some studies,^{20, 21} we examined whether the association of CAC with dementia differed by APOE4 status (carrier/non-carrier) inserting a product term in the full adjustment model.

Statistical significance was at P < 0.05 (2-sided P value). SAS software (version 9.4; SAS Institute, Inc., Cary, North Carolina) was used for all statistical analyses.

RESULTS

We analyzed 6,293 participants (47.5% men, 12.2% Chinese, 26.1% Black, 22.5% Hispanic, and 39.2% White) and observed 271 dementia cases, 173 interim strokes, and 385 interim CVD events during the median follow-up of 12.2 (inter-quartile range: 11.6–12.7) years. The proportions with baseline CAC score of 0, 1–400, 401–1000, and ≥1001 were 49.8% (n=3,132), 40.3% (2,539), 6.1% (384), and 3.8% (238), respectively. Those with higher baseline CAC tended to be older, white, male, less physically active, former smoker, user of antihypertensive and lipid-lowering medications, having higher systolic blood pressure, and diabetics (Table 1). Crude dementia rates in both sexes increased steeply with age (Supplementary Table 1). Table 2 shows mutually adjusted HRs for the covariates in the “full adjustment” plus interim stroke. Interim stroke had the greatest magnitude of association with dementia risk (HR=5.66) followed by older age. Male gender, current smoker and CAC were positively associated with dementia, while physical activity was inversely associated. Chinese participants had lower dementia risk than white participants. APOE4 genotype had a graded association with dementia; HRs of heterozygous and homozygous compared to non-carriers were 1.33 and 2.20, respectively. Interim CVD was also strongly associated with dementia (HR=5.08, Supplementary Table 2).

Table 1.

Demographics of the participants according to coronary artery calcium score at baseline (N=6293, Baseline 2000–2002)

	Coronary artery calcium (CAC) score at baseline				P-value
	0 (n=3,132)	1 to 400 (n=2,539)	401 to 1000 (n=384)	≥ 1001 (n=238)
Age, years (sd)	58.0 (9.2)	65.4 (9.7)	69.8 (7.9)	71.7 (7.5)	<0.001
Men, no. (%)	1156 (36.9)	1393 (54.9)	248 (64.6)	193 (81.1)	<0.001
Race/ethnicity, no. (%)					<0.001
White	1068 (34.1)	1073 (42.3)	196 (51.0)	127 (53.4)
Chinese	380 (12.1)	337 (13.3)	38 (9.9)	13 (5.5)
Black	915 (29.2)	595 (23.4)	81 (21.1)	53 (22.3)
Hispanic	769 (24.6)	534 (21.0)	69 (18.0)	45 (18.9)
Highest education level attained, no. (%)					0.053
to high school	1074 (34.3)	983 (38.7)	155 (40.4)	85 (35.7)
to associate degree	931 (29.7)	689 (27.1)	98 (25.5)	64 (26.9)
No health insurance, no. (%)	345 (11.0)	192 (7.6)	15 (3.9)	10 (4.2)	<0.001
Physical activity, MET-min/week*	4271 (2040, 8096)	3840 (1905, 7073)	3705 (1808, 7035)	3450 (1793, 5955)	<0.001†
Smoking, no. (%)					<0.001
current	413 (13.2)	336 (13.2)	43 (11.2)	32 (13.5)
former	954 (30.5)	1031 (40.6)	185 (48.2)	117 (49.2)
Body mass index, kg/m2 (sd)	28.2 (5.6)	28.3 (5.4)	28.7 (5.1)	28.1 (4.5)	0.488
Obesity (BMI ≥30), no. (%)	981 (31.3)	804 (31.7)	138 (35.9)	66 (27.7)	0.800
Systolic blood pressure, mmHg (sd)	122 (20)	130 (22)	136 (22)	135 (21)	<0.001
Total cholesterol, mg/dL (sd)	194 (35)	195 (36)	196 (37)	186 (37)	0.435
HDL-cholesterol, mg/dL (sd)	52 (15)	50 (14)	49 (15)	48 (14)	<0.001
Non HDL-cholesterol, mg/dL (sd)	141 (35)	146 (36)	147 (37)	138 (36)	0.019
Antihypertensive medication, no. (%)	896 (28.6)	1088 (42.9)	208 (54.2)	147 (61.8)	<0.001
Lipid-lowering medication, no. (%)	331 (10.6)	510 (20.1)	104 (27.1)	74 (31.1)	<0.001
Diabetes, no. (%)	281 (9.0)	355 (14.0)	86 (22.4)	51 (21.4)	<0.001
APOEε-4 genotype, no. (%)‡					0.136
heterozygous	770 (24.6)	649 (25.6)	78 (20.3)	50 (21.0)
homozygous	79 (2.5)	54 (2.1)	8 (2.1)	6 (2.5)

Values were means (standard deviations) unless otherwise specified. P-values were calculated by linearly regressing on the CAC categories treated as ordinal for continuous variables, and by Mantel-Haenszel Chi-Square test for categorical variables.

^*Median (25th, 75th percentile) was given as the distribution was skewed.

^†For physical activity, log-transformed value of (1+ MET-min/week) was used to obtain the P-value).

^‡Observed numbers for each genotype in total sample were consistent with Hardy-Weinberg equilibrium (P=0.212).

Abbreviations: HDL, high-density lipoprotein; APOE, Apolipoprotein E; MET, metabolic equivalent..

Table 2.

Multivariable adjusted hazard ratio (HR) and 95% confidence interval (95%CI) for dementia risk (N=6293, from 2000 through 2013)

	HR	95% CI		P
Age, per 1SD* higher	4.83	3.96	5.90	<0.001
Female gender	0.62	0.48	0.81	<0.001
Race/Ethnicity (ref. White)
Black	1.16	0.86	1.57	0.340
Chinese	0.50	0.30	0.83	0.008
Hispanic	0.82	0.57	1.16	0.256
Higher education level attained †	0.89	0.76	1.04	0.128
No health insurance	1.36	0.72	2.59	0.341
Physical activity, per 1 SD higher *	0.88	0.80	0.98	0.015
Smoking habit (ref. never)
current	1.83	1.23	2.73	0.003
former	0.80	0.61	1.06	0.114
Obesity (BMI ≥ 30kg/m2)	0.86	0.64	1.15	0.308
Systolic blood pressure, per 1SD* higher	1.10	0.98	1.23	0.119
Antihypertensive medication	1.00	0.77	1.29	0.995
Non HDL-cholesterol, per 1SD* higher	0.97	0.84	1.11	0.626
Lipid-lowering medication	0.82	0.60	1.11	0.199
Diabetes	1.22	0.87	1.71	0.257
APOEε-4 genotype (ref. non-carrier)
heterozygous	1.33	1.02	1.74	0.035
homozygous	2.20	1.15	4.20	0.018
Interim stroke‡	5.66	3.97	8.05	<0.001
Log2[CAC+1], per 1SD* higher	1.19	1.04	1.36	0.013

All the above factors were adjusted simultaneously in Cox model with 271 dementia cases over the median follow-up of 12.2 years.

^*1 standard deviation (SD) for each variable was as follows: age, 10.3 years; physical activity (in log[MET-min/week]), 1.34; systolic blood pressure, 21.5mmHg; non HDL-cholesterol, 36.0 mg/dL; log2[CAC+1], 3.63 which approximates 12.4-times greater CAC-score.

^†Education level was categorized into the following three groups and treated as an ordinal variable: 0=up to high school, 1= up to associate degree, 2= bachelors’ degree or higher.

^‡Interim stroke was treated as a time-dependent variable.

Crude rate ratio of dementia, relative to CAC score zero, for those with CAC score of 1 to 400, 401 to 1000, and ≥1001 were 3.12, 5.05, and 8.44, respectively (Table 3). In the minimal adjustment and the full adjustment models, results were attenuated but the graded positive trend remained (Table 4). The HR associated with 1 standard deviation (SD) higher log2[CAC+1] (1 SD=3.634, equivalent to multiplying CAC score by 12.4) was 1.24 (95%CI 1.08, 1.41, P=0.002) in the full adjustment model. Further adjustment for interim stroke, or the competing risk model attenuated the relation, but the overall association lost statistical significance only in a mediation model, adjusting for interim CVD (P=0.084). The model using age as time-scale, stratified by 3 birth-cohorts, yielded similar results to the full adjustment model. In secondary analyses excluding the participants who developed interim stroke or CVD, the associations were decreased, but remained significant using log2[CAC+1]. In stratified analysis by baseline age < 75 years (n=5388, 117 events) versus ≥75 years (n=905, 154 events), the association was attenuated in the older group, but a positive association remained significant in both age groups with full adjustment. The HRs (95%CI; p-value) per 1 SD higher log2[CAC+1] were 1.24 (1.01, 1.51; 0.038) and 1.21 (1.01, 1.45; 0.035) in the younger and older group, respectively (data not tabulated). We observed no statistical evidence supporting interaction by either sex, race/ethnicity, or age (<75y or ≥75y) regardless of representation of CAC as categories or log2[CAC+1] (p for interaction >0.5 in all models tested). There was no indication supporting interaction by APOE4 on the association between CAC and dementia (p-values for product term were ≥0.14 whether CAC was in categories or log2[CAC+1]).

Table 3.

Crude rates of dementia according to coronary artery calcium score at baseline (N=6293, from 2000 through 2013)

	Coronary artery calcium score at baseline				Total
	0	1 to 400	401 to 1000	≥ 1001
No. at Risk (baseline)	3,132	2,539	384	238	6,293
Follow-up period, person-years	36,231	27,618	3,880	2,251	69,981
Dementia, no.	61	145	33	32	271
Dementia rate, per 1,000 person-years	1.68	5.25	8.50	14.22	3.87
Crude rate ratio for dementia	1.0 (ref)	3.12	5.05	8.44	-

Table 4.

Hazard ratios of dementia according to coronary artery calcium (CAC) score at baseline (N=6293, from 2000 through 2013)

	CAC category										Continuous CAC
	0	1 to 400			401 to 1000			≥ 1001			per 1SD5) higher CAC [log2(CAC+1)]
		HR	(95% CI)	P	HR	(95% CI)	P	HR	(95% CI)	P	HR	(95% CI)	P
Model
Minimal adjustment1)	1.0 (ref)	1.27	(0.93, 1.74)	0.127	1.37	(0.89, 2.13)	0.157	1.65	(1.04, 2.63)	0.033	1.23	(1.08, 1.40)	0.002
Full adjustment1)	1.0 (ref)	1.23	(0.90, 1.69)	0.189	1.35	(0.86, 2.11)	0.187	1.71	(1.07, 2.73)	0.026	1.24	(1.08, 1.41)	0.002
Full adjustment + interim stroke2) adjusted	1.0 (ref)	1.13	(0.82, 1.56)	0.445	1.21	(0.77, 1.90)	0.412	1.51	(0.94, 2.42)	0.087	1.19	(1.04, 1.36)	0.013
Full adjustment + interim CVD2) adjusted	1.0 (ref)	1.07	(0.78, 1.48)	0.664	1.06	(0.67, 1.67)	0.799	1.26	(0.78, 2.04)	0.341	1.13	(0.98, 1.29)	0.084
Competing risk model (full adjustment) 3)	1.0 (ref)	1.18	(0.86, 1.62)	0.310	1.23	(0.78, 1.95)	0.364	1.40	(0.85, 2.29)	0.183	1.17	(1.02, 1.35)	0.023
Age as time-scale, stratified by birth cohort 4)	1.0 (ref)	1.22	(0.89, 1.67)	0.226	1.33	(0.85, 2.08)	0.220	1.65	(1.03, 2.63)	0.037	1.22	(1.07, 1.40)	0.003
Full adjustment with exclusion
interim stroke excluded (n=6120, case=241)	1.0 (ref)	1.11	(0.81, 1.54)	0.513	1.16	(0.72, 1.87)	0.553	1.50	(0.92, 2.46)	0.106	1.18	(1.03, 1.36)	0.018
interim CVD excluded (n=5908, case=220)	1.0 (ref)	1.10	(0.79, 1.53)	0.569	0.96	(0.57, 1.63)	0.883	1.60	(0.95, 2.68)	0.075	1.17	(1.01, 1.35)	0.039

¹⁾Adjusting covariates in “Minimal adjustment”: age (years), sex, and race (White/Black/Chinese/Hispanic). “Full adjustment” further included education level (to high school/to associate degree/bachelor’s degree or higher), having health insurance (yes/no), physical activity (MET-minutes/week, log-transformed), smoking (current/former/never), obesity (yes/no), hypertension (yes/no), medication(s) for hypertension or dyslipidemia (yes/no), systolic blood pressure (mmHg), non high-density lipoprotein cholesterol (mg/dL), diabetes mellitus (yes/no), APOE ε4 genotype (homo/hetero/none).

²⁾Interim stroke and interim CVD were treated as time-dependent variables.

³⁾In the competing risk model, death without dementia diagnosis was treated as the competing risk to dementia.

⁴⁾The results were based on cox regression model stratified with 3-birth cohorts (1917 to <1925, 1925 to <1935, and thereafter).

⁵⁾1SD (standard deviation) of log2[CAC+1] =3.634, equivalent to multiplying CAC score by 12.4.

The corresponding HRs (95%CIs) per doubling [CAC+1] in models from the top to the bottom rows were 1.06 (1.02, 1.10), 1.06 (1.02, 1.10), 1.05 (1.01, 1.09), 1.03 (1.00, 1.07), 1.05 (1.01, 1.09), 1.06 (1.02, 1.10), 1.05 (1.01, 1.09), and 1.04 (1.00, 1.09).

DISCUSSION

In this community-based multiethnic cohort sample, we found that baseline CAC was associated with future dementia risk independent of baseline vascular risk factors and APOE4 genotype. Although the positive association was attenuated to non-significance by adjustment for interim CVD, it remained significant after excluding interim stroke/CVD.

With some inconsistency,²² vascular risk factors have been shown to predict subsequent cognitive decline or dementia,^1–3 perhaps depending on when risk factors were measured over the life course, as some risk factors such as blood pressure and smoking habits change over time. CAC, in contrast, is a cumulative, quantitative biomarker of vascular injury. To our knowledge, this is the first population-based study that prospectively showed a positive graded association of CAC with dementia risk not only independent of vascular risk factors and APOE4, but also in the absence of clinical CVD. Our finding is consistent with the conjecture that vascular injury plays an important pathogenetic role in dementia, since CAC may correlate well with subclinical vascular changes in the brain.^{23, 24} A few longitudinal studies,^{4, 5, 7} have reported an association of carotid atherosclerosis with dementia. In many of those, only higher levels of IMT (e.g. >80^th percentile) was associated with (total) dementia with no clear dose-response relation.^{4, 5, 7} In longitudinal analysis of a subgroup of the Cardiovascular Health Study (CHS), the participants with CAC score of ≥400 at baseline, had higher dementia risk than those with low score of 0–10.⁶ However the average age of the participants was much older (80 years) and the sample size was smaller (N=532) than our study.

We observed remarkably elevated dementia risk among those who developed interim stroke or CVD. Exclusion of those participants attenuated the association between CAC and dementia. Certainly, stroke increases the risk of vascular-related dementia, and CAC is known to predict stroke.²⁵ Beyond that, our results suggest that overall CVD is a risk factor for dementia, conceivably related to subclinical cerebrovascular changes, which is consistent with the CHS report of elevated dementia risk among those with prevalent CVD, other than stroke.⁷ The pathway via clinical stroke to dementia is known to be only the tip of the iceberg. Recent evidence suggests an important role of cerebral small vessel disease, manifested as lacunes, microbleeds, and white matter hyperintensities in magnetic resonance imaging (MRI), for both vascular and neurodegenerative dementia.^{26, 27} From this perspective, previous studies give insights into potential mechanisms linking CAC and dementia. The Age, Gene, Environment Susceptibility (AGES)-Reykjavik Study, a cross-sectional study in Iceland, for example, reported not only a positive association of CAC with dementia/cognitive performance, but also an association between CAC and cerebral pathologies (infarcts, microbleeds, white matter lesions, and lower brain volume) assessed with brain MRI. Interestingly, in their analysis, the relation of dementia/cognitive performance to CAC was significantly attenuated after adjustment for those brain findings.²³ Similarly, the Rotterdam study and the CHS reported cross-sectional associations of CAC with (subclinical) brain pathologies and abnormal cognitive status.^{24, 28, 29} In our model, adjustment for interim CVD attenuated the association of CAC with dementia to non-significance. This may be because those with interim CVD had more advanced atherosclerosis in the brain at the CAC measurement than those without, providing a mechanism for the relationship and attenuating the overall association. Although exclusion of interim events may result in a biased estimate, that estimate is more conservative than in the full adjustment model. Therefore, the existing literature, together with our findings, suggests that CAC indicates presence of the coexisting brain pathologies that increase dementia risk.^{23, 24, 28, 29}

MESA previously found a positive association of continuous CAC with dementia risk, one of the multiple outcomes evaluated in that report.¹⁷ Compared to the earlier report, our dementia ascertainment was more thorough and validated,¹⁵ utilizing not only hospital records used in the previous study but also death certificates and more inclusive dementia-defining codes, which enabled us to almost double the number of cases. Other novel findings shown in our study but not in the previous one include a clear dose-response relation of CAC with dementia in multivariable adjustment including APOE4, and persistent association of CAC with dementia after excluding or adjustment for interim stroke events, supporting the importance of subclinical vascular injury in occurrence of dementia.

APOE4 is an established risk factor for Alzheimer’s disease. Effect size of APOE4 carriers observed in our study seems to be weaker than other cohort studies.^30,31 Because those previous reports studied only on Alzheimer disease, our estimate is likely diluted due to inclusion of other type of dementia. Our results did not support interaction by APOE4 on the association of atherosclerosis (CAC) with dementia in contrast with some,^{20, 21} but not all,^{4, 7} previous studies. This question, therefore, warrants further investigation.

Validity of ICD-based code

Since we utilized ICD-codes from hospitalization and death certificate in identifying dementia cases, and given that dementia is a non-primary event of interest in MESA, we may have missed cases from non-hospitalized, and or less advanced dementia (false negatives).^{32, 33} Our estimate is, therefore, likely more applicable to advanced cases of dementia. Nevertheless, our findings were generally consistent with other population-based studies that utilized more clinically-oriented case ascertainments (i.e. cognitive screening tests on the entire cohort followed by clinical assessment as indicated).^{4, 34} In addition, our model found multiple factors associated with dementia, directionally consistent with the current literature: increased risk with current smoking,³⁵ less physical activity,³⁵ and APOE4 carriers³⁵ in a graded fashion. All those findings suggest internal validity of our case identification.

Limitations and Strengths

We were unable to study the association of CAC with major dementia subtypes: Alzheimer and vascular type. These subtypes may be difficult to differentiate as necropsy studies have shown that each neuropathological feature commonly coexists in demented elderly adults.^{36, 37} We believe that such differentiation is of interest, but not necessarily crucial in examining our hypothesis because vascular injuries have been related not only to vascular type dementia, but also to Alzheimer types,^{4, 5, 7, 38} and these two sub-types can be viewed as a continuum (i.e. “mixed type” dementia).^{26, 27}

Importantly, we are unable to isolate the potential independent contribution of CAC from that of cerebral small vessel disease²⁶ with which it is known to be related. This does not, however, diminish the important strong predictive value of the CAC measure we demonstrated. An advantage of the ICD-based approach compared to cognitive function testing alone is that the former is based on a clinical diagnosis. Any type of cognitive function test administered at one time has an inherent limitation, influenced by fluctuation of concentration, and inability to diagnose cognitive decline from their baseline. A strength of our study was verification of cases by review on medical records to minimize false positives.

In conclusion, our study showed a positive graded association between baseline CAC score and future dementia risk independent of baseline vascular risk factors, APOE4 genotype and interim stroke events in a community-based sample of men and women. The results support the important role of vascular injuries in pathogenesis of dementia, although whether CAC indicates presence of dementia-related pathologies in the brain warrants further investigation.

Clinical Perspective.

Although studies suggest that vascular injuries play a role in pathogenesis of dementia, only a few studies examined a longitudinal relation of subclinical vascular disease with dementia. We examined whether baseline coronary artery calcium (CAC), a quantitative biomarker of subclinical vascular disease, is associated with incident dementia independent of vascular risk factors (VRF) measured at baseline, and APOE-ε4 genotype in a community-based sample. We analyzed 6,293 participants of the Multi-Ethnic Study of Atherosclerosis (MESA), aged 45–84 years at baseline (2000–2002), initially free of cardiovascular disease (CVD: stroke included) and noticeable cognitive deficit. Two hundred seventy-one dementia cases were observed during a median follow-up of 12.2 years. Baseline CAC had a positive association with dementia risk in multivariable adjustment: CAC score of 1–400, 401–1000, and ≥1001 had elevated risk of dementia by 23%, 35%, and 71%, respectively, compared to no CAC; Doubling of CAC score was associated with 6% (95%CI: 2–10%, P=0.002) higher dementia risk. The association remained significant regardless of baseline age (<75y vs ≥ 75y) or after excluding those who developed CVD during the follow-up. The results are consistent with the “vascular injury to dementia” hypothesis, suggesting that prevention of vascular injury may reduce future dementia risk. However, more study is needed to prove this hypothesis.