Hostname: page-component-7c8c6479df-fqc5m Total loading time: 0 Render date: 2024-03-29T10:04:52.173Z Has data issue: false hasContentIssue false

n–3 Fatty acids, hypertension and risk of cognitive decline among older adults in the Atherosclerosis Risk in Communities (ARIC) study

Published online by Cambridge University Press:  01 January 2008

May A Beydoun*
Affiliation:
Center for Human Nutrition, Department of International Health, Johns Hopkins Bloomberg School of Public Health, 615 N Wolfe Street E2610, Baltimore, MD 21205, USA
Jay S Kaufman
Affiliation:
Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA Carolina Population Center, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
Philip D Sloane
Affiliation:
Department of Family Medicine, School of Medicine, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
Gerardo Heiss
Affiliation:
Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
Joseph Ibrahim
Affiliation:
Department of Biostatistics, School of Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
*
*Corresponding author: Email mbaydoun@jhsph.edu
Rights & Permissions [Opens in a new window]

Abstract

Objective

Recent research indicates that n–3 fatty acids can inhibit cognitive decline, perhaps differentially by hypertensive status.

Design

We tested these hypotheses in a prospective cohort study (the Atherosclerosis Risk in Communities). Dietary assessment using a food-frequency questionnaire and plasma fatty acid exposure by gas chromatography were completed in 1987–1989 (visit 1), while cognitive assessment with three screening tools – the Delayed Word Recall Test, the Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised and the Word Fluency Test (WFT) – was completed in 1990–1992 (visit 2) and 1996–1998 (visit 4). Regression calibration and simulation extrapolation were used to control for measurement error in dietary exposures.

Setting

Four US communities – Forsyth County (North Carolina), Jackson (Mississippi), suburbs of Minneapolis (Minnesota) and Washington County (Maryland).

Subjects

Men and women aged 50–65 years at visit 1 with complete dietary data (n = 7814); white men and women in same age group in the Minnesota field centre with complete plasma fatty acid data (n = 2251).

Results

Findings indicated that an increase of one standard deviation in dietary long-chain n–3 fatty acids (% of energy intake) and balancing long-chain n−3/n–6 decreased the risk of 6-year cognitive decline in verbal fluency with an odds ratio (95% confidence interval) of 0.79 (0.66–0.95) and 0.81 (0.68–0.96), respectively, among hypertensives. An interaction with hypertensive status was found for dietary long-chain n–3 fatty acids (g day−1) and WFT decline (likelihood ratio test, P = 0.06). This exposure in plasma cholesteryl esters was also protective against WFT decline, particularly among hypertensives (OR = 0.51, P < 0.05).

Conclusion

One implication from our study is that diets rich in fatty acids of marine origin should be considered for middle-aged hypertensive subjects. To this end, randomised clinical trials are needed.

Type
Research Paper
Copyright
Copyright © The Authors 2007

On the basis of recent United Nations estimates, the proportion of the US population aged 65 years and over was 12.3% in 2000 and is projected to increase rapidly in the coming decades to reach 20% by 20501. As populations age, all cognitive disorders, including dementia, become more common. Recent research indicates that n−3 fatty acids may be important in preventing cognitive decline. So far, epidemiological evidence, although inconclusive, suggests a protective effect of n–3 fatty acid intake in the dietReference Morris, Evans, Bienias, Tangney, Bennett and Wilson2Reference Kalmijn, Feskens, Launer and Kromhout4. Essential fatty acids are linked to several biochemical and biophysical functions, including structural integrity and fluidity of membranes, enzyme activities, lipid–protein interactions and serving as precursors for eicosanoids such as prostaglandins, leukotrienes and thromboxanesReference Youdim, Martin and Joseph5. The fatty acid composition of neuronal cell-membrane phospholipids reflects the intake of fatty acids in the dietReference Haag6 and fish oils, which contain high levels of C20 and C22 polyunsaturated fatty acids (PUFAs), exert the most profound influence on brain PUFA concentration. According to experimental animal studies, there is a plausible pathway by which hypertension and low dietary n–3 fatty acid intake may interact in increasing the risk of cognitive decline. In fact, hypertensive rats tended to have lower brain concentrations of monounsaturated fatty acids and PUFAs than normotensive ratsReference de Wilde, Hogyes, Kiliaan, Farkas, Luiten and Farkas7, possibly due to pressure-induced endothelial dysfunction at the blood–brain barrier or exhausted astrocytic metabolism. Oxidative stress which accompanies high blood pressure leads to increased peroxidation of unsaturated fatty acids and a reduction in their concentration in the brain represents an alternative explanation.

Despite animal experimental evidence for a possible biological interaction between dietary intake of n–3 fatty acids and hypertensive statusReference de Wilde, Hogyes, Kiliaan, Farkas, Luiten and Farkas7Reference Engler, Engler, Pierson, Molteni and Molteni10 in affecting cognitive decline, no epidemiological study has attempted to test this hypothesis to date. The present observational prospective study assessed the effect of low n–3 fatty acid status on 6-year cognitive decline in men and women aged 50 years and older. A secondary objective was to explore whether hypertensive subjects would benefit to a larger extent than normotensive subjects from this increased intake.

Data and methods

Study sample

The Atherosclerosis Risk in Communities (ARIC) study is an ongoing prospective cohort study aimed at investigating the aetiology of atherosclerosis and its clinical sequelae, and the longitudinal impact of variation in cardiovascular risk factors, medical care and disease, by race, sex, place and time. In each of four US communities – Forsyth County (North Carolina), Jackson (Mississippi), suburbs of Minneapolis (Minnesota) and Washington County (Maryland) – 4000 adults aged 45–64 years were examined four times, 3 years apart (visits 1 to 4). Three out of the four cohorts represented the ethnic mix of their communities, while at Jackson (Mississippi) only African American residents were recruited11. Out of the total sample examined at baseline (N = 15 792) we restricted these analyses to 11 557 individuals aged 50 years or older at baseline since research clearly shows that risk of cognitive decline in general, and of dementia in particular, is negligible prior to the age of 60 years (which is the age at which the youngest individuals in this cohort were re-examined at visit 4)Reference Abate, Ferrari-Ramondo and Di Iorio12. Eligibility for these analyses further required survival to visit 4 (n = 8346), complete data on cognitive functioning at visits 2 (1990–1992) and 4 (1996–1998) (n = 8012) and complete dietary intake at visit 1 (1987–1989), which yielded 7814 men and women. Of these, plasma fatty acid data at visit 1 were available for a subset of the Minneapolis (Minnesota) cohort (n = 2251).

Outcome assessment

Three measures of cognitive functioning were made only at visits 2 and 4 among the total ARIC cohort, and these measures relied on the following instruments: the Delayed Word Recall Test (DWRT)Reference Knopman and Ryberg13, the Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised (DSST/WAIS-R)Reference Wechsler14, and the Word Fluency Test (WFT) of the Multilingual Aphasia Examination, also known as the controlled oral word associationReference Lezak15.

DWRT. This screening tool assesses verbal learning and recent memory. It requires the respondent to recall 10 common words after a 5-min interval during which another test is administered. Test scores may range between 0 and 10 words recalled and the time limit for recall is set at 60 s. The 6-month test–retest reliability of DWRT was previously shown to be high among 26 normal elderly individuals (Pearson correlation coefficient, r = 0.75)Reference Knopman and Ryberg13.

DSST/WAIS-R. This test is a paper-and-pencil test requiring timed translation of numbers 1–9 to symbols using a key. The test measures psychomotor performance and is relatively unaffected by intellectual ability, memory or learning for most adultsReference Lezak15. It appears to be a sensitive and reliable marker of brain damageReference Russell16. The test score can range between 0 and 93 and reflects the correctly translated number of digit–symbol pairs within a time limit of 90 s. Short-term test–retest reliability over 2–5 weeks has been found to be high in individuals aged 45–54 years (r = 0.82)Reference Wechsler14.

WFT. This test requires subjects to record as many words as possible using the initial letters F, A and S, and to list these words, the subject is given only 60 s per letter. The total score corresponds to the total number of words generated during these three trials. The test is particularly sensitive to linguistic impairmentReference Lezak15, Reference Tranel17 and early mental decline in older personsReference Benton, Eslinger and Damasio18. It is also a sensitive marker of damage in the left lateral frontal lobeReference Lezak15, Reference Tranel17. The immediate test–retest correlation coefficient based on an alternative test form has been found to be high (r = 0.82)Reference Franzen19.

Preliminary analysis suggested that while visits 2 and 4 scores of DSST/WAIS-R and WFT had a correlation coefficient close to 0.5, the correlation between DSST/WAIS-R and DWRT was 0.4 and that between DWRT and WFT was about 0.4. However cognitive changes (visit 4–visit 2) in each of these scales had much weaker correlations with each other, ranging between 0.06 and 0.09.

Cut-off points were determined for decline in each of the three cognitive status tests using the Reliable Change Index (RCI) method, in order to correct for measurement error and practice effectsReference Frerichs and Tuokko20. RCI is defined as , where X 1 is the individual’s score at baseline, X 2 the individual’s score at follow-up, M 1 and M 2 are the group mean pre-test and follow-up scores, respectively, and SED is the observed standard error of the difference scores. Scoring below an RCI of −1.645 was regarded as a ‘statistically reliable’ deterioration in the test scores.

A composite measure of the three RCIs to assess global cognitive decline was created using principal components analysis, a data-reduction techniqueReference Mueller and Kim21. Similarly, the cut-off point of the composite score for statistically reliable global cognitive decline was chosen to be −1.645.

Exposure assessment

Usual dietary intake was estimated from an interviewer-administered 61-item semi-quantitative food-frequency questionnaire (FFQ) previously developed and validated by Willett et al. against multiple food records among a sub-sample of the Nurses’ Health Study cohortReference Willett, Sampson, Stampfer, Rosner, Bain and Witschi22. Dietary intake of essential fatty acids and their elongated and desaturated products was expressed as a % of total energy intake and grouped under four main categories, as suggested by Lands et al.Reference Lands, Libelt, Morris, Kramer, Prewitt and Bowen23, Reference Lands24.

  1. 1. C18 n–3 PUFAs, 18:3 + 18:4n−3 (3P);

  2. 2. C18 n–6 PUFAs, 18:2 + 18:3n−6 (6P);

  3. 3. C20 and C22 n–3 highly unsaturated fatty acids (HUFAs): 20:5 + 22:5 + 22:6n−3 (3 H); and

  4. 4. C20 and C22 n–6 HUFAs: 20:3 + 20:4 + 22:4 + 22:5n−6 (6 H).

Sums of fatty acid intake as % of energy included 3 = 3P + 3 H and 6 = 6P + 6 H. Ratios of interest included 3P/6P, 3H/6H and (3P + 3H)/(6P + 6H), also denoted 3/6. In multivariate models, all these variables were standardised by subtracting each observation from the variable mean and dividing the difference by the standard deviation (SD). Hence, the main exposures of interest were 3P, 3H, 3 (as % of energy intake), 3P/6P, 3H/6H, 3/6 and total 3H (in g day−1). Adjustment was made for the other fatty acid variables, and total energy intake was considered as a potential confounder to emulate a multivariate nutrient density modelReference Willett25.

Twelve-hour fasting blood was collected according to the ARIC study-wide protocol. The Minneapolis field centre conducted fatty acid analysis of plasma phospholipid and cholesteryl ester fractions on visit 1 blood specimens. The procedure is described in detail elsewhereReference Shahar, Boland, Folsom, Tockman, McGovern and Eckfeldt26. The identity of 28 fatty acid peaks was revealed by gas chromatography by comparing each peak’s retention time with the retention times of fatty acids in synthetic standards of known composition. The relative amount of each fatty acid (as % of all fatty acids) was calculated by integrating the area under the peak, dividing the result by the total area for all fatty acids, and multiplying by 100. Data from the chromatogram were transferred electronically to a computer for analysis. Plasma exposures are expressed as % of total fatty acids in each fraction and were grouped similarly to dietary exposure. Test–retest reliability coefficients (individuals sampled three times, 2 weeks apart) for various plasma fatty acids ranged from 0.50–0.93 for cholesteryl esters to 0.89 for phospholipidsReference Ma, Folsom, Eckfeldt, Lewis and Chambless27. However, only 3H and the ratio 3H/6H were considered in these analyses.

Replicate dietary measures

Dietary intake was assessed among the surviving ARIC sample at visit 3 (1992–1994), using the same semi-quantitative FFQ that was administered at baseline. At visit 2, a sub-sample of ARIC (around 10% of the original sample) was asked to repeat the FFQ. As stated earlier, of our eligible subset with baseline data on exposure and complete outcome data (n = 7814), 657 had data on visit 2 exposure, 7482 had complete data at visit 3, while 634 had both.

Covariates

Most covariates considered as potential confounders were measured at visit 1, although some were defined according to criteria that spanned all four visits. Covariates can be subdivided into sociodemographic, genetic, health behaviours and nutritional. Age, gender, ethnicity and education were all reported by the respondent. Apolipoprotein E (ApoE) genotype was categorised as 0 to indicate the absence of an ε4 allele vs. 1 to denote carrier status for at least one ε4 allele. Among the behavioural factors (all measured at visit 1), smoking was represented on a three-level categorical scale, i.e. never smoked, smoked previously and current smoker. FFQ-derived values of alcohol (g day−1) and caffeine (mg day−1) were considered as well. Physical activity was assessed using a questionnaire developed by Baecke et al. that included 16 items about usual exertionReference Baecke, Burema and Frijters28. A validated index of physical activity was derived at visit 1, summing sports, work and leisure indices which ranged from a score of 1 (low) to 5 (high)Reference Richardson, Ainsworth, Wu, Jacobs and Leon29. Body mass index at visit 1 was computed by dividing weight (kg) by the square of height (m2). Baseline dietary intake of antioxidants and other micronutrients (mainly vitamins B6, B12 and folate) was also consideredReference Willett, Sampson, Stampfer, Rosner, Bain and Witschi22. The association of these covariates with our outcome has been documented previously by similar cohort studies based on ARIC dataReference Cerhan, Folsom, Mortimer, Shahar, Knopman and McGovern30Reference Blair, Folsom, Knopman, Bray, Mosley and Boerwinkle33.

Our main effect modifier, hypertension, was operationalised using measured systolic and diastolic blood pressure at each visit as well as use of antihypertensive medication over the past two weeks. Seated blood pressure levels were calculated as the average of the second and third of three consecutive measurements with a random-zero sphygmomanometer. Hypertension was defined as systolic blood pressure ≥140 mmHg and diastolic blood pressure ≥90 mmHg or the use of antihypertensive medication during the two weeks prior to examination on any of visits 1 to 4.

Statistical analysis

We carried out univariate analyses of predictor and outcome variables as well as covariates. For bivariate analyses of exposure and outcome, we computed means of predictor variables across outcome groups (0 = no decline; 1 = decline) and assessed the statistical significance of differences using the independent-samples t-test at an α level of 0.05. We computed odds ratios of decline with increase in each exposure by 1SD through a multivariate logistic regression analysis. Control for confounding was accomplished using backward elimination and an overall change in estimate criterion of 5%. Covariates which changed the estimated effect of the exposure by more than 5% were retained in the final modelReference Maldonado and Greenland34.

Hypertension was considered as a potential effect modifier. Likelihood ratio tests were used to assess statistical significance of the interaction between exposure and hypertensive status at a type I error level of 0.20, after obtaining the final parsimonious modelReference Selvin35, Reference Hernan, Hernandez-Diaz, Werler and Mitchell36. The multivariate models can be summarised by equations

(1)
(1)

and

(2)

:

(2)

In the above equations, Q 1 is the main exposure of interest as derived from the FFQ, Q 2i are the other fatty acids that might act as confounders, Zj is a vector of potential confounders that are assumed to be perfectly measured (i.e. no error variance associated with them) and H is the potential effect modifier ‘hypertensive status’, also assumed to be perfectly measured. The same process was used with the plasma exposures in cholesteryl esters and phospholipids. To correct for measurement error in dietary exposure, a sensitivity analysis was conducted for models (1) and (2) whereby regression calibration (RCAL) and simulation extrapolation (SIMEX) were applied to the final parsimonious models for each outcome/exposure pairReference Hardin, Schemiediche and Carroll37, Reference Hardin, Schmiediche and Carroll38. In a multivariate setting, both RCAL and SIMEX rely on the method of moments and attempt to estimate the error variance in the error-prone exposure and adjust the exposure–outcome effect using different procedures. In both cases, replicate measures of FFQ measurements at visits 2 and 3 were used for the correction. The two methods are described in more detail in Appendix A. Statistical analyses were conducted using STATA version 9.039.

Results

Characteristics of study subjects

Table 1 shows the characteristics of study subjects according to availability of dietary and plasma fatty acid data as well as cognitive assessment data at both points in time. Subjects in the plasma fatty acid group (‘plasma group’ hereafter) consisted of whites residing in the suburbs of Minneapolis. They were in general more educated than the dietary group, which was a mix from all ARIC centres. They had a lower proportion of women (50.7% vs. 54.6%), a lower prevalence of the ApoE ε4 allele (28.8% vs. 30.0%), a higher proportion of ‘ever smoked’ status (59.6% vs. 55.5%), and greater consumption of alcohol and caffeine. Some differences were noted for other behavioural and nutritional factors as well. Hypertensive status was particularly high in the dietary group (56.0%) compared with the plasma group (49.3%). Raw mean scores of baseline cognitive function were greater among those in the plasma than in the dietary group and average declines between visits 2 and 4 were found to be steeper in the former. Table 2 summarises the distribution of dietary and plasma fatty acid exposures considered. The SDs for Q1, M and N are of particular importance in interpreting multivariate logistic regression analysis. Q2 and Q3 represent replicate measurements on Q1 at visits 2 and 3 which were used to correct for measurement error in the exposure.

Table 1 Characteristics of study subjects with complete cognitive and dietary data between visits 1 and 4 (dietary group; N = 7814) and those with complete cognitive and plasma data (plasma group; N = 2251); ARIC, 1987–1998

ARIC – Atherosclerosis Risk in Communities study; ApoE – apolipoprotein E; DWRT – Delayed Word Recall Test; DSST/WAIS-R – Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised; WFT – Word Fluency Test.

Data are mean (standard deviation) or %.

* P < 0.05 for null hypothesis that means or proportions are equal between plasma and non-plasma groups.

† Covariate measured at visit 1.

‡ Covariate measured at visits 1 to 4: period prevalence over 9 years.

§ Covariate with other time frame.

Table 2 Distribution of fatty acid groups and ratios for Q1, M, N, Q2 and Q3: mean (SD); ARIC, 1987–1995

SD – standard deviation; ARIC – Atherosclerosis Risk in Communities study; Q1 – food-frequency questionnaire measurement at visit 1 of fatty acid group intake as % of energy intake or ratio of n–3 to n–6 groups; M – biomarker of fatty acid intake in cholesteryl ester fraction of plasma; N – biomarker of fatty acid intake in phospholipid fraction of plasma; Q2 – repeat of Q1 measured at visit 2 among a subset of the cohort; Q3 – repeat of Q1 measured at visit 3 among the surviving baseline cohort; fatty acid exposure categories defined under ‘Exposure assessment’ in Data and methods section.

Multivariate analysis findings: dietary exposures

Multivariate logistic regression models of the relationship between dietary exposure and cognitive decline are presented in Table 3. Results indicate that risk of clinically significant decline in DWRT over the period of 6 years was reduced modestly with every 1SD increase in long-chain n–3 fatty acid intake (3H) as % of total energy intake. This was observed in the total population and among the hypertensive subgroup. For DSST/WAIS-R, although the ratio 3H/6H was protective against decline, this effect did not reach statistical significance. However, the likelihood ratio test indicated a significant level of interaction between this exposure and hypertensive status, shown by the variation in effect across strata of the effect modifier (1.09 among normotensives vs. 0.88 among hypertensives). Risk of decline in WFT was reduced by long-chain and all types of n–3 fatty acid intake (3H and 3) as % of total energy intake, by the ratio 3H/6H and by 3H in g day−1. This relationship was stronger among hypertensive subjects and a significant interaction was noted for 3H in g day−1 (likelihood ratio test, P = 0.06). No statistically significant results were observed for global cognitive decline or other dietary exposures. After adjusting for measurement error in the dietary covariate using RCAL, loss of precision in measures of effect was observed. In most cases, bias seemed to be towards the null when comparing naive and calibrated estimates. In a few instances significance of odds ratios was preserved, such as in the case of 3H, 3H (g day−1), 3 and 3H/6H with WFT decline as the outcome among the hypertensive stratum (data not shown). In these four exposures, the calibrated odds ratio (95% confidence interval) for WFT decline in the hypertensive stratum was 0.68 (0.49–0.95), 0.69 (0.47–1.00), 0.65 (0.43–1.00) and 0.70 (0.52–0.95), respectively. Using SIMEX for selected associations, similar results were obtained. Figure 1 shows two examples of stratified models, models 3.2b and 3.2g of Table 3 (effect of 3H and 3H/6H ratio on decline in WFT). The figure shows the extent to which naive estimates are biased towards the null when compared with the corrected regression coefficients using the SIMEX method. This method is described in detail in Appendix A.

Table 3 Multivariate logistic models of cognitive decline and dietary n–3 fatty acid exposures†‡: naive and regression calibrated OR (95% CI); ARIC, 1987–1998

OR – odds ratio; CI – confidence interval; ARIC – Atherosclerosis Risk in Communities study; RCI – Reliable Change Index; DWRT – Delayed Word Recall Test; DSST/WAIS-R – Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised; WFT – Word Fluency Test; GCD – global cognitive decline; Q1 – food-frequency questionnaire measurement at visit 1 of fatty acid group intake as % of energy intake or ratio of n–3 to n–6 groups; fatty acid exposure categories defined under ‘Exposure assessment’ in Data and methods section.

* P < 0.05 for testing the null hypothesis that β 1 = 0. See equations (1) and (2).

**P < 0.10 for testing the null hypotheses that γ = 0 using the likelihood ratio test. See equation (2).

† Exposures were standardised by subtracting each observation from its mean and dividing it by its standard deviation. Each model (e.g. 1a) has one exposure/outcome pair.

‡ Control for confounding was done using backward elimination and an overall change in estimate criterion of 5%. Covariates which changed the estimate of exposure by more than 5% were retained in the final model. Covariates considered as potential confounders were: sociodemographics (age, sex, education, race); genetic factors (apolipoprotein E ε4 allele); behavioural factors (smoking, alcohol, caffeine consumption and physical activity) and nutritional factors (body mass index, energy intake, other fatty acids, intake of antioxidants and vitamins B6, B12 and folate). Hypertension was considered as a potential effect modifier in separate models. Multiplicative interaction was tested using the likelihood ratio test at an α level of 0.10.

§ Regression calibrated OR (adjusted for measurement error in dietary fatty acids) with its 95% CI, using replicate dietary fatty acid measures at visits 2 and 3 (Q2 and Q3).

Fig. 1 Simulation extrapolation (SIMEX) plot of corrected coefficients for stratified models 3b and 3e of Table 3; Atherosclerosis Risk in Communities study, 1987–1998. pct3 h is dietary intake of long-chain n−3 fatty acids expressed as % of energy intake (3H); r3hr6h is ratio of long-chain n−3 to long-chain n−6 fatty acids (3H/6H) (for definition of 3H and 6H see ‘Exposure assessment’ in Data and methods section). λ is equivalent to θ = {0.5,1,1.5,2} and is a scale factor used to add error to the covariate and estimate βm=f(θ, βm 59) starting from the naive estimate in which θ = 0.Hence, the naive estimate of the regression coefficient β is the one estimated by generalised linear models without measurement error correction. See Appendix A for more details

Multivariate analysis findings: plasma exposures

Multivariate logistic regression analyses of the plasma fatty acid data (Table 4) indicated generally lower odds of cognitive decline among subjects with a higher concentration of long-chain n–3 fatty acid in their plasma cholesteryl esters and phospholipids, and an elevated ratio of long-chain n−3/n−6 fatty acids. An interaction was noted between WFT for absolute 3H in cholesteryl esters (OR = 0.74 among normotensives vs. 0.51 among hypertensives) without reaching statistical significance at the level of 0.10 (P = 0.25). The same pattern was observed for the ratio of 3H/6H in cholesteryl esters and both exposures in the phospholipid fraction.

Table 4 Multivariate logistic models of cognitive decline and plasma n–3 fatty acid exposures†‡: OR (95% CI) for interaction with hypertensive status; ARIC, 1987–1998

OR – odds ratio; CI – confidence interval; ARIC – Atherosclerosis Risk in Communities study; RCI – Reliable Change Index; DWRT – Delayed Word Recall Test; DSST/WAIS-R – Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised; WFT – Word Fluency Test; GCD – global cognitive decline; fatty acid exposure categories defined under ‘Exposure assessment’ in Data and methods section.

* P < 0.05 for testing the null hypothesis that β 1 = 0. See equations (1) and (2).

† Exposures were standardised by subtracting each observation from its mean and dividing it by its standard deviation (SD). Each model (e.g. 1.1a) has one exposure/outcome pair. These models are then stratified by hypertensive status for each of the two exposures.

‡ Same analytic approach was used as in Table 3.

§ Plasma cholesteryl ester levels of fatty acids (%); plasma phospholipids levels of fatty acids (%); M and N in reference to SD values from Table 2.

Discussion

This population-based prospective study conducted among middle-aged men and women at baseline showed that increased dietary intake of long-chain n–3 fatty acids and balancing long-chain n–3/n–6 decreased the risk of cognitive decline in verbal fluency, particularly among hypertensive subjects. This finding also held for the corresponding plasma analytes in the cholesteryl ester and phospholipid fractions. This finding may be due to the higher sensitivity of WFT to early mental decline (i.e. among those aged 55 years or more at baseline) compared with DWRT and DSST/WAIS-RReference Benton, Eslinger and Damasio18. Limitations of the study include the lack of psychometric diagnosis for mild cognitive impairment, which might have been a more definite and clinically relevant outcomeReference Petersen, Smith, Waring, Ivnik, Tangalos and Kokmen40. However, the neuropsychological tests used represent some of the domains reported to be most sensitive in discriminating between normal ageing and mild cognitive impairmentReference Ritchie, Artero and Touchon41. Another limitation relates to the nature of dietary exposures in general, which are often prone to measurement error both in terms of validity and short-term reliability. We corrected for validity by using replicate FFQ measures but failed to correct for short-term reliability of the FFQ due to the lack of adequate short-term replicates in ARIC. Nevertheless, a previous study by Ma et al.Reference Ma, Folsom, Eckfeldt, Lewis and Chambless27 provides estimates of short-term reliability for each of the fatty acids that were considered. Moreover, residual confounding due to inadequate control or measurement of potential confounders cannot be totally ruled out. It is important to note also that exposure timing (1 year prior to visit 1) did not coincide with the baseline measurement of outcome (visit 2), which constitutes another major limitation.

One of the main strengths of this study is its prospective design which, as stated earlier, thus far is unique in the literature testing this particular hypothesisReference Solfrizzi, D’Introno, Colacicco, Capurso, Del Parigi and Capurso42. An evidence-based report suggested a need to look for the effect of n–3 fatty acids on cognitive decline by cardiovascular disease status and to define exposure in terms of absolute value of medium- and long-chain fatty acids, as well as the ratio between n–3 and n−6 fatty acids in diet and plasmaReference Maclean, Issa, Newberry, Mojica, Morton and Garland43. All of these suggestions were implemented in the present study. Moreover, this is the first study to assess effect modification by hypertensive status and to test at the population level a biological interaction documented in animal experimental work. Measurement error, which almost always accompanies dietary assessment, was corrected for in this study using RCAL for all associations and SIMEX for a selected number of these associations.

Previous epidemiological studies have shown that the fatty acids composition of plasma differs significantly between subjects with normal cognitive functioning and patients with some form of cognitive impairmentReference Conquer, Tierney, Zecevic, Bettger and Fisher44Reference Tully, Roche, Doyle, Fallon, Bruce and Lawlor46. While the majority showed a beneficial effect of plasma and erythrocyte n–3 fatty acids on cognition, a case–control study – the Canadian Study of Health and Ageing – reported that the mean relative plasma concentration of n–3 fatty acids as well as total PUFAs was higher among subjects aged 65 years or more with cognitive impairment or dementia after controlling for demographic, behavioural and genetic factorsReference Laurin, Verreault, Lindsay, Dewailly and Holub47. Epidemiological studies based on dietary assessments of n–3 fatty acids also had suggestive but somewhat controversial results. While most leaned towards a protective effect of increasing intake of these fatty acids in the dietReference Morris, Evans, Bienias, Tangney, Bennett and Wilson2, Reference Kalmijn, van Boxtel, Ocke, Verschuren, Kromhout and Launer3, Reference Kalmijn, Launer, Ott, Witteman, Hofman and Breteler48Reference Morris, Evans, Tangney, Bienias and Wilson50, others found no effect or the opposite effect on cognitive functioning and declineReference Kalmijn, Feskens, Launer and Kromhout4, Reference Engelhart, Geerlings, Ruitenberg, Van Swieten, Hofman and Witteman51.

One possibly important implication from our study’s results is that diets rich in fatty acids of marine origin should be considered for middle-aged subjects. We explored whether such an association was differential according to hypertensive status. Although no statistically significant interactions were found, the results suggest that hypertensive subjects (e.g. odds ratios for 3H and WFT were <1 with P < 0.05) may benefit from supplementation of their diets to a larger extent than the normotensive group. These results merit replication given the large public health potential that would be associated with results that unequivocally indicate an inverse association between fatty acid intake and reduced cognitive decline in the general population. The literature indicates that these fatty acids were frequently found associated with reduced risk of cardiovascular disease, including strokeReference He, Song, Daviglus, Liu, Van Horn and Dyer52 and coronary heart diseaseReference Whelton, He, Whelton and Muntner53, Reference He, Song, Daviglus, Liu, Van Horn and Dyer54, although thus far all the evidence is of an observational nature. They have also been associated with improved insulin sensitivityReference Feskens, Virtanen, Rasanen, Tuomilehto, Stengard and Pekkanen55, reduced risk of dyslipidaemiaReference Harris, Windsor and Dujovne56 and a hypocoagulable profileReference Shahar, Folsom, Wu, Dennis, Shimakawa and Conlan57 among other health benefits. Because many of these conditions are also related to cognitive impairment, future research should focus on disentangling the direct and indirect effects of fatty acids (using plasma biomarkers) on cognition and uncover the main mechanism involved in their ability to prevent clinically significant decline in ageing populations. Finally, these findings suggest the utility of randomised clinical trials that would augment intake of marine fatty acids in the treatment group and give a non-enriched diet to the placebo group while allowing for stratification by baseline hypertensive status.

Acknowledgements

Sources of funding:ARIC is carried out as a collaborative study supported by the National Heart, Lung, and Blood Institute, contracts N01-HC-55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021 and N01-HC-55022.

Conflict of interest declaration:None.

Authorship responsibilities: M.A.B. contributed to conceptualisation, literature review, statistical analysis and data management, write-up of the manuscript. J.S.K. contributed to statistical and epidemiological methods, literature review, write-up of parts of the manuscript, final revision of the manuscript. P.D.S. made substantive contribution in the area of cognitive decline and conceptualisation of outcome variables, literature review, revision of the final manuscript. G.H. contributed to dataset/variable acquisition, operationalisation and management, write-up of parts of the manuscript, revision of the final manuscript. J.I. contributed to write-up on parts on measurement error adjustment methods, revision of the final manuscript.

Acknowledgements:The authors thank all ARIC staff and participants for their important contributions; Aaron Folsom (University of Minnesota, Twin Cities, Department of Epidemiology and ARIC principal investigator) for making available to us the plasma fatty acid data for the Minneapolis baseline population of ARIC; William EM Lands for his assistance in clarifying concepts related to his previously published empirical equations and the classification of fatty acid groups; and Eliseo Guallar and Woody Chambless for giving us primary and statistical reviews on the manuscript through the ARIC publication committee.

Appendix A – Measurement error correction

We considered a structural model in which i stands for individual and j for dietary variable and Tij (a latent variable) is the true value of dietary intake of nutrient or ratio of nutrients j for subject i. For that subject i, Qij is the value of dietary variable j derived from the FFQ. Q 1ij and their replicate values from visits 2 and 3, Q 2ij and Q 3ij, were entered into the model as Z-scored manifest variables.

(A1)

An estimate of the error variance covariance matrix was computed using a formula suggested previouslyReference Selvin35, which takes into account the intra-class correlation between replicate measures. For RCAL and SIMEX, the method of moments is used to correct for measurement error in covariates and can be summarised as follows39, Reference Carroll, Ruppert and Stefanski58:

(A2)

The generalised linear model in which measurement error correction of covariates is conducted can be written as:

(A3)

SIMEX is a procedure consisting of four main steps.

Step 1. Fitting the causal model to obtain the estimated coefficients β naive and an estimate of the measurement error variance .

Step 2. Generating random pseudo errors for a scale factor θ times the estimated error variance, . These pseudo errors are added to the original values of the error-prone covariate. Fitting the model to obtain β{naive, θj}. This is repeated r times to obtain the mean coefficient vector, .

Step 3. The previous step is repeated for scale factors, where typically we use , although individual researchers may choose a longer list of scale factors. Using the typical list of scale factors we have k = 5 estimated coefficient vectors, since k* = 4 for the list above and we have the estimated coefficient vector from the initial step (k = k* + 1).

Step 4. For each regression coefficient βm () in the model, we consider the estimated coefficient as a function of the scale factor θj for . Formally, we specify a function f( ) such that . We estimate this relationship and then extrapolate back the final estimates (no measurement error). Researchers are free to choose the form of the function f( ), but we point out that there are relatively few – in this case, five – observations available to estimate the parameters of f( ). The function f( ) used to model the relationship between the estimated coefficient and θ is called the extrapolant functionReference Holcomb59. Although deciding which model to fit is a valid question when performing SIMEX, it has been shown that conservative estimates with a quadratic curve do improve over the naive estimator without any correction. Investigators may also use model-fitting techniques to decide which model to fit and then extrapolate with. Calculating the standard error of the SIMEX estimator requires 100 simulations on its own. With the ever-increasing speed of computers, the necessary computing power is widely availableReference Hardin, Schemiediche and Carroll37.

Appendix B – Unadjusted mean (SD) of fatty acid exposures by cognitive decline status in three domains (DWRT, DSST/WAIS-R and WFT) and GCD, stratified by hypertensive status; ARIC, 1987–1998

SD – standard deviation; RCI – Reliable Change Index; DWRT – Delayed Word Recall Test; DSST/WAIS-R – Digit Symbol Substitution Test of the Wechsler Adult Intelligence Scale–Revised; WFT – Word Fluency Test; GCD – global cognitive decline; ARIC – Atherosclerosis Risk in Communities study; fatty acid exposure categories defined under ‘Exposure assessment’ in Data and methods section.

*P < 0.05 for null hypothesis that means of exposures are equal to each other between cognitive decline categories or hypertensive status.

† ‘Hypertensive’: screened positive on measured hypertension at either visits 1 to 4 or was taking antihypertensive medication in the two weeks prior to examination at any of the four visits.

‡ Plasma cholesteryl ester levels of fatty acids (%); plasma phospholipids levels of fatty acids (%).

Appendix C – Correlations and mean differences between covariates and main fatty acid exposures among all subjects in the dietary group (N = 7814) and the plasma group (N = 2251); ARIC, 1987–1998

ARIC – Atherosclerosis Risk in Communities study; SD – standard deviation; HS – high school; ApoE – apolipoprotein E; fatty acid exposure categories defined under ‘Exposure assessment’ in Data and methods section.

* P < 0.05 for hypothesis test of no difference between categories (in categorical covariates) and Pearson correlation coefficient r = 0 for continuous covariates.

References

1United Nations (UN) . World Population Prospects: The 2002 Revision. New York: UN, 2002.Google Scholar
2Morris, MC, Evans, DA, Bienias, JL, Tangney, CC, Bennett, DA, Wilson, RS, et al. . Consumption of fish and n–3 fatty acids and risk of incident Alzheimer disease. Archives of Neurology 2003; 60 (7): 940946.CrossRefGoogle ScholarPubMed
3Kalmijn, S, van Boxtel, MP, Ocke, M, Verschuren, WM, Kromhout, D, Launer, LJ. Dietary intake of fatty acids and fish in relation to cognitive performance at middle age. Neurology 2004; 62 (2): 275280.CrossRefGoogle ScholarPubMed
4Kalmijn, S, Feskens, EJ, Launer, LJ, Kromhout, D. Polyunsaturated fatty acids, antioxidants, and cognitive function in very old men. American Journal of Epidemiology 1997; 145 (1): 3341.CrossRefGoogle ScholarPubMed
5Youdim, KA, Martin, A, Joseph, JA. Essential fatty acids and the brain: possible health implications. International Journal of Developmental Neuroscience 2000; 18 (4–5): 383399.CrossRefGoogle ScholarPubMed
6Haag, M . Essential fatty acids and the brain. Canadian Journal of Psychiatry 2003; 48 (3): 195203.CrossRefGoogle ScholarPubMed
7de Wilde, MC, Hogyes, E, Kiliaan, AJ, Farkas, T, Luiten, PG, Farkas, E. Dietary fatty acids alter blood pressure, behavior and brain membrane composition of hypertensive rats. Brain Research 2003; 988 (1–2): 919.CrossRefGoogle ScholarPubMed
8Frenoux, JM, Prost, ED, Belleville, JL, Prost, JL. A polyunsaturated fatty acid diet lowers blood pressure and improves antioxidant status in spontaneously hypertensive rats. Journal of Nutrition 2001; 131 (1): 3945.CrossRefGoogle ScholarPubMed
9Bellenger-Germain, S, Poisson, JP, Narce, M. Antihypertensive effects of a dietary unsaturated FA mixture in spontaneously hypertensive rats. Lipids 2002; 37 (6): 561567.CrossRefGoogle ScholarPubMed
10Engler, MM, Engler, MB, Pierson, DM, Molteni, LB, Molteni, A. Effects of docosahexaenoic acid on vascular pathology and reactivity in hypertension. Experimental Biology and Medicine 2003; 228 (3): 299307.CrossRefGoogle ScholarPubMed
11The ARIC investigators . The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. American Journal of Epidemiology 1989; 129 (4): 687702.CrossRefGoogle Scholar
12Abate, C, Ferrari-Ramondo, V, Di Iorio, A. Risk factors for cognitive disorders in the elderly: a review. Archives of Gerontology and Geriatrics 1998; (Suppl. 6): 715.CrossRefGoogle Scholar
13Knopman, DS, Ryberg, S. A verbal memory test with high predictive accuracy for dementia of the Alzheimer type. Archives of Neurology 1989; 46 (2): 141145.CrossRefGoogle ScholarPubMed
14Wechsler, D . WAIS-R Manual. Cleveland, OH: The Psychological Corporation, 1981.Google Scholar
15Lezak, MD . Neuropsychological Assessment, 2nd ed. New York: Oxford University Press, 1983.Google Scholar
16Russell, EW . WAIS factor analysis with brain-damaged subjects using criterion measures. Journal of Consulting and Clinical Psychology 1972; 39 (1): 133139.CrossRefGoogle ScholarPubMed
17Tranel, D . Neuropsychological assessment. Psychiatric Clinics of North America 1992; 15 (2): 283299.CrossRefGoogle ScholarPubMed
18Benton, AL, Eslinger, PJ, Damasio, AR. Normative observations on neuropsychological test performances in old age. Journal of Clinical Neuropsychology 1981; 3 (1): 3342.CrossRefGoogle ScholarPubMed
19Franzen, MD , ed. Mutlilingual Aphasia Examination. Kansas City, MO: Test Corporation of America, 1986.Google Scholar
20Frerichs, RJ, Tuokko, HA. Reliable change scores and their relation to perceived change in memory: implications for the diagnosis of mild cognitive impairment. Archives of Clinical Neuropsychology 2006; 21 (1): 109115.CrossRefGoogle ScholarPubMed
21Mueller, CW, Kim, JO. Factor Analysis: Statistical Methods and Practical Issues. London: Sage Publications, 1978.Google Scholar
22Willett, WC, Sampson, L, Stampfer, MJ, Rosner, B, Bain, C, Witschi, J, et al. . Reproducibility and validity of a semiquantitative food frequency questionnaire. American Journal of Epidemiology 1985; 122 (1): 5165.CrossRefGoogle ScholarPubMed
23Lands, WE, Libelt, B, Morris, A, Kramer, NC, Prewitt, TE, Bowen, P, et al. . Maintenance of lower proportions of (n−6) eicosanoid precursors in phospholipids of human plasma in response to added dietary (n−3) fatty acids. Biochimica et Biophysica Acta 1992; 1180 (2): 147162.CrossRefGoogle ScholarPubMed
24Lands, WE . Long-term fat intake and biomarkers. American Journal of Clinical Nutrition 1995; 61 (3 Suppl.): 721S725S.CrossRefGoogle ScholarPubMed
25Willett, WC . Nutritional Epidemiology. New York: Oxford University Press, 1990.Google Scholar
26Shahar, E, Boland, LL, Folsom, AR, Tockman, MS, McGovern, PG, Eckfeldt, JH. Docosahexaenoic acid and smoking-related chronic obstructive pulmonary disease. The Atherosclerosis Risk in Communities Study Investigators. American Journal of Respiratory and Critical Care Medicine 1999; 159 (6): 17801785.CrossRefGoogle ScholarPubMed
27Ma, J, Folsom, AR, Eckfeldt, JH, Lewis, L, Chambless, LE. Short- and long-term repeatability of fatty acid composition of human plasma phospholipids and cholesterol esters. The Atherosclerosis Risk in Communities (ARIC) Study Investigators. American Journal of Clinical Nutrition 1995; 62 (3): 572578.CrossRefGoogle Scholar
28Baecke, JA, Burema, J, Frijters, JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. American Journal of Clinical Nutrition 1982; 36 (5): 936942.CrossRefGoogle Scholar
29Richardson, MT, Ainsworth, BE, Wu, HC, JrJacobs, DR, Leon, AS. Ability of the Atherosclerosis Risk in Communities (ARIC)/Baecke Questionnaire to assess leisure-time physical activity. International Journal of Epidemiology 1995; 24 (4): 685693.CrossRefGoogle ScholarPubMed
30Cerhan, JR, Folsom, AR, Mortimer, JA, Shahar, E, Knopman, DS, McGovern, PG, et al. . Correlates of cognitive function in middle-aged adults. Atherosclerosis Risk in Communities (ARIC) Study Investigators. Gerontology 1998; 44 (2): 95105.CrossRefGoogle ScholarPubMed
31Peacock, JM, Folsom, AR, Knopman, DS, Mosley, TH, JrGoff, DC, Szklo, M. Dietary antioxidant intake and cognitive performance in middle-aged adults. The Atherosclerosis Risk in Communities (ARIC) Study investigators. Public Health Nutrition 2000; 3 (3): 337343.CrossRefGoogle ScholarPubMed
32Knopman, D, Boland, LL, Mosley, T, Howard, G, Liao, D, Szklo, M, et al. . Cardiovascular risk factors and cognitive decline in middle-aged adults. Neurology 2001; 56 (1): 4248.CrossRefGoogle ScholarPubMed
33Blair, CK, Folsom, AR, Knopman, DS, Bray, MS, Mosley, TH, Boerwinkle, E. APOE genotype and cognitive decline in a middle-aged cohort. Neurology 2005; 64 (2): 268276.CrossRefGoogle Scholar
34Maldonado, G, Greenland, S. Simulation study of confounder-selection strategies. American Journal of Epidemiology 1993; 138 (11): 923936.CrossRefGoogle ScholarPubMed
35Selvin, S . Statistical Analysis of Epidemiologic Data, 3rd ed. New York: Oxford University Press, 2004.CrossRefGoogle Scholar
36Hernan, MA, Hernandez-Diaz, S, Werler, MM, Mitchell, AA. Causal knowledge as a prerequisite for confounding evaluation: an application to birth defects epidemiology. American Journal of Epidemiology 2002; 155 (2): 176184.CrossRefGoogle ScholarPubMed
37Hardin, JW, Schemiediche, H, Carroll, RJ. The simulation extrapolation method for fitting generalized linear models with additive measurement error. The STATA Journal 2003; 4: 373385.CrossRefGoogle Scholar
38Hardin, JW, Schmiediche, H, Carroll, RJ. The regression calibration method for fitting generalized linear models with additive measurement error. The STATA Journal 2003; 4: 112.Google Scholar
39Stata Corporation . Statistics/Data Analysis: Release 8.2 [program]. College Station, TX: Stata Corporation, 2002.Google Scholar
40Petersen, RC, Smith, GE, Waring, SC, Ivnik, RJ, Tangalos, EG, Kokmen, E. Mild cognitive impairment: clinical characterization and outcome. Archives of Neurology 1999; 56 (3): 303308.CrossRefGoogle ScholarPubMed
41Ritchie, K, Artero, S, Touchon, J. Classification criteria for mild cognitive impairment: a population-based validation study. Neurology 2001; 56 (1): 3742.CrossRefGoogle ScholarPubMed
42Solfrizzi, V, D’Introno, A, Colacicco, AM, Capurso, C, Del Parigi, A, Capurso, S, et al. . Dietary fatty acids intake: possible role in cognitive decline and dementia. Experimental Gerontology 2005; 40 (4): 257270.CrossRefGoogle ScholarPubMed
43Maclean, CH, Issa, AM, Newberry, SJ, Mojica, WA, Morton, SC, Garland, RH, et al. . Effects of omega-3 fatty acids on cognitive function with aging, dementia, and neurological diseases. Evidence Report/Technology Assessment (Summary) 2005; (114): 13.Google ScholarPubMed
44Conquer, JA, Tierney, MC, Zecevic, J, Bettger, WJ, Fisher, RH. Fatty acid analysis of blood plasma of patients with Alzheimer’s disease, other types of dementia, and cognitive impairment. Lipids 2000; 35 (12): 13051312.CrossRefGoogle ScholarPubMed
45Heude, B, Ducimetiere, P, Berr, C. Cognitive decline and fatty acid composition of erythrocyte membranes – The EVA Study. American Journal of Clinical Nutrition 2003; 77 (4): 803808.CrossRefGoogle ScholarPubMed
46Tully, AM, Roche, HM, Doyle, R, Fallon, C, Bruce, I, Lawlor, B, et al. . Low serum cholesteryl ester-docosahexaenoic acid levels in Alzheimer’s disease: a case–control study. British Journal of Nutrition 2003; 89 (4): 483489.CrossRefGoogle ScholarPubMed
47Laurin, D, Verreault, R, Lindsay, J, Dewailly, E, Holub, BJ. Omega-3 fatty acids and risk of cognitive impairment and dementia. Journal of Alzheimer’s Disease 2003; 5 (4): 315322.CrossRefGoogle ScholarPubMed
48Kalmijn, S, Launer, LJ, Ott, A, Witteman, JC, Hofman, A, Breteler, MM. Dietary fat intake and the risk of incident dementia in the Rotterdam Study. Annals of Neurology 1997; 42 (5): 776782.CrossRefGoogle ScholarPubMed
49Whalley, LJ, Fox, HC, Wahle, KW, Starr, JM, Deary, IJ. Cognitive aging, childhood intelligence, and the use of food supplements: possible involvement of n–3 fatty acids. American Journal of Clinical Nutrition 2004; 80 (6): 16501657.CrossRefGoogle ScholarPubMed
50Morris, MC, Evans, DA, Tangney, CC, Bienias, JL, Wilson, RS. Fish consumption and cognitive decline with age in a large community study. Archives of Neurology 2005; 62 (12): 18491853.CrossRefGoogle Scholar
51Engelhart, MJ, Geerlings, MI, Ruitenberg, A, Van Swieten, JC, Hofman, A, Witteman, JC, et al. . Diet and risk of dementia: does fat matter? The Rotterdam Study. Neurology 2002; 59 (12): 19151921.CrossRefGoogle ScholarPubMed
52He, K, Song, Y, Daviglus, ML, Liu, K, Van Horn, L, Dyer, AR, et al. . Fish consumption and incidence of stroke: a meta-analysis of cohort studies. Stroke 2004; 35 (7): 15381542.CrossRefGoogle ScholarPubMed
53Whelton, SP, He, J, Whelton, PK, Muntner, P. Meta-analysis of observational studies on fish intake and coronary heart disease. American Journal of Cardiology 2004; 93 (9): 11191123.CrossRefGoogle ScholarPubMed
54He, K, Song, Y, Daviglus, ML, Liu, K, Van Horn, L, Dyer, AR, et al. . Accumulated evidence on fish consumption and coronary heart disease mortality: a meta-analysis of cohort studies. Circulation 2004; 109 (22): 27052711.CrossRefGoogle ScholarPubMed
55Feskens, EJ, Virtanen, SM, Rasanen, L, Tuomilehto, J, Stengard, J, Pekkanen, J, et al. . Dietary factors determining diabetes and impaired glucose tolerance. A 20-year follow-up of the Finnish and Dutch cohorts of the Seven Countries Study. Diabetes Care 1995; 18 (8): 11041112.CrossRefGoogle ScholarPubMed
56Harris, WS, Windsor, SL, Dujovne, CA. Effects of four doses of n–3 fatty acids given to hyperlipidemic patients for six months. Journal of the American College of Nutrition 1991; 10 (3): 220227.CrossRefGoogle ScholarPubMed
57Shahar, E, Folsom, AR, Wu, KK, Dennis, BH, Shimakawa, T, Conlan, MG, et al. . Associations of fish intake and dietary n–3 polyunsaturated fatty acids with a hypocoagulable profile. The Atherosclerosis Risk in Communities (ARIC) Study. Arteriosclerosis and Thrombosis 1993; 13 (8): 12051212.CrossRefGoogle ScholarPubMed
58Carroll, RJ, Ruppert, D, Stefanski, LA. Measurement Error in Nonlinear Models. Boca Raton, FL: CRC Press, 1995.CrossRefGoogle Scholar
59JrHolcomb, JP . Regression with covariates and outcome calculated from a common set of variables measured with error: estimation using the SIMEX method. Statistics in Medicine 1999; 18 (21): 28472862.3.0.CO;2-V>CrossRefGoogle ScholarPubMed
Figure 0

Table 1 Characteristics of study subjects with complete cognitive and dietary data between visits 1 and 4 (dietary group; N = 7814) and those with complete cognitive and plasma data (plasma group; N = 2251); ARIC, 1987–1998

Figure 1

Table 2 Distribution of fatty acid groups and ratios for Q1, M, N, Q2 and Q3: mean (SD); ARIC, 1987–1995

Figure 2

Table 3 Multivariate logistic models of cognitive decline and dietary n–3 fatty acid exposures†‡: naive and regression calibrated OR (95% CI); ARIC, 1987–1998

Figure 3

Fig. 1 Simulation extrapolation (SIMEX) plot of corrected coefficients for stratified models 3b and 3e of Table 3; Atherosclerosis Risk in Communities study, 1987–1998. pct3 h is dietary intake of long-chain n−3 fatty acids expressed as % of energy intake (3H); r3hr6h is ratio of long-chain n−3 to long-chain n−6 fatty acids (3H/6H) (for definition of 3H and 6H see ‘Exposure assessment’ in Data and methods section). λ is equivalent to θ = {0.5,1,1.5,2} and is a scale factor used to add error to the covariate and estimate βm=f(θ, βm59) starting from the naive estimate in which θ = 0.Hence, the naive estimate of the regression coefficient β is the one estimated by generalised linear models without measurement error correction. See Appendix A for more details

Figure 4

Table 4 Multivariate logistic models of cognitive decline and plasma n–3 fatty acid exposures†‡: OR (95% CI) for interaction with hypertensive status; ARIC, 1987–1998