Circulating Omega‐3 Polyunsaturated Fatty Acids and Subclinical Brain Abnormalities on MRI in Older Adults: The Cardiovascular Health Study

The Cardiovascular Health Study (CHS) is a prospective cohort study of 5888 older adults. The design and recruitment experience have been described.^{24, 25} Briefly, 5201 men and women aged ≥65 at baseline were randomly selected and enrolled in 1989–1990 from Medicare eligibility lists in 4 US communities: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Allegheny County, Pennsylvania. An additional 687 black participants were similarly recruited and enrolled in 1992–1993. Each center's institutional review committee approved the study. All subjects gave informed written consent.

Plasma phospholipid omega‐3 PUFAs were measured in stored plasma specimens taken at the 1992–1993 examination in 3941 participants, as described.²⁶ Blood was drawn after 12 hours of fasting, stored at −70°C, and shipped on dry ice for long‐term storage at −80°C until analyzed.²⁶ As previously described,²⁶ lipid fractions were separated by thin‐layer chromatography, and 45 individual fatty acids were measured as a percentage of total concentrations using gas chromatography. Phospholipid fatty acids represent a biomarker of longer‐term (4‐ to 8‐week) circulating concentrations, with similar responses as fatty acids in erythrocyte membranes.²⁷ No degradation, lipolysis, or oxidation has been observed after 10 years in the blood storage conditions in CHS.²⁸ Interassay CVs were 2.1% for EPA, 1.5% for DPA, 1.6% for DHA, and 3.1% for ALA. Long‐term reliability of these measures has also been described²⁶ and is similar to other major risk factors such as blood pressure.²⁹

All CHS participants were invited to undergo MRI scanning in 1991–1994. The average time between blood collection and MRI scanning was 112 days (range, −517 to 650 days). A total of 3660 (62%) underwent scanning and were slightly younger and healthier than those who did not.¹ All participants were again invited to undergo MRI scanning 5 years later, in 1997–1999, and 2313 were scanned. A total of 2116 participants underwent both scans and were healthier than the 1544 who underwent only the initial scan, including a lower prevalence of cardiovascular disease, hypertension, diabetes, current smoking, and higher income and education.⁵

The cranial MRI scanning protocol included sagittal T1‐weighted localizer images and axial T1‐, spin‐density‐, and T2‐weighted images.³⁰ Without knowledge of participants’ clinical information, neuroradiologists at the centralized CHS reading center identified infarcts and estimated white matter, ventricular, and sulcal grades, as detailed previously.^{1, 31, 32} Brain infarct was defined as an area of abnormal signal intensity ≥3 mm in size in a vascular distribution that lacked mass effect.³² Grades were defined using a semiquantitative 10‐point scale from 0 to 9 (most abnormal) based on comparison with templates.¹ Ventricular grades ranged from slit‐like ventricles (grade 0) to markedly enlarged ventricles (grade 9). Sulcal grades ranged in a similar fashion. White matter grades were estimated by the total extent of periventricular and subcortical white matter signal abnormality on spin‐density‐weighted axial images graded by successive increase from no changes or barely detectable changes (grades 0 and 1, respectively) to almost all white matter involved (grade 9).³¹ As described,³³ ventricular and sulcal grades were grouped for analysis as ≤2, 3, 4, and ≥5 and white matter grade as ≤1, 2, 3, and ≥4.

To evaluate changes in white matter between the serial MRI scans, all scans were reread side by side without knowledge of their order or their previous readings to minimize any potential reader bias.⁹ Changes in ventricular and sulcal grades were not estimated. Because of technical limitations, 197 of the 2116 paired scans could not be reread. Demographics, cardiovascular risk factors, and prevalent cardiovascular disease were similar in the 1919 included and 197 excluded participants (data not shown).

Consistent with prior analyses,¹⁷ for cross‐sectional analyses of subclinical infarcts, we excluded participants with prevalent clinical TIAs or stroke or with possible (<3‐mm) infarcts or prior hemorrhage at the first MRI (Figure 1). For prospective analyses we also excluded participants with subclinical infarcts at the first MRI or with clinically diagnosed TIAs or stroke between the 2 MRIs (Figure 2). For cross‐sectional analyses of white matter, ventricles, and sulci, we excluded participants with prevalent clinical TIAs or stroke. For serial changes in white matter grade we also excluded participants with TIAs or stroke before the second MRI.

All participants underwent extensive baseline evaluations including standard questionnaires, physical examination, performance measures, and laboratory testing.^{24, 25, 32} Prevalent coronary heart disease, stroke, TIA, hypertension, and diabetes were defined using patients’ reports and confirmed by centralized review of hospital and clinic records.^{24, 25} Dietary habits were assessed by reproducible and validated semiquantitative food‐frequency questionnaires (FFQ) in 1989–1990 and in 1995–1996.

For the analyses with incident subclinical infarcts in the secondary analyses with dietary ALA, we calculated the cumulative average intake of the 2 FFQs or used data from only 1 if the other was not available. In the analyses with prevalent subclinical findings, we used data from the first FFQ. We also examined whether our previously observed inverse association of consumption of tuna/other fish with risk of MRI findings in this cohort¹⁷ would be affected by adjustment for phospholipid long‐chain omega‐3 PUFA. Fish intake was assessed in 1989–1990 as previously described.¹⁷

We explored the univariate relationships between phospholipid EPA+DPA+DHA and ALA and baseline characteristics by means and linear regression (for continuous variables) or χ² tests (for dichotomous variables). We related the phospholipid PUFA to the risk of subclinical infarcts and worsening white matter grade by logistic regression. Only 61 subjects experienced white matter worsening by ≥2 grades between scans, so they were grouped with those experiencing worsening by 1 grade. Risk of number of subclinical infarcts (range, 0 to 4) was evaluated using ordinal logistic regression. Sulcal, ventricular, and white matter grades in quartiles of phospholipid PUFAs were evaluated using analysis of covariance (see list of covariates below). Absolute risk reduction was calculated by multiplying the absolute risk in the reference group by the multivariable‐adjusted risk reduction in the comparison group.

Multivariate models included confounders that were selected on the basis of the previously observed associations between fish consumption and MRI findings.¹⁷ The final models included age (years), sex, race (white/other), enrollment center (4 sites), diabetes (yes/no), education (<high school, high school, >high school), smoking (never/former/current), pack‐years of smoking, body mass index (kg/m²), prevalent coronary heart disease (yes/no), alcohol use (beverages/week), physical activity (kcal/week), energy intake (kcal/day), and meat consumption and vegetable consumption (servings/week in quartiles). Further adjustment for proportions of other phospholipid fatty acid biomarkers or the time between blood collection and the MRI scanning did not appreciably alter (<5%) the risk estimates (data not shown). The cohort mean was used to replace missing values in covariates (<12.5% in dietary variables, <3.0% in others). Results were similar if missing values were excluded. We also conducted a sensitivity analysis by evaluating the impact of adjusting for phospholipid long‐chain omega‐3 PUFAs on the previously observed association between tuna/other fish intake and MRI findings¹⁰ in participants with data on both exposures (n=2079). Likelihood ratio tests using multiplicative interaction terms were used to explore potential effect modification by age, sex, race, education, diabetes, coronary heart disease, hypertension, alcohol intake, smoking, or use of aspirin, lipid‐lowering, or hypertension medications. Correlations were estimated by Spearman correlation coefficients. All P values were 2‐tailed (α=0.05). Analyses were performed using SPSS 19.0 for Windows (SPSS Inc, Chicago, IL).

The mean (±SD) phospholipid proportions were 0.60% (0.39%) for EPA, 0.84% (0.17%) for DPA, 3.06% (0.98%) for DHA and 0.15% (0.05%) for ALA of total fatty acids. At baseline, higher EPA+DPA+DHA proportion was related to nonwhite race, higher education, and lower triglyceride concentrations (Table 1). Higher EPA+DPA+DHA proportion was also positively associated with consumption of tuna/other fish and estimated EPA+DHA consumption, but not with fried fish consumption, as well as with lower intake of red meat and higher intake of fruits and vegetables. Baseline characteristics according to EPA, DPA, and DHA have been published previously.²⁶

Higher phospholipid ALA proportion was related to female sex, white race, current smoking, and higher education (Table 1). Higher ALA proportion was also related to lower BMI, serum LDL cholesterol, and intake of red meat; higher intakes of tuna/other fish, fruits, and vegetables; and higher estimated intakes of EPA+DHA and ALA.

A total of 534 of 2293 participants (23.3%) had 1 or more prevalent subclinical infarcts on the first MRI. In multivariable‐adjusted analyses, the odds for having a subclinical infarct were 40% lower (OR, 0.60; 95% CI, 0.44 to 0.82; P for trend=0.001) in the highest versus the lowest phospholipid EPA+DPA+DHA quartile (Table 2). On the basis of an absolute risk of 27.1% in the first quartile, the absolute risk reduction for low EPA+DHA+DHA consumption was 10.8%. Evaluating the number of subclinical infarcts, the odds of each additional multiple infarct was 41% lower in the highest versus the lowest quartile (OR, 0.59; 95% CI, 0.44 to 0.80; P for trend<0.001). When each long‐chain omega‐3 PUFA was evaluated individually, only DHA was significantly associated with lower risk (Table 2) (reference group absolute risk=25.1%; absolute risk reduction in the highest DHA quartile=8.0%). The OR of each additional multiple infarct in the highest versus the lowest DHA quartile was 0.67 (95% CI, 0.49 to 0.90; P for trend=0.001). Simultaneous adjustment for EPA and DPA had no effect on this association (data not shown).

A total of 170 participants (16.1%) experienced an incident first subclinical infarct between the 2 MRI scans. The associations between total or individual long‐chain omega‐3 PUFAs and incident subclinical infarct did not achieve statistical significance (Table 3), but directions of association were similar to those seen for prevalent subclinical infarcts.

In cross‐sectional analyses, higher phospholipid EPA+DPA+DHA was associated with better white matter grade (Table 4). After multivariable adjustment, those in the highest quartile had 9.2% better white matter grade, compared with those in the lowest quartile (P for trend=0.001). The associations were similar for EPA (difference between the extreme quartiles 9.3%, P for trend=0.002) and DHA (difference, 8.5%; P for trend=0.001), but appeared weaker with DPA (difference, 3.6%; P for trend=0.08). When EPA, DPA, and DHA were included in the model simultaneously, only the association with DHA remained statistically significant (difference, 7.2%; P for trend=0.01). EPA+DPA+DHA or individual fatty acids were not significantly associated with markers of brain atrophy, that is, sulcal or ventricular grades (P>0.05 for each, Table 4).

In prospective analyses, 383 of 1385 participants (27.7%) experienced a worsening of white matter grade (a decrement of ≥1 grade) between the MRI scans. In multivariable‐adjusted analyses, participants in the highest EPA+DPA+DHA quartile had 42% lower odds of developing worsening white matter (OR, 0.58; 95% CI, 0.40 to 0.84; P for trend=0.002) compared with the lowest quartile (Table 5). When individual long‐chain omega‐3 PUFAs were evaluated separately, only DHA was significantly associated with lower risk (Table 5).

Phospholipid ALA content was not associated with prevalent or incident subclinical infarcts or white matter grade (Tables 2 through 5). However, higher phospholipid ALA was associated with modestly better sulcal grade (in the highest quartile, 3.8% better; P for trend=0.02) and ventricular grade (3.0% better, P for trend=0.03) compared with the lowest quartile. Adjustment for phospholipid long‐chain n‐3 PUFAs had no effect on these associations (data not shown).

In secondary analyses, dietary ALA was not significantly associated with any of the MRI findings, except for a cross‐sectional association with white matter grade, for which those in the highest quartile of consumption had a 4.4% better grade (P for trend=0.04) compared with the lowest quartile (other data not shown).

Before adjustment for phospholipid long‐chain omega‐3 PUFAs, the multivariable‐adjusted ORs of prevalent subclinical infarct across categories of fish intake were 1, 0.95, 0.84, and 0.70 (P for trend=0.09). Adjustment for phospholipid DHA substantially attenuated the association; the ORs were 1, 0.98, 0.96, and 0.87 (P for trend=0.54). In contrast, adjustment for fish intake had no effect on the association between phospholipid long‐chain omega‐3 PUFAs and MRI findings. For example, the extreme‐quartile OR for prevalent subclinical infarct in the quartiles of phospholipid DHA was 0.70 (95% CI, 0.51 to 0.96; P for trend=0.003) after adjustment. Similar results were also observed with the other MRI findings (data not shown).

We did not find evidence for significant effect modification by age, sex, race, education, diabetes, coronary heart disease, hypertension, alcohol intake, smoking, or use of aspirin, lipid‐lowering, or hypertension medications (P for interactions>0.05). We also did not find evidence for interaction between phospholipid long‐chain n‐3 PUFAs and ALA proportions (P>0.05). Exclusion of participants who used fish oil supplements (3.8%) did not change the associations (data not shown).