Vitamin E and Cognitive Decline in Older Persons FREE

OXIDATIVE MECHANISMS may play important roles in neurodegenerative diseases, such as Alzheimer disease^{1 - 3} and Parkinson disease,⁴ and in cellular processes associated with aging.^{5 - 7} The brain is especially vulnerable to free radical damage because of its high oxygen consumption rate, its abundance of easily peroxidized lipid membranes, and the presence of relatively few antioxidant enzymes.⁶ Oxidative reactions induced by reactive oxygen species are thought to cause the degeneration of neurons. Antioxidant nutrients, including vitamin E, vitamin C, and carotene, counteract these processes by inhibiting lipid peroxidation,^{6 ,8 - 9} the generation of reactive oxygen species,^{10 - 11} apoptosis,^{10 - 12} mitochondrial dysfunction,¹¹ cytotoxic damage to cell membranes,¹³ and oxidative damage to proteins^{14 - 16} and DNA.^{17 - 19}

Oxidative processes and the intake of antioxidant nutrients may also affect the rate of cognitive decline, but epidemiologic investigations^{20 - 21} of this issue have been limited. In a clinical trial²² of patients with Alzheimer disease, vitamin E supplementation delayed progression to institutionalization but had no effect on cognitive decline. Whether dietary intake of antioxidants can prevent cognitive decline earlier in the disease process is not known. In this report, we describe the association between dietary antioxidants and 3-year change in cognitive function in a large biracial population.

Study participants are from the Chicago Health and Aging Project (CHAP), a longitudinal study of a geographically defined community on the south side of Chicago.²³ From September 17, 1993, to May 5, 1997, we conducted a complete census, followed by home interviews with age-eligible residents that included cognitive testing. A total of 6158 residents 65 years and older participated (79% of eligible residents). Follow-up interviews were conducted approximately 3 years later for 4320 participants (87% of the 4983 survivors), from January 8, 1997, to February 22, 2000. In addition, 729 randomly selected persons underwent cognitive testing 6 months after the baseline interview. The Institutional Review Board of Rush-Presbyterian-St Luke's Medical Center approved the study, and all participants gave written consent.

Diet was assessed after baseline using a modified Harvard self-administered food frequency questionnaire (FFQ)²⁴ that measured usual consumption during the past year of 139 food items and vitamin supplements.²⁵ Of the 4276 persons with cognitive data at the 3-year interview, 4226 (99%) completed the dietary questionnaire. We excluded 140 persons with incomplete or potentially invalid dietary data and 8 who were not black or white. Because dietary changes can occur with the onset of illnesses associated with dementia, we eliminated 1189 persons for whom FFQ completion exceeded the a priori defined period of 2½ years after baseline, leaving 2889 participants for analysis. For these participants, the FFQs were completed a mean of 18 (SD, ±8) months after baseline and 21 (SD, ±8) months before the 3-year follow-up. A third cognitive assessment was available for 288 of these participants, of 729 tested at 6 months. (Of the remaining sample, 160 died, 134 had incomplete or invalid dietary data, and 147 exceeded the 2½-year time restriction.) Compared with the analyzed group, unanalyzed participants (36% of whom died before follow-up) were older and less educated, had lower cognitive scores, and had lower intakes of vitamins E and C (Table 1).

Four cognitive tests were administered at each time point: the East Boston Memory Test,^{26 - 27} which involves immediate and delayed recall of a brief story; the oral form of the Symbol Digit Modalities Test,²⁸ a measure of perceptual speed; and the Mini-Mental State Examination.²⁹ Analysis of decline on individual tests is problematic because of floor and ceiling effects and other sources of measurement error.³⁰ In a previous factor analysis of the data, all 4 tests loaded on a single factor in a principal components factor analysis, supporting the idea that a global measure could adequately summarize performance on the 4 tests; a global measure of cognitive function was formed by averaging the z scores of the 4 individual tests.²³ We used this previously established approach in our analyses as well. We computed the z scores at all time points based on the means and SDs of the test scores of the entire population at the baseline interview so that we could examine change in this global measure over time.

Daily intake of each dietary component was computed by multiplying the nutrient content of the food item (from the Harvard nutrient database) by reported frequency of intake, and summing over all food items. All nutrients were calorie adjusted using the regression-residual method.³¹

In a validation study of 232 randomly selected CHAP participants, Pearson product moment correlation coefficients between nutrient intakes measured by the FFQ and the average of repeated 24-hour recalls were as follows: r = 0.67 for total vitamin E (r = 0.41 without supplements) and r = 0.60 for total vitamin C (r = 0.46 without supplements). The correlation between total vitamin E intake from the FFQ and serum levels of α-tocopherol among 56 of these participants was r = 0.63. For all correlations, P<.001.

Sex and race were recorded at the census and verified at the baseline interview. Race questions and categories were those used by the 1990 US census. Other nondietary covariates were obtained at the baseline interview, and included age (computed from self-reported birth date and date of the baseline interview), level of education (years of regular schooling), and current smoking ("Do you smoke cigarettes now?"). Daily alcohol intake (in grams) was computed from separate questions on the CHAP FFQ about usual consumption of beer, wine, and other alcoholic liquor.

We used random-effects models³² in SAS statistical software³³ to examine the association of nutrient intake with change in the global measure of cognitive function, which was approximately normally distributed, at the baseline, 6-month, and 3-year evaluations. The random-effects model allows for within-person variability in initial cognitive score and in slope of change, while simultaneously estimating the linear effects of model covariates on overall level of cognitive score across the 3 time points (covariate term) and on the rate of change in score (covariate-time interaction term). Each random-effects model included terms for time, the main effect term for each covariate, and interaction terms between time and each covariate. Intakes of the antioxidant nutrients were modeled in quintiles, with and without intake from vitamin supplements, using the lowest quintile as the referent category. We used a multivariable model to simultaneously control for factors potentially related to change in cognitive function. Confounding by other dietary components was examined by adding continuous log-transformed terms for these variables to the multivariable model. We examined effect modification in separate models that included main effect terms from the multivariable model; 2-way interaction terms between each pair of time, the effect modifier, and the antioxidant nutrient (continuous); and a 3-way interaction term between these variables. Model assumptions (eg, normality, bivariate normality of the random effects, and homoscedasticity) were evaluated using standard analytical and graphical techniques. P values were 2 sided, and the type I error rate for statistical significance (α) was .05.

Analyzed participants included 1594 black and 1295 white persons, aged 65 to 102 years (mean, 73.9 years), with a mean educational level of 12.5 years (SD, ±3.7 years). The average length of follow-up was 3.2 years (range, 1.8-5.9 years). The median baseline global z score was 39 × 10⁻² standardized units per year (SU/y). Cognitive scores declined, on average, by 5.0 × 10⁻² SU/y; however, many participants (39%) experienced no change or even an improvement in scores.

A higher intake of total vitamin E (from foods and supplements) was associated with less change in cognitive score per year. In the age-adjusted model, scores for persons in the lowest quintile of total vitamin E intake declined by 6.5 × 10⁻² SU/y (Table 2). The rate decrease for the fifth quintile was statistically significant, as was the test for linear trend. The rate decreases in the upper quintiles changed little after adjustment for age, race, sex, educational level, current smoking, alcohol consumption, and total calorie (energy) intake. Further adjustment for intakes of total vitamin A, carotene, and vitamin C had little effect on the results. The rate for persons in the highest vitamin E quintile (−4.3 × 10⁻² SU/y) was lower by 2.4 × 10⁻² SU/y compared with that among persons in the lowest quintile (−6.7 × 10⁻² SU/y), a statistically significant 36% reduction (P = .05). The rate differences for quintiles of total vitamin C intake were smaller than those for vitamin E and were nonsignificant in the age-adjusted and multivariable models (Table 2). There was little or no evidence of an association between carotene intake and cognitive decline. In the multivariable model, the differences in the rates of cognitive change (×10⁻² SU/y) from the lowest to highest quintiles of total carotene intake were as follows: 0.0, 0.3, 1.1, 0.7, and 0.8 (P = .77 for trend).

We observed a linear inverse association with cognitive decline in age-adjusted analyses considering vitamin E intake from food sources only (Table 3). In this model, rates of decline decreased significantly by 2.0 × 10⁻² SU/y in the fourth quintile and by 2.1 × 10⁻² SU/y in the fifth quintile compared with the first. The rate differences did not change appreciably with adjustment for other risk factors in the multivariable model, and were slightly increased with additional adjustment for other antioxidant nutrients. The multivariable model is presented in Figure 1. Persons in the upper quintiles of intake had higher predicted baseline scores than those in the lowest quintile. In addition, the rates of change in score per year were smaller in quintiles 4 and 5 than in quintile 1. Rates of change (×10−²) for quintiles 1 through 5 were as follows: −6.7, −5.1, −6.5, −4.6, and −4.3, respectively (P = .02 for trend). The protective effect remained when we adjusted for use of multivitamin and vitamin E supplements in the multivariable model (P = .04 for trend); however, neither multivitamin use (β = −.9 × 10⁻² SU/y; P = .20) nor vitamin E supplement use overall (β = 1.4 × 10⁻² SU/y; P = .10) was associated with cognitive change. Further investigation revealed that the vitamin E supplement users had significantly less cognitive decline than the nonusers who had low food intake of vitamin E, but no difference in decline from the nonusers with high food intake of vitamin E (P = .04 for the interaction term between supplement use and food intake, modeled continuously in the multivariable model). There was no association between vitamin C intake from food sources and change in cognitive function (Table 3). However, vitamin C supplement use was significantly associated with less cognitive decline, by 1.8 × 10⁻² SU/y (P = .03), in a multivariable model adjusted for quintiles of vitamin C intake from foods and multivitamin use. We found no evidence of interaction between the use of vitamin E and vitamin C supplements (β = .2 × 10⁻² SU/y; P = .91) in the multivariable model, although we had limited power to test this adequately.

As seen in Table 4, the potential confounders were different for vitamin E supplement use and quintiles of intake from foods. Vitamin E supplement users tended to have more years of education and a higher intake of other antioxidant nutrients. The lowest quintile group of vitamin E intake from foods had higher alcohol consumption and more smokers. Black participants had a somewhat higher intake of vitamin E from foods, but were less likely to use either vitamin E supplements or multivitamins.

The association between cognitive change and total vitamin E intake did not change appreciably in separate multivariable models adjusted for intake of either total fat or different types of fat (polyunsaturated, monounsaturated, and saturated). We also explored whether the associations with total intakes of the antioxidant nutrients differed by level of age, sex, race, education, smoking, alcohol intake, and analyzed dietary components. The only evidence of effect modification was a marginally significant lower rate of cognitive decline with high vitamin E intake among smokers compared with nonsmokers (P = .08 in the multivariable model).

When we reanalyzed the data excluding persons with possible cognitive impairment at baseline (Mini-Mental State Examination score, <25), the association with total vitamin E intake was strengthened (a rate difference of 3.6 × 10⁻² SU/y in the fifth quintile compared with the first; P = .003) in the multivariable model, adjusting for other antioxidant nutrients. In a reanalysis without the time restriction on FFQ completion (n = 4072), the vitamin E association was in the protective direction but not statistically significant. The rate differences (×10⁻² SU/y) for quintiles 2 through 5 of vitamin E intake from foods in the multivariable model, adjusted for other antioxidant nutrients, were as follows: 0.5, −0.5, 1.1, and 1.3, respectively (P = .10 for trend).

Vitamin E intake from foods and supplements was associated with reduced cognitive decline in this older biracial population. In contrast, there was no association between cognitive decline and carotene intake, and limited evidence for an association with vitamin C. The effects on cognitive decline in the highest quintiles of vitamin E intake (total or from foods only) were equivalent to a corresponding decrease in age of 8 to 9 years. The median level of vitamin E intake (6.8 IU/d) for the lowest fifth of the CHAP population was higher than previously reported by other US studies.^{34 - 37} This would seem to indicate that increasing vitamin E intake in the population to at least the recommended levels of 18 to 22 IU/d would have important public health implications.

We were able to detect small changes in cognitive function in the population because of the many participants. In addition, the use of a combined measure of 4 cognitive tests, and up to 3 points of cognitive assessment, provided a reliable measure of cognitive change. The multiple points of cognitive assessment allowed for the estimation of within-person changes in cognition, thus reducing potential confounding from between-person differences in cognitive performance. The fewer number of participants with a third cognitive assessment may have resulted in less precise but unbiased estimates of the rates of cognitive change.

The fact that associations were observed for food and supplement sources of intake strengthens the inference that vitamin E is associated with reduced cognitive decline. However, the many participants eliminated from the analyses and the assessment of diet midway between the cognitive assessments make a causal interpretation more tenuous. We excluded participants from the analyses whose dietary assessments were 2½ years after baseline because dietary behaviors can change with onset of conditions and diseases associated with cognitive decline.^{38 - 39} Such bias due to dietary changes may explain the weakened vitamin E effect in the analyses of all participants, although it is also possible that the true effect is smaller than that observed with the exclusion of participants with more cross-sectional assessments of diet. We also observed secular changes in the use of vitamin E supplements as the survey progressed; 9% of the participants who completed the FFQ in 1994 reported taking a vitamin E supplement compared with 19% of those who completed the FFQ in 1997. Recent supplement use could have weakened the observed association if the increase occurred among persons who were experiencing problems in cognition. Further follow-up of the study population will allow for greater explication of the association between cognitive decline and vitamin E intake from supplements.

Much of the available epidemiologic evidence on this issue is from cross-sectional studies. Among 6 cross-sectional studies^{40 - 45} that examined vitamin E in serum or plasma, 5^{40 - 44} found significant protective associations against low cognitive function, as did one dietary study.⁴⁵ A cross-sectional study⁴⁶ of dietary vitamin E found no association with cognitive impairment. The findings from cross-sectional studies of vitamin C have been equivocal, with some reporting protective associations against low cognitive function^{45 - 50} and others reporting no association.^{20 ,40 - 41 ,51 - 52} Inconsistent findings have also been reported for beta carotene, with 5^{42 - 43 ,45 - 46 ,53} of 8 studies^{20 ,40 - 43 ,45 - 46 ,53} showing protective associations.

Few epidemiologic studies have examined the association between antioxidant nutrients and cognitive decline or incident dementia. One study²⁰ observed no association with a 2-point decrease in cognitive score in 53 of 342 men over 3 years. Of 2 other studies that related cognitive score to past levels of antioxidant nutrients in plasma⁴⁵ and from diet,⁴⁸ one observed a protective association with vitamin E⁴⁸ and one with beta carotene and vitamin C.⁴⁵ All of these studies were small and likely had low power to detect nutrient effects on small changes in cognition. Two prospective studies have reported protective associations between antioxidant vitamin supplements and incident²¹ and prevalent⁵⁴ dementia. A study²¹ of 633 East Boston residents with 91 incident cases of Alzheimer disease found that 0 of 38 persons consuming vitamin E or C supplements developed incident disease over 4½ years; as in the present study, there was no association with multivitamin use. Another investigation⁵⁴ reported lower dementia among men who reported use of vitamin E and C supplements in earlier surveys. In a clinically evaluated sample of the CHAP population, we found that vitamin E intake from foods was associated with reduced risk of developing Alzheimer disease, and the protective benefit appeared to occur only among persons who were APOE-ϵ4–negative.⁵⁵ Because APOE genotyping was possible only among the clinically evaluated participants, we could not examine the interaction between APOE-ϵ4 and vitamin E on cognitive decline in the population.This report provides evidence that high intakes of vitamin E from food or supplements are associated with a reduced risk of cognitive decline. The relation is supported by animal studies, in which rodents fed diets high in vitamin E demonstrated improved cognitive functioning during aging, with fewer errors in maze tests,⁶ a greater rate of learning,⁷ and greater memory retention.^{7 ,56} A causal interpretation of the observed findings would require additional evidence by other longitudinal epidemiologic studies and/or primary prevention trials.

Accepted for publication March 6, 2002.

Author contributions: Study concept and design (Drs Morris, Evans, Bienias, and Wilson); acquisition of data (Drs Evans and Tangney); analysis and interpretation of data (Drs Morris, Evans, Bienias, Tangney, and Wilson); drafting of the manuscript (Drs Morris, Evans, and Wilson); critical revision of the manuscript for important intellectual content (Drs Morris, Evans, Bienias, Tangney, and Wilson); statistical expertise (Drs Morris, Bienias, and Tangney); obtained funding (Dr Morris); administrative, technical, and material support (Drs Evans and Wilson); and study supervision (Dr Morris).

This study was supported by grants AG11101 and AG13170 from the National Institute on Aging, Bethesda, Md.

We thank the residents of Morgan Park, Washington Heights, and Beverly, Ill, for their cooperation and support; the coordinators, Cheryl Bibbs, BA, Michelle Bos, BA, and Flavio Lamorticella, BA, and their staffs; and the analytic programmer, Woojeong Bang, MS.

Corresponding author and reprints: Martha Clare Morris, ScD, Rush Institute for Healthy Aging, Rush-Presbyterian-St Luke's Medical Center, 1645 W Jackson, Suite 675, Chicago, IL 60612 (e-mail: mmorris@rush.edu).