Caffeine and Alcohol Intakes and Overall Nutrient Adequacy Are Associated with Longitudinal Cognitive Performance among U.S. Adults^,,

Abstract

Among modifiable lifestyle factors, diet may affect cognitive health. Cross-sectional and longitudinal associations may exist between dietary exposures [e.g., caffeine (mg/d), alcohol (g/d), and nutrient adequacy] and cognitive performance and change over time. This was a prospective cohort study, the Baltimore Longitudinal Study of Aging (n = 628–1305 persons depending on the cognitive outcome; ∼2 visits/person). Outcomes included 10 cognitive scores, spanning various domains of cognition. Caffeine and alcohol intakes and a nutrient adequacy score (NAS) were estimated from 7-d food diaries. Among key findings, caffeine intake was associated with better baseline global cognition among participants with a baseline age (Age_base) of ≥70 y. A higher NAS was associated with better baseline global cognition performance (overall, women, Age_base <70 y), better baseline verbal memory (immediate and delayed recall, Age_base ≥70 y), and slower rate of decline or faster improvement in the attention domain (women). For an Age_base of <70 y, alcohol consumption was associated with slower improvement on letter fluency and global cognition over time. Conversely, for an Age_base of ≥70 y and among women, alcohol intake was related to better baseline attention and working memory. In sum, patterns of diet and cognition associations indicate stratum-specific associations by sex and baseline age. The general observed trend was that of putative beneficial effects of caffeine intake and nutrient adequacy on domains of global cognition, verbal memory, and attention, and mixed effects of alcohol on domains of letter fluency, attention, and working memory. Further longitudinal studies conducted on larger samples of adults are needed to determine whether dietary factors individually or in combination are modifiers of cognitive trajectories among adults.

Introduction

Preventing age-related cognitive decline can help maintain quality of life (1). Dietary factors, such as the neuroactive compounds caffeine and alcohol, may affect cognitive health (2–4). Cognitive health benefits were also ascribed to healthy patterns of dietary intake and dietary quality (5). However, few studies with a prospective cohort design have examined all 3 predictors (i.e., caffeine and alcohol intakes and dietary quality) simultaneously, covarying for the others. In fact, to our knowledge, research has to date restricted its aim to a single cognitive test score, thus failing to incorporate multiple domains of cognition. Consequently, well-designed cohort studies are needed to clarify independent associations of dietary quality and caffeine and alcohol intakes with cognition. Such studies would ascertain temporality, include multiple cognitive domains, and would account for potential confounding effects within the diet.

Caffeine, primarily obtained from coffee, is the most widely used neuroactive compound worldwide (6). Acting as a brain stimulant, it causes heightened alertness and arousal (6) and can improve perceptual speed, vigilance, and even memory (7, 8). As a methylxanthine, caffeine blocks brain adenosine receptors, triggering cholinergic stimulation and potentially improving cognitive performance and slowing age-related cognitive decline (8). Caffeine’s putative beneficial effects on cognition are domain-specific (9–17). However, several studies found no association (18–20), whereas others noted differential associations by gender (9, 21).

Alcohol, on the other hand, is a well-known depressant drug widely consumed in Western diets. Although many large epidemiologic studies have examined cognition in relation to alcohol consumption, the direction of the association remains uncertain (22–51).

Importantly, caffeine- and alcohol-containing drinks are often consumed with other foods and with each other at different meals. Thus, intakes of caffeine and alcohol may be correlated with each other and with dietary quality or nutrient adequacy (i.e., an index of dietary quality based solely on nutrients). Furthermore, poor dietary quality has been associated with adverse cognitive outcomes, including decline and poor function (2, 52–58). Thus, dietary quality may confound relations of caffeine and alcohol intakes with cognition. Previous studies failed to account for this potential confounding effect. In our present study, we evaluated the independent associations of caffeine and alcohol intake and nutrient adequacy with cross-sectional and longitudinal cognitive performance in a U.S. population of older adults. We hypothesized caffeine to have putative beneficial and sex-specific effects for certain domains (e.g., attention, perceptual speed, and memory), and alcohol to have both beneficial and deleterious effects, depending on the population subgroup (age group or gender) and the cognitive domain. In contrast, nutrient adequacy was hypothesized to affect a wide range of domains in a positive manner.

Based on previous evidence, particularly in the case of caffeine and alcohol intake (9, 14, 21, 36, 40), we presented sex-specific findings. In addition, variations in nutrient requirements by age at cut-points of 50 y and 70 y had been noted [e.g., fiber (lower at ≥50 y vs. <50 y), sodium (lower at ≥70 y vs. <70 y), calcium (different ranges for ≥50 y vs. <50 y), iron (reduced with age), phosphorus [upper limit (UL)⁹ reduced at 70 y], vitamin B-6 (increases with age, including ≥50 y), and vitamin D (increases at age ≥70 y)] (59–62). Moreover, studies have commonly stratified by age associations of dietary components with cognition (15, 17, 40, 57). Therefore, we stratified our results by sex and age accordingly, similar to previous studies (9, 14, 15, 17, 21, 36, 40, 57).

Materials and Methods

Database and study population

The Baltimore Longitudinal Study of Aging (BLSA) is an ongoing prospective open cohort study of community-dwelling adults that was initiated in 1958 by the National Institute on Aging. BLSA participants were generally highly educated adults with a first-visit age of 17 to 97 y (median = 60.7; means ± SDs = 58.9 ± 18.0), and around 60% were men. Total enrollment included n₁ = 3047 participants (n′₁ = 20,385 visits, 1958–2009) (63). Exclusionary criteria are summarized elsewhere (64). Examinations were conducted at ∼2-y intervals, and the protocol was approved by the Medstar Research Institute’s Institutional Review Board. Examinations included physical, neurocognitive, medical history, dietary assessment, laboratory, and radiologic tests and measurements. Participants completed a written informed consent form per visit (63).

Eligible participants (n), observations or visits (n′), and visits/participants (n″) were as follows: 1) had complete dietary intake data from 1961 to 2007 (n₂ = 1821 participants; n′₂ = 4537 visits; n″₂ range = 1–12 visits/participant, mean = 2.5 visits/participant); 2) had complete cognitive data from 1962 to 2008 (n_3a-3j = 1199–2704 participants; n′_3a-3j = 5111–10,704 visits depending on the cognitive test score; n″₃ range = 1–22 visits/participant; mean = 3.4–4.4 visits/participant); and 3) had ≥1 visits that were concurrent (i.e., during the same visit/year) between dietary and cognitive data (n_4a-4j = 628–1305 participants; n′_4a-4j = 1218–2528 visits; n″₄ range = 1–10 visits/participant; mean = 1.9–2.0 visits/participant). Moreover, the mean (range) first-visit age for final samples with cognitive and dietary data (4a-4j) was 62 to 69 y (17–99 y), depending on the cognitive test. However, a distinction was made between first-visit characteristics (including age) and baseline characteristics. The “baseline” (base) visit was the earliest visit/year (Year_base) with concurrent data on diet and cognition. Timing was similar for most tests, except for the Benton Visual Retention Test (BVRT), historically initiated earlier in the BLSA. In samples 4a-4j, Year_base ranged from 1961 to 2007 for BVRT and 1985 to 2007 for other cognitive tests. The baseline age (Age_base) range for samples 4a-4j was 18 to 93 y (mean: 62 y) for BVRT and 27 to 96 y (mean: 68–72 y) for other cognitive tests.

Dietary assessment: caffeine and alcohol intake and nutrient adequacy score

Dietary intake was assessed with 7-d dietary records. BLSA participants were instructed by trained dietitians to estimate portion size, weigh foods, and complete the records (65–67). Intake was assessed in a noncontinuous fashion: 1961 to 1965 (1.01% of n′₂ = 4537), 1968 to 1975 (31.10% of n′₂), 1984 to 1992 (23.46% of n′₂), and 1994 to 2007 (44.3% of n′₂). Overall (n′₂ = 4537, 1961–2007), the means ± SDs of completed dietary records was 6.04 ± 1.73 (IQR: 6–7). Food codes and amounts were recorded for each diary, with nutrient intakes [absolute and relative amounts (i.e., per 1000 kcal or % energy)] estimated using a revised and up-to-date nutrient database (68) and averaged over available diaries per individual visit. Specifically, caffeine (per 100 mg/d) and alcohol (g/d) intakes and the nutrient adequacy score (NAS) were of primary interest. Moderate alcohol consumption was defined as 14 to 28 g/d and was compared with lower and higher amounts of intake as a sensitivity analysis, based on previous studies (22–35, 37–39).

To estimate the NAS, age/sex-specific DRIs for U.S. adults were used among others to categorize individuals according to adequacy of dietary intake for macronutrients (e.g., protein, carbohydrates, fat) and micronutrients (vitamins and minerals). Among DRI, adequate intake (AI) was used to reference an amount of vitamins and minerals above which a participant’s intake was adequate without exceeding the UL. For carbohydrates, protein, and total fat (% energy), the acceptable macronutrient distribution range was used instead. Saturated fat (% energy) intake in moderation was estimated using the 2005 Healthy Eating Index complete score. Similarly, cholesterol (mg) intake was determined adequate based on the 1995 Healthy Eating Index complete score (60–62).

For all NAS components, AI or AI and UL in combination were used as follows: 1) AI only: total fiber, potassium, thiamin, riboflavin, and vitamin E; and 2) AI and UL: sodium, calcium, iron, magnesium, phosphorus, zinc, retinol, niacin, vitamin B-6, folate (UL applied only to synthetic folic acid), and vitamins C and D (59–62). Nutrient adequacy was determined for each nutrient in relation to its age/sex-specific recommended intake (0: inadequate; 1: adequate). The NAS (range: 0–22) was computed as the sum of 22 nutrient components, with a higher score reflecting better overall nutrient adequacy, similar to a previous study (69).

Cognitive assessment

A battery of 6 cognitive tests was used.

Mini Mental State Examination.

Administered in the BLSA since the mid 1980s, the Mini Mental State Examination (MMSE) is a brief mental status test measuring orientation, concentration, immediate and delayed memory, language, and constructional praxis (70). Scores range from 0 to 30, with higher scores indicating better cognitive performance.

BVRT.

The BVRT is a test of short-term visual memory and constructional abilities (71). Administration A has been used in the BLSA since 1960, with a modified error scoring system, based on the BVRT manual scoring, such that higher scores indicate poorer visual memory.

California Verbal Learning Test.

Administered in the BLSA since 1993, the California Verbal Learning Test (CVLT) is a 16-item shopping list measuring verbal learning and memory. The variables of interest in this study were List A sum across 5 learning trials and long-delay free recall. Scores ranged from 0 to 80 for List A sum and 0 to 16 for long-delay free recall. Higher scores indicate better verbal memory (72).

Verbal Fluency Tests.

Administered in the BLSA since the mid 1980s, the Verbal Fluency Test includes the letter (F, A, S) assessment measuring phonemic fluency (VFT-L) (73, 74) and the categorical (fruits, animals, vegetables) assessment measuring semantic fluency (VFT-C) (75). Participants were required to generate as many words as possible for 60 s, starting with either a specific letter or category. Higher scores indicate better verbal fluency, with the total number of words, minus intrusions and perseverations, analyzed for each test.

Trail Making Tests.

Trail Making Tests A and B are tests of attention (Trails A) and executive functioning (Trails B), specifically cognitive control and visuo-motor scanning (76). Participants corrected incident errors by returning to their last correct response and continuing from there. The stopwatch recorded the time while corrections were made. Scores reflected time to completion (in seconds) separately for Trails A and B. Higher scores indicate poorer performance.

Digits Span Forward and Backward.

The Wechsler Adult Intelligence Scale-Revised, digits span-forward (DS-F) and digits span-backward (DS-B) (77), assesses attention and working memory, respectively. DS-F involves orally presenting a series of single-digit numbers at increasing digit span lengths for participants to repeat in the same order. The numbers’ span length ranges from 3 to 9 digits. Two trials at each span are presented. The test is discontinued when participants incorrectly repeat both trials at a specified span. DS-B is similar to DS-F, except that participants repeat a series of increasingly longer spans of single digit numbers in reverse order. The numbers’ span length ranges from 2 to 8 digits. The total score for both DS-F and DS-B is 14.

In sum, most cognitive test scores’ direction was “better performance with a higher score” the reverse was true for the BVRT and Trails A and B.

Covariates

Potentially confounding covariates were Age_base, Year_base, sex, race/ethnicity (non-Hispanic white, non-Hispanic black, and other ethnicity), education (y), smoking_base (“never,” “former”, or “current smoker”), and measured baseline BMI [BMI_base, weight/(height)² in kg/m²]. First-visit measures (i.e., visit 1 of the BLSA) were also considered for the descriptive part of the analysis.

Statistical methods

Analyses were performed using Stata version 11.0 (78). Participant characteristics (fixed and at first-visit) were described and compared by data availability and by sex using a 1-factor ANOVA, t test, and χ² test.

Mixed-effects linear regression models were used to examine associations of baseline caffeine and alcohol intakes and NAS with baseline cognitive performance (cross-sectional effect) and their relations with cognitive change over time (longitudinal effect), controlling for Age_base, sex, and other “baseline” or fixed covariates, including race/ethnicity, education, baseline smoking status, and baseline BMI. These models, which we term time-interval, mixed-effects regression models, were adapted from a previously published study that described the methodology in detail (79). Time elapsed (y) was measured from Age_base (per cognitive test).

Equations 1.1–1.4

Multilevel models vs. composite models

where Y_ij represents cognitive test scores for individual “i” and visit “j” , level-1 intercept for individual i; , level-1 slope for individual i; , level-2 intercept of the random intercept ; , level-2 intercept of the slope ; and , a vector of individual-level fixed baseline covariates (including Age_base) predicting level-1 intercepts () and level-1 slopes (). Among covariates, education (y) was centered at 16y (approximate mean for BLSA), BMI at 25 kg/m², total energy intake at 2000 kcal/d, Age_base at 50 y, and Year_base at 2000. X_ija are the main predictor variables: “caffeine centered at 0 mg/d,” “NAS centered at 10,” and “alcohol centered at 10 g/d” and are level-2 disturbances; and is the within-person level-1 disturbance. In a sensitivity analysis, alcohol was entered as categorical variable and interacted as such with time elapsed: 0 = 14 to 28 g/d, 1 = <14 g/d, 2 = >28 g/d. Thus, having lower and higher than moderate consumption was compared with moderate consumption.

Estimated parameters with SE and P values reflected rate of cognitive change over time (γ₁₀), the effects of caffeine and alcohol intakes and NAS on baseline cognitive performance (time = 0) (γ₀₁, γ₀₂, and γ₀₃; γ₀₃₁ and γ₀₃₂ for categorical alcohol), and effects of caffeine and alcohol intakes and NAS on annual rate of cognitive change over time (γ₁₁, γ₁₂, and γ₁₃; γ₁₃₁ and γ₁₃₂ for categorical alcohol). Analysis was presented for the overall sample and was further stratified by sex and Age_base (<70 y vs. ≥70 y).

The effect modification by sex and Age_base was tested by including additional interaction terms in the “overall population” model (e.g., Age_base× caffeine and Age_base× caffeine × time, separately). We used a 2-stage Heckman selection model adjusting for bias because of nonrandom participant selection for final analyses (80, 81).

We further estimated cognitive test scores and plotted their predicted means against time, with Age_base set alternatively at 50 y and 70 y. Each exposure was examined separately controlling for the other covariates. Caffeine intake was alternatively set at 0 mg/d vs. 300 mg/d; alcohol intake at 10 g/d vs. 50 g/d; and NAS at 5 vs. 15. Thus, cognitive performance trajectories for the hypothetical population with set covariate distribution was examined over time and compared by exposure level to illustrate direction and magnitude of fixed effects γ_0a and γ_1a.

Type I error was set at 0.05 for each of the 3 exposure variable-related hypotheses. Adjustment for multiple testing reduced the type I error to 0.05/3 = 0.017, and thus only P values <0.017 were considered statistically significant. However, type I error for 3-way interaction terms was set to 0.10 because of reduced power to detect significant associations (82).

Results

As shown in Table 1, compared with men, women were generally older and had more ethnic diversity, performed better on the MMSE, and had lower prevalence of current smoking. The distribution of study characteristics by sex and data completeness is presented in Supplemental Table 1 and Supplemental Figure 1.

TABLE 1.

Baseline characteristics of participants included in the final analysis with the MMSE (global cognitive function) and dietary data available, stratified by sex, BLSA, 1962–2008¹

	n2	n′3	Values
Men, n	415
First-visit age, y	407		66.8 ± 13.9*
Race/ethnicity	407
Non-Hispanic white	347		85.3*
Non-Hispanic black	50		12.3
Other	10		2.5
First-visit education, y	378		16.9 ± 2.7*
First-visit smoking	366
Never smoker	120		32.8*
Former smoker	184		50.3
Current smoker	62		16.9
First-visit BMI, kg/m2	393		26.3 ± 3.7
Energy intake, kcal/d		1458	2156 ± 565*
NAS		1458	11.10 ± 3.02*
>10 (above median), %		1458	56.5*
Caffeine, mg/d		1458	127.4 ± 211.0
100–300 mg/d (1–3 cups of coffee), %		1458	29.8*
Alcohol, g/d		1458	11.5 ± 0.4*
14–28 g/d (1–2 drinks), %		1458	18.2
MMSE total score		753	28.4 ± 2.2*
Women, n	312
First-visit age, y	309		69.8 ± 12.1
Race/ethnicity	311
Non-Hispanic white	234		75.2
Non-Hispanic black	62		19.9
Other	15		4.8
First-visit education, y	282		15.8 ± 2.6
First-visit smoking	251
Never smoker	102		40.6
Former smoker	124		49.4
Current smoker	25		10.0
First-visit BMI, kg/m2	301		25.9 ± 4.6
Energy intake, kcal/d		978	1697 ± 427
NAS		978	12.87 ± 3.16
>10 (above median), %		978	77.4
Caffeine, mg/d		978	138.7 ± 194.0
100–300 mg/d (1–3 cups of coffee), %		978	40.8
Alcohol, g/d		978	5.8 ± 0.3
14–28 g/d (1–2 drinks), %		978	9.8
MMSE total score		680	28.8 ± 2.1

¹Values are means ± SDs or percentages. *P < 0.05 for null hypothesis of no sex difference between means or proportions using the t test and χ² test, respectively, within each of the samples. BLSA, Baltimore Longitudinal Study of Aging; MMSE, Mini Mental State Examination; NAS, nutrient adequacy score.

²Number of participants in the analysis.

³Total number of visits included in the analysis.

Several key findings emerged from the time-interval, mixed-effects regression models. For most cognitive tests, younger participants at baseline performed better than older participants (γ_Age in the direction of poorer performance with higher age, Table 2). For a few tests, there was also an appreciable decline over time (γ₁₀ in the direction of decline), controlling for Age_base, whereas for others there was a learning effect over time that was reduced or tapered off with increasing age (γ₁₀ in the direction of improvement over time, P < 0.05, but with a γ_Age×Time in the direction of tapered-off learning and an eventual time-related decline at higher Age_base values).

TABLE 2.

Analysis of baseline caffeine intake (continuous, 100 mg/d), alcohol intake (g/d), and the NAS, and longitudinal change in cognitive performance (total and sex-stratified), time-interval mixed-effects linear regression analysis, BLSA, 1962–2008¹

	Total: model 1		Men: model 2		Women: model 3
	γ ± SEE2	P3	γ ± SEE2	P3	γ ± SEE2	P3
MMSE, total score4	n = 555	n′ = 1102	n = 328	n′ = 595	n = 227	n′ = 507
Fixed effects
Intercept (γ00 for π0i)	+29.48 ± 0.27	<0.001	+29.76 ± 0.39	<0.001	+29.22 ± 0.36	<0.001
Time (γ10 for π1i)	+0.135 ± 0.106	0.201	+0.086 ± 0.148	0.559	+0.191 ± 0.182	0.294
Agebase	−0.052 ± 0.009	<0.001	−0.067 ± 0.013	<0.001	−0.032 ± 0.011	0.004
Agebase × time	−0.009 ± 0.004	0.014	−0.005 ± 0.006	0.386	−0.011 ± 0.006	0.044
Gender (women vs. men)	+0.182 ± 0.159	0.253	—	—	—	—
Gender × time	+0.020 ± 0.059	0.733	—	—	—	—
Caffeine (γ01 for π0i)	+0.094 ± 0.047	0.043	+0.001 ± 0.001	0.099	+0.060 ± 0.072	0.407
Caffeine × time (γ11 for π1i)	−0.004 ± 0.016	0.795	−0.000 ± 0.000	0.322	+0.008 ± 0.032	0.813
NAS (γ02 for π0i)	+0.072 ± 0.025	0.004	+0.058 ± 0.034	0.094	+0.104 ± 0.036	0.003
NAS × time (γ12 for π1i)	−0.004 ± 0.009	0.624	+0.003 ± 0.012	0.838	−0.013 ± 0.016	0.405
Alcohol (γ03 for π0i)	+0.006 ± 0.006	0.295	+0.008 ± 0.006	0.198*	−0.008 ± 0.011	0.432
Alcohol × time (γ13 for π1i)	−0.003 ± 0.002	0.095	−0.001 ± 0.002	0.692*	−0.008 ± 0.005	0.132
Random effects
Level 1 residuals (Rij)	+0.79 ± 0.03	<0.001	+0.93 ± 0.05	<0.001	+0.622 ± 0.034	<0.001
Level 2 residuals
Intercept (ξ0i)	+1.38 ± 0.05	<0.001	+1.36 ± 0.08	<0.001	+1.305 ± 0.034	<0.001
Linear slope (ξ1i)	+0.32 ± 0.02	<0.001	+0.30 ± 0.03	<0.001	+0.389 ± 0.037	<0.001
CVLT-List A, total score	n = 568	n’ = 1111	n = 296	n’ = 559	n = 272	n’ = 552
Intercept (γ00 for π0i)	+58.32 ± 1.29	<0.001	+59.57 ± 1.85	<0.001	+63.18 ± 1.71	<0.001
Time (γ10 for π1i)	+0.060 ± 0.393	0.878	+0.191 ± 0.535	0.720	−0.175 ± 0.52	0.751
Agebase	−0.424 ± 0.036	<0.001	−0.472 ± 0.058	<0.001	−0.380 ± 0.046	<0.001
Agebase × time	−0.010 ± 0.010	0.317	−0.004 ± 0.015	0.789	−0.008 ± 0.014	0.554
Gender (women vs. men)	+6.547 ± 0.957	<0.001	—	—	—	—
Gender × time	−0.269 ± 0.249	0.280	—	—	—	—
Caffeine (γ01 for π0i)	+0.090 ± 0.276	0.745	−0.204 ± 0.348	0.559	+0.587 ± 0.480	0.222
Caffeine × time (γ11 for π1i)	+0.004 ± 0.064	0.944	−0.063 ± 0.078	0.419	+0.076 ± 0.109	0.484
NAS (γ02 for π0i)	+0.291 ± 0.152	0.055	+0.365 ± 0.231	0.114	+0.282 ± 0.203	0.164
NAS × time (γ12 for π1i)	+0.086 ± 0.040	0.031	+0.116 ± 0.056	0.040	+0.081 ± 0.059	0.169
Alcohol (γ03 for π0i)	+0.053 ± 0.031	0.088	+0.050 ± 0.038	0.191	+0.059 ± 0.057	0.295
Alcohol × time (γ13 for π1i)	−0.008 ± 0.007	0.208	−0.007 ± 0.007	0.298*	+0.016 ± 0.017	0.351
CVLT-delayed recall, total score	n = 568	n’ = 1111	n = 296	n’ = 559	n = 272	n’ = 552
Intercept (γ00 for π0i)	+12.52 ± 0.40	<0.001	+12.94 ± 0.58	<0.001	+13.47 ± 0.52	<0.001
Time (γ10 for π1i)	−0.021 ± 0.113	0.850	+0.008 ± 0.159	0.959	−0.072 ± 0.151	0.631
Agebase	−0.112 ± 0.011	<0.001	−0.139 ± 0.018	<0.001	−0.089 ± 0.014	<0.001
Agebase × time	−0.004 ± 0.003	0.157	−0.004 ± 0.004	0.376	−0.003 ± 0.004	0.502
Gender (women vs. men)	+1.281 ± 0.299	<0.001	—	—	—	—
Gender × time	−0.024 ± 0.071	0.738	—	—	—	—
Caffeine (γ01 for π0i)	+0.015 ± 0.086	0.858	+0.006 ± 0.110	0.953	−0.017 ± 0.146	0.909
Caffeine × time (γ11 for π1i)	−0.083 ± 0.018	0.964	−0.033 ± 0.023	0.148*	+0.042 ± 0.030	0.157
NAS (γ02 for π0i)	+0.102 ± 0.047	0.030	+0.131 ± 0.073	0.071	+0.092 ± 0.062	0.140
NAS × time (γ12 for π1i)	+0.017 ± 0.011	0.131	+0.029 ± 0.017	0.081	+0.014 ± 0.016	0.373
Alcohol (γ03 for π0i)	+0.013 ± 0.010	0.174	+0.013 ± 0.012	0.550	+0.012 ± 0.017	0.477
Alcohol × time (γ13 for π1i)	−0.001 ± 0.002	0.498	−0.001 ± 0.002	0.550	+0.004 ± 0.005	0.366
BVRT, total errors	n = 1005	n’ = 1975	n = 620	n’ = 1152	n = 385	n’ = 822
Intercept (γ00 for π0i)	+3.68 ± 0.31	<0.001	+4.03 ± 0.39	<0.001	+3.49 ± 0.43	<0.001
Time (γ10 for π1i)	+0.051 ± 0.048	0.288	+0.051 ± 0.058	0.384	−0.072 ± 0.082	0.377
Agebase	+0.112 ± 0.006	<0.001	+0.110 ± 0.008	<0.001	+0.111 ± 0.010	<0.001
Agebase × time	+0.005 ± 0.001	<0.001	+0.004 ± 0.001	<0.001	+0.005 ± 0.002	0.004
Gender (women vs. men)	−0.005 ± 0.244	0.985	—	—	—	—
Gender × time	−0.054 ± 0.035	0.128	—	—	—	—
Caffeine (γ01 for π0i)	−0.157 ± 0.070	0.024	−0.181 ± 0.092	0.050	−0.164 ± 0.112	0.144
Caffeine × time (γ11 for π1i)	+0.012 ± 0.005	0.031	+0.008 ± 0.012	0.507	+0.011 ± 0.007	0.138
NAS (γ02 for π0i)	−0.065 ± 0.037	0.076	−0.052 ± 0.051	0.311	−0.090 ± 0.054	0.099
NAS × time (γ12 for π1i)	−0.001 ± 0.005	0.878	+0.002 ± 0.007	0.795	−0.002 ± 0.008	0.776
Alcohol (γ03 for π0i)	−0.007 ± 0.007	0.307	−0.001 ± 0.007	0.933	−0.020 ± 0.016	0.220
Alcohol × time (γ13 for π1i)	−0.000 ± 0.001	0.851	−0.000 ± 0.001	0.539	+0.001 ± 0.002	0.718
VFT-C, total score	n = 602	n’ = 1236	n = 346	n’ = 655	n = 256	n’ = 581
Intercept (γ00 for π0i)	+18.29 ± 0.47	<0.001	+18.21 ± 0.65	<0.001	+20.07 ± 0.70	<0.001
Time (γ10 for π1i)	+0.167 ± 0.093	0.073	+0.128 ± 0.140	0.362	+0.314 ± 0.138	0.022
Agebase	−0.169 ± 0.015	<0.001	−0.173 ± 0.022	<0.001	−0.165 ± 0.023	<0.001
Agebase × time	−0.014 ± 0.003	<0.001	−0.012 ± 0.005	0.030	−0.017 ± 0.004	<0.001
Gender (women vs. men)	−0.341 ± 0.460	0.458	—	—	—	—
Gender × time	−0.084 ± 0.139	0.545	—	—	—	—
Caffeine (γ01 for π0i)	−0.036 ± 0.009	0.680	+0.002 ± 0.108	0.985	−0.001 ± 0.002	0.481
Caffeine × time (γ11 for π1i)	+0.006 ± 0.014	0.633	−0.003 ± 0.018	0.871	+0.000 ± 0.000	0.231
NAS (γ02 for π0i)	+0.051 ± 0.047	0.279	+0.052 ± 0.062	0.410	+0.056 ± 0.074	0.443
NAS × time (γ12 for π1i)	+0.003 ± 0.008	0.741	+0.001 ± 0.012	0.907	+0.004 ± 0.012	0.739
Alcohol (γ03 for π0i)	+0.012 ± 0.010	0.257	+0.006 ± 0.011	0.576	+0.025 ± 0.023	0.263
Alcohol × time (γ13 for π1i)	−0.000 ± 0.001	0.975	−0.000 ± 0.002	0.967	+0.003 ± 0.003	0.352
VFT-L, total score	n = 601	n’ = 1233	n = 346	n′ = 645	n = 255	n’ = 577
Intercept (γ00 for π0i)	+15.13 ± 0.60	<0.001	+14.43 ± 0.87	<0.001	+16.25 ± 0.84	<0.001
Time (γ10 for π1i)	+0.239 ± 0.102	0.020	+0.273 ± 0.140	0.052	+0.190 ± 0.174	0.277
Agebase	−0.040 ± 0.019	0.041	−0.019 ± 0.029	0.515	−0.057 ± 0.027	0.036
Agebase × time	−0.012 ± 0.003	0.001	−0.013 ± 0.005	0.012	−0.012 ± 0.005	0.026
Gender (women vs. men)	+0.729 ± 0.387	0.059	—	—	—	—
Gender × time	−0.001 ± 0.055	0.987	—	—	—	—
Caffeine (γ01 for π0i)	+0.028 ± 0.113	0.980	+0.097 ± 0.145	0.502*	−0.116 ± 0.187	0.533
Caffeine × time (γ11 for π1i)	−0.023 ± 0.015	0.121	−0.038 ± 0.018	0.039	−0.000 ± 0.000	0.994
NAS (γ02 for π0i)	−0.051 ± 0.060	0.393	+0.033 ± 0.084	0.687	−0.139 ± 0.088	0.113
NAS × time (γ12 for π1i)	+0.001 ± 0.009	0.891	+0.006 ± 0.012	0.586	+0.001 ± 0.015	0.941
Alcohol (γ03 for π0i)	+0.021 ± 0.013	0.107	+0.028 ± 0.015	0.066	−0.006 ± 0.027	0.827
Alcohol × time (γ13 for π1i)	−0.002 ± 0.002	0.141	−0.002 ± 0.002	0.294	−0.002 ± 0.004	0.652
DS-F, total score	n = 541	n’ = 1067	n = 285	n’ = 544	n = 256	n’ = 523
Intercept (γ00 for π0i)	+9.50 ± 0.30	<0.001	+10.04 ± 0.43	<0.001	+8.74 ± 0.40	<0.001
Time (γ10 for π1i)	−0.002 ± 0.060	0.978	+0.015 ± 0.093	0.869	+0.013 ± 0.077	0.862
Agebase	−0.036 ± 0.008	<0.001	−0.053 ± 0.013	<0.001	−0.026 ± 0.010	0.011
Agebase × time	−0.002 ± 0.001	0.237	−0.002 ± 0.002	0.293	−0.001 ± 0.002	0.588
Gender (women vs. men)	−0.493 ± 0.215	0.022	—	—	—	—
Gender × time	+0.012 ± 0.034	0.729	—	—	—	—
Caffeine (γ01 for π0i)	−0.039 ± 0.061	0.519	−0.077 ± 0.076	0.312	+0.031 ± 0.106	0.773
Caffeine × time (γ11 for π1i)	−0.010 ± 0.009	0.286	−0.005 ± 0.013	0.706	−0.022 ± 0.014	0.111
NAS (γ02 for π0i)	−0.003 ± 0.035	0.934	−0.002 ± 0.053	0.975	−0.010 ± 0.047	0.835
NAS × time (γ12 for π1i)	+0.010 ± 0.006	0.085	+0.001 ± 0.009	0.988	+0.022 ± 0.008	0.005
Alcohol (γ03 for π0i)	+0.015 ± 0.007	0.036	+0.022 ± 0.009	0.012	+0.003 ± 0.013	0.788
Alcohol × time (γ13 for π1i)	+0.000 ± 0.001	0.850	−0.001 ± 0.001	0.406	+0.003 ± 0.002	0.156
DS-B, total score	n = 545	n’ = 1066	n = 284	n’ = 540	n = 261	n’ = 526
Intercept (γ00 for π0i)	+8.90 ± 0.32	<0.001	+9.08 ± 0.48	<0.001	+8.57 ± 0.40	<0.001
Time (γ10 for π1i)	−0.367 ± 0.098	<0.001	−0.383 ± 0.143	0.008	−0.298 ± 0.132	0.024
Agebase	−0.036 ± 0.009	<0.001	−0.044 ± 0.015	0.003	−0.032 ± 0.011	0.002
Agebase × time	+0.006 ± 0.002	0.008	+0.008 ± 0.004	0.041	+0.006 ± 0.003	0.050
Gender (women vs. men)	−0.257 ± 0.224	0.251	—	—	—	—
Gender × time	+0.103 ± 0.057	0.073	—	—	—	—
Caffeine (γ01 for π0i)	−0.040 ± 0.064	0.533	−0.068 ± 0.084	0.418	−0.034 ± 0.111	0.759
Caffeine × time (γ11 for π1i)	−0.008 ± 0.015	0.576	−0.022 ± 0.020	0.263	+0.014 ± 0.024	0.558
NAS (γ02 for π0i)	−0.051 ± 0.036	0.160	−0.052 ± 0.058	0.373	−0.035 ± 0.047	0.462
NAS × time (γ12 for π1i)	+0.003 ± 0.010	0.742	−0.003 ± 0.015	0.849	+0.012 ± 0.013	0.378
Alcohol (γ03 for π0i)	+0.021 ± 0.007	0.004	+0.024 ± 0.009	0.013	+0.018 ± 0.013	0.171
Alcohol × time (γ13 for π1i)	−0.002 ± 0.002	0.106	−0.003 ± 0.002	0.078	+0.001 ± 0.004	0.788

¹Models were further adjusted for baseline year of intake, race/ethnicity, education (y), baseline smoking status, and baseline BMI. See Materials and Methods for more details on covariate coding and model specifications. Random effects are presented only for the MMSE, for simplicity. *P < 0.10 for interaction with gender to test effect modification by gender for each of the 3 predictors’ effects (i.e., caffeine intake, alcohol intake, and NAS) on cognitive performance at baseline and cognitive change over time. BLSA, Baltimore Longitudinal Study of Aging; BVRT, Benton Visual Retention Test; CVLT, California Verbal Learning Test; DS-B, digits span-backward; DS-F, digits span-forward; MMSE, Mini Mental State Examination; NAS, nutrient adequacy score; Trails A, Trail Making Test, part A; Trails B, Trail Making Test, part B; VFT-C, Verbal Fluency Test-Categorical; VFT-L, Verbal Fluency Test-Letter.

²n = number of participants in the analysis.

³n′ = total number of visits included in the analysis.

⁴Cognitive scores were in the direction of higher score indicated better performance with the exception of the BVRT and Trails A and B.

Importantly, when examining associations of caffeine intake, NAS, and alcohol intake with cognitive performance at baseline (Tables 2 and 3; γ₀₁, γ₀₂, and γ₀₃), and change over time (γ₁₁, γ₁₂, and γ₁₃), there was apparent effect modification by Age_base (<70 y vs. ≥70 y) and, occasionally, by sex. When testing interaction terms, sex and Age_base differentials in diet’s association with cognition (cross-sectional and longitudinal) were significant for some but not all associations (P < 0.10).

TABLE 3.

Analysis of baseline caffeine intake (continuous, 100 mg/d), alcohol intake (g/d), and the NAS, and longitudinal change in cognitive performance (Age_base-stratified), time-interval, mixed-effects linear regression analysis, BLSA, 1962–2008¹

	Baseline age <70 y: model 4		Baseline age ≥70 y: model 5
	γ ± SEE2	P3	γ ± SEE2	P3
MMSE, total score4	n = 243	n′ = 514	n = 312	n′ = 588
Fixed effects
Intercept (γ00 for π0i)	+29.07 ± 0.40	<0.001	+29.49 ± 0.62	<0.001
Time (γ10 for π1i)	+0.164 ± 0.099	0.097	+0.056 ± 0.363	0.878
Agebase	+0.001 ± 0.024	0.979	−0.064 ± 0.019	0.001
Agebase × time	−0.009 ± 0.005	0.087	−0.007 ± 0.011	0.532
Gender (women vs. men)	−0.149 ± 0.182	0.412	+0.512 ± 0.247	0.032
Gender × time	+0.050 ± 0.037	0.178	+0.017 ± 0.127	0.894
Caffeine (γ01 for π0i)	−0.004 ± 0.005	0.922*	+0.218 ± 0.083	0.008
Caffeine × time (γ11 for π1i)	−0.008 ± 0.010	0.377*	+0.016 ± 0.040	0.700
NAS (γ02 for π0i)	+0.084 ± 0.027	0.002	+0.077 ± 0.040	0.055
NAS × time (γ12 for π1i)	−0.014 ± 0.006	0.017	−0.007 ± 0.020	0.723
Alcohol (γ03 for π0i)	+0.005 ± 0.006	0.325	+0.010 ± 0.010	0.288
Alcohol × time (γ13 for π1i)	−0.002 ± 0.001	0.008	−0.008 ± 0.005	0.139
Random effects
Level 1 residuals (Rij)	+0.80 ± 0.05	<0.001	+0.78 ± 0.04	<0.001
Level 2 residuals
Intercept (ξ0i)	+0.91 ± 0.07	<0.001	+1.61 ± 0.08	<0.001
Linear slope (ξ1i)	+0.10 ± 0.02	<0.001	+0.49 ± 0.04	<0.001
CVLT-List A, total score	n = 321	n’ = 579	n = 247	n’ = 532
Intercept (γ00 for π0i)	+57.20 ± 1.49	<0.001	+68.51 ± 4.02	<0.001
Time (γ10 for π1i)	+0.215 ± 0.559	0.701	−0.672 ± 1.061	0.526
Agebase	−0.264 ± 0.070	<0.001	−0.747 ± 0.121	<0.001
Agebase × time	−0.010 ± 0.020	0.628	+0.006 ± 0.035	0.852
Gender (women vs. men)	+6.768 ± 1.230	<0.001	+7.449 ± 1.552	<0.001
Gender × time	−0.297 ± 0.347	0.392	−0.218 ± 0.457	0.633
Caffeine (γ01 for π0i)	+0.479 ± 0.323	0.138*	−0.796 ± 0.503	0.113
Caffeine × time (γ11 for π1i)	+0.005 ± 0.074	0.944	−0.043 ± 0.141	0.763
NAS (γ02 for π0i)	+0.018 ± 0.183	0.919	+0.630 ± 0.260	0.015
NAS × time (γ12 for π1i)	+0.128 ± 0.054	0.019	+0.030 ± 0.068	0.654
Alcohol (γ03 for π0i)	+0.035 ± 0.035	0.311	+0.098 ± 0.061	0.110
Alcohol × time (γ13 for π1i)	−0.009 ± 0.008	0.221	−0.010 ± 0.017	0.563
CVLT-delayed recall, total score	n = 321	n’ = 579	n = 247	n’ = 532
Intercept (γ00 for π0i)	+12.73 ± 0.45	<0.001	+14.97 ± 1.29	<0.001
Time (γ10 for π1i)	+0.060 ± 0.146	0.682	−0.344 ± 0.331	0.299
Agebase	−0.053 ± 0.021	0.011	−0.217 ± 0.039	<0.001
Agebase × time	−0.003 ± 0.005	0.521	+0.001 ± 0.011	0.899
Gender (women vs. men)	+1.091 ± 0.369	0.003	+1.921 ± 0.498	<0.001
Gender × time	−0.075 ± 0.090	0.407	−0.004 ± 0.498	0.975
Caffeine (γ01 for π0i)	+0.053 ± 0.097	0.580	−0.111 ± 0.161	0.490
Caffeine × time (γ11 for π1i)	+0.009 ± 0.019	0.656*	−0.051 ± 0.044	0.247
NAS (γ02 for π0i)	+0.005 ± 0.055	0.921*	+0.230 ± 0.083	0.006
NAS × time (γ12 for π1i)	+0.025 ± 0.014	0.077	+0.014 ± 0.021	0.489
Alcohol (γ03 for π0i)	+0.010 ± 0.010	0.336	+0.016 ± 0.019	0.410
Alcohol × time (γ13 for π1i)	−0.002 ± 0.002	0.242	−0.000 ± 0.005	0.967
BVRT, total errors	n = 667	n′ = 1354	n = 338	n’ = 620
Intercept (γ00 for π0i)	+3.13 ± 0.34	<0.001	+2.09 ± 1.25	<0.001
Time (γ10 for π1i)	+0.002 ± 0.041	0.956	+0.745 ± 0.408	0.068
Agebase	+0.073 ± 0.008	<0.001	0.183 ± 0.038	<0.001
Agebase × time	+0.003 ± 0.001	<0.001	−0.013 ± 0.013	0.292
Gender (women vs. men)	+0.263 ± 0.272	0.336	−0.341 ± 0.460	0.458
Gender × time	−0.043 ± 0.032	0.173	−0.084 ± 0.139	0.545
Caffeine (γ01 for π0i)	−0.096 ± 0.071	0.179	−0.082 ± 0.161	0.611
Caffeine × time (γ11 for π1i)	+0.005 ± 0.005	0.278	+0.073 ± 0.046	0.116
NAS (γ02 for π0i)	−0.073 ± 0.039	0.060	−0.013 ± 0.077	0.866
NAS × time (γ12 for π1i)	+0.009 ± 0.004	0.044	−0.037 ± 0.021	0.075
Alcohol (γ03 for π0i)	−0.008 ± 0.007	0.226	−0.004 ± 0.017	0.814
Alcohol × time (γ13 for π1i)	−0.000 ± 0.000	0.620	+0.003 ± 0.005	0.533
VFT-C, total score	n = 275	n’ = 561	n = 327	n’ = 675
Intercept (γ00 for π0i)	+17.32 ± 0.68	<0.001	+20.08 ± 1.06	<0.001
Time (γ10 for π1i)	+0.532 ± 0.177	0.003	+0.152 ± 0.242	0.528
Agebase	−0.149 ± 0.038	<0.001	−0.219 ± 0.032	<0.001
Agebase × time	−0.034 ± 0.010	<0.001	−0.018 ± 0.008	0.027
Gender (women vs. men)	+1.771 ± 0.458	<0.001	+1.780 ± 0.411	<0.001
Gender × time	+0.087 ± 0.072	0.230	+0.011 ± 0.081	0.896
Caffeine (γ01 for π0i)	+0.021 ± 0.116	0.853	−0.001 ± 0.001	0.553
Caffeine × time (γ11 for π1i)	−0.012 ± 0.017	0.462	+0.000 ± 0.000	0.213
NAS (γ02 for π0i)	+0.086 ± 0.066	0.193	+0.012 ± 0.067	0.857
NAS × time (γ12 for π1i)	−0.012 ± 0.011	0.260*	+0.019 ± 0.013	0.139
Alcohol (γ03 for π0i)	+0.002 ± 0.013	0.890	+0.030 ± 0.016	0.059
Alcohol × time (γ13 for π1i)	−0.001 ± 0.002	0.658	+0.002 ± 0.003	0.627
VFT-L, total score	n = 275	n’ = 561	n = 326	n’ = 672
Intercept (γ00 for π0i)	+14.74 ± 0.84	<0.001	+16.95 ± 1.44	<0.001
Time (γ10 for π1i)	+0.700 ± 0.185	<0.001	+0.111 ± 0.259	0.667
Agebase	−0.002 ± 0.047	0.973	−0.113 ± 0.044	0.008
Agebase × time	−0.036 ± 0.010	<0.001	−0.011 ± 0.009	0.212
Gender (women vs. men)	+0.792 ± 0.555	0.153	+0.896 ± 0.558	0.108
Gender × time	−0.018 ± 0.070	0.792	−0.077 ± 0.086	0.371
Caffeine (γ01 for π0i)	−0.147 ± 0.142	0.300	+0.241 ± 0.187	0.198
Caffeine × time (γ11 for π1i)	−0.026 ± 0.018	0.136	−0.016 ± 0.028	0.552
NAS (γ02 for π0i)	−0.084 ± 0.081	0.298	+0.006 ± 0.090	0.951
NAS × time (γ12 for π1i)	−0.012 ± 0.011	0.273	+0.013 ± 0.014	0.334
Alcohol (γ03 for π0i)	+0.027 ± 0.016	0.096	+0.018 ± 0.022	0.420
Alcohol × time (γ13 for π1i)	−0.006 ± 0.002	0.001	+0.003 ± 0.004	0.348
DS-F, total score	n = 287	n’ = 507	n = 254	n’ = 560
Intercept (γ00 for π0i)	+9.80 ± 0.39	<0.001	+10.00 ± 0.82	<0.001
Time (γ10 for π1i)	−0.021 ± 0.098	0.826	−0.095 ± 0.136	0.481
Agebase	−0.024 ± 0.018	0.181	−0.068 ± 0.024	0.006
Agebase × time	−0.001 ± 0.003	0.708	+0.004 ± 0.004	0.371
Gender (women vs. men)	−0.652 ± 0.313	0.037	−0.227 ± 0.316	0.472
Gender × time	−0.035 ± 0.053	0.514	+0.051 ± 0.048	0.283
Caffeine (γ01 for π0i)	−0.099 ± 0.080	0.994	+0.053 ± 0.101	0.600
Caffeine × time (γ11 for π1i)	−0.001 ± 0.001	0.994	−0.030 ± 0.016	0.062
NAS (γ02 for π0i)	−0.044 ± 0.048	0.357	+0.059 ± 0.053	0.264
NAS × time (γ12 for π1i)	+0.018 ± 0.009	0.040	−0.001 ± 0.008	0.930
Alcohol (γ03 for π0i)	+0.009 ± 0.009	0.291	+0.024 ± 0.012	0.045
Alcohol × time (γ13 for π1i)	−0.000 ± 0.001	0.694	−0.000 ± 0.002	0.962
DS-B, total score	n = 289	n’ = 511	n = 256	n’ = 555
Intercept (γ00 for π0i)	+8.93 ± 0.42	<0.001	+9.38 ± 0.86	<0.001
Time (γ10 for π1i)	−0.478 ± 0.148	0.001	−0.186 ± 0.242	0.444
Agebase	−0.036 ± 0.020	0.068	−0.054 ± 0.026	0.037
Agebase × time	+0.011 ± 0.004	0.018	+0.001 ± 0.008	0.849
Gender (women vs. men)	−0.251 ± 0.336	0.455	−0.213 ± 0.330	0.518
Gender × time	+0.141 ± 0.084	0.093	+0.126 ± 0.089	0.154
Caffeine (γ01 for π0i)	−0.027 ± 0.086	0.756	−0.052 ± 0.108	0.626
Caffeine × time (γ11 for π1i)	−0.004 ± 0.017	0.980	−0.030 ± 0.029	0.296
NAS (γ02 for π0i)	−0.075 ± 0.051	0.138	−0.024 ± 0.056	0.659
NAS × time (γ12 for π1i)	−0.062 ± 0.013	0.963	+0.06 ± 0.016	0.700
Alcohol (γ03 for π0i)	+0.015 ± 0.010	0.116	+0.030 ± 0.012	0.015
Alcohol × time (γ13 for π1i)	−0.002 ± 0.002	0.303	−0.006 ± 0.004	0.121

¹Models were further adjusted for baseline year of intake, race/ethnicity, education (y), baseline smoking status and baseline BMI. See Materials and Methods for more details on covariate coding and model specifications. Random effects are presented only for the MMSE, for simplicity. *P < 0.10 for interaction with baseline age to test effect modification by age for each of the 3 predictors’ effects (i.e., caffeine intake, alcohol intake, and NAS) on cognitive performance at baseline and cognitive change over time. BLSA, Baltimore Longitudinal Study of Aging; BVRT, Benton Visual Retention Test; CVLT, California Verbal Learning Test; DS-B, digits span-backward; DS-F, digits span-forward; MMSE, Mini Mental State Examination; NAS, nutrient adequacy score; Trails A, Trail Making Test, part A; Trails B, Trail Making Test, part B; VFT-C, Verbal Fluency Test-Categorical; VFT-L, Verbal Fluency Test-Letter.

²n = number of participants in the analysis.

³n′ = total number of visits included in the analysis.

⁴Cognitive scores were in the direction of higher score indicated better performance with the exception of the BVRT and Trails A and B.

Among key findings, caffeine intake was associated with better global cognitive function (MMSE) at baseline for those ≥70 y (P = 0.008), independently of potential confounding covariates. Second, the NAS was associated with better baseline performance on the MMSE overall (P = 0.004), in women (P = 0.003), and in those <70 y (P = 0.003). When Age_base was ≥70 y, the NAS was associated with better baseline performance on the CVLT-List A and CVLT-Delayed Recall. Similarly, the NAS was associated with slower decline or improvement over time on the DS-F test among women. Finally, alcohol intake was associated with faster decline or slower improvement on the MMSE (P = 0.008) and on the VFT-L test (P = 0.001) when Age_base was <70 y. However, when Age_base was ≥70 y, alcohol was related to better baseline performance on the DS-F and DS-B (Tables 2 and 3). There were no significant associations between any of the continuous dietary exposures and Trails A or B (Supplemental Table 2).

Our key findings are illustrated visually in Supplemental Figure 2 if a population with fixed characteristics (listed in the footnote of the supplemental figures) was followed up from ages 50 y and 70 y for a period of ∼9y and was alternatively exposed to 2 different levels of each dietary exposure, keeping the other exposures and covariates constant.

In a sensitivity analysis (Supplemental Table 3), categorical alcohol intake was entered into time-interval, mixed-effects regression models. Compared with 14 to 28 g/d consumption, individuals with >28 g/d of alcohol intake had faster decline or slower improvement on the MMSE, particularly among women and in the older group (Age_base ≥70 y, γ₁₃₂ = −0.59 ± 0.22, P = 0.009). This finding is at slight odds from our previous results with continuous alcohol intake where we found this relation in the younger group. Moreover, consuming <14 g/d was associated with slower decline or faster improvement in the VFT-L compared with a moderate intake of 14 to 28 g/d (Age_base <70 y, γ₁₃₁ = +0.24 ± 0.07, P < 0.001). A similar pattern was noted also in the younger group for both Trails A (Age_base <70 y, γ₁₃₁ = −1.40 ± 0.53, P = 0.009) and Trails B (<70 y at baseline, γ₁₃₁ = −3.41 ± 1.23, P = 0.006). Overall, among men, and for Age_base ≥70 y, lower alcohol intake compared with moderate consumption was associated with poorer performance on the DS-B (overall, γ₀₃₁ = −0.76 ± 0.28, P = 0.008). However, and particularly among men, lower alcohol intake compared with moderate alcohol consumption was linked to slower decline on that test over time (men, γ₁₃₁ = +0.19 ± 0.08, P = 0.014).

Discussion

We examined cross-sectional and longitudinal relations of caffeine and alcohol intakes and nutrient adequacy with cognitive performance in the BLSA. Outcomes included 10 cognitive test scores spanning the domains of global cognition, verbal memory, visual memory/visuo-constructive ability, verbal fluency, attention, working memory, and executive function. Caffeine and alcohol intakes and the NAS were estimated from 7-d food diaries. Using time-interval, mixed-effect regression models, with baseline defined as the earliest available visit with dietary and cognitive data, caffeine intake was associated with better baseline global cognition (MMSE), particularly when baseline age was ≥70 y, independently of key potential confounders. A higher NAS was associated with slower decline or faster improvement on a test of attention (DS-F, for women), and with better baseline performance on immediate and delayed recall for verbal memory (CVLT-List A and DR for participants aged ≥70 y at baseline). A higher NAS was also associated with better baseline performance on global cognition overall among women and among participants aged <70 y at baseline. Alcohol intake, on the other hand, was associated with slower improvement on letter fluency (VFT-L) and global cognition among those aged <70 y at baseline. Conversely, alcohol intake was associated with better attention (DS-F) and working memory (DS-B) performance, particularly among men and individuals ≥70 y at baseline. Some nonlinear associations were found, with moderate alcohol consumption only showing a beneficial effect on baseline DS-B (a measure of working memory), specifically when compared with lower intakes. However, longitudinal associations indicated that alcohol has potentially deleterious effects over time with lower intake being a better choice than moderate intake.

Caffeine and alcohol consumption and the NAS have been associated with cognition in some studies, with mixed findings with respect to the associations’ directionality. Although 2 cross-sectional studies found habitual caffeine intake to be linked with better cognitive or long-term memory performance (12, 17), 2 others failed to detect an association (19, 20). However, using data from the same cohort as in a previous study (12), after a 6-y follow-up no association was found (15). The Longitudinal Lothian Birth Cohort 1936 Study found a potential neuroprotective effect of caffeine intake, but only for coffee (83). Two other longitudinal studies reported such effects of caffeine intake in older women, but not men (9, 21). In contrast, inverse J-shaped associations between coffee consumption and 10-y cognitive declines were seen in elderly men, with the least decline occurring for men consuming 3 cups of coffee per day in the Finland, Italy, and the Netherlands Elderly (FINE) cohort (16). However, among cohort studies, the Finnish Twin Cohort Study failed to detect an association after a 28-y follow-up (18). Protective associations between tea consumption, another source of caffeine, and cognitive decline have been demonstrated in Chinese adults (10, 11). The results of our study suggest that there might be potential acute beneficial effects of caffeine on global cognition, but not other domains. Despite no sex differences, age differentials were significant whereby the putative beneficial effect on global cognition was more noticeable among older adults aged ≥70 y.

Among studies on alcohol intake and cognition (22–51), the Rotterdam Study (36) found that past alcohol consumption was predictive of speed and flexibility in a U-shaped manner, with the best performance among those drinking 1 to 4 glasses per day, particularly women, as was found in other studies (22–35, 37–39). However, a linear dose-response relation has also been shown, sometimes with differences by sex (26). One cohort study found that overall, moderate consumption was protective against poor cognitive function, but that the reverse was true among ApoE4⁺ individuals (22). This effect modification was not found in another study (43). Slower memory decline with increased alcohol consumption in men was found in one study, although the opposite relation was found in the case of psychomotor speed among women (44). A cross-sectional positive relation between alcohol intake and memory was noted in one study with both men and women (45). However, heavy alcohol use has also been linked to poorer cognitive outcomes (34, 46–48). Finally, few studies found no association between alcohol consumption and cognitive outcomes (49–51). Our study detected an association between alcohol intake and faster decline in global cognition and letter fluency, as well as attention and executive function, when comparing moderate consumption with lower intake—findings that were not previously replicated. In contrast, an acutely beneficial effect of alcohol in domains of attention and working memory was found in our study, which shows that alcohol can potentially alter cognitive trajectories and cross-sectional performance differently across different domains.

Individual nutrients were shown to affect cognition with the most widely studied ones being n–3 fatty acids (84–86), some B vitamins (87–91), and antioxidants (92–94), all potentially protective against cognitive impairment. Recent studies are beginning to explore how complete dietary patterns may improve cognitive performance (52, 54, 58) and slow age-related cognitive decline (2, 55, 57), particularly diets high in fruits, vegetables, nuts, unsaturated fats from fish or olive oil, and whole grain breads/cereals, and low in red and processed meats, high fat dairy, and desserts. Consistent with previous studies, we observed that a nutrient-adequate diet was associated with better performance on global cognition, particularly among those aged <70 y at baseline.

Among studies examining dietary quality or patterns in relation to cognition (2, 52–58), limited research has explored specific cognitive domains (i.e., memory, executive function, attention). Nevertheless, some studies suggest that diets with adequate nutrients are associated with better verbal memory (58) and selective attention (52), which is consistent with our findings of better baseline performance on verbal memory and slower rates of decline on verbal memory and attention. In fact, rodent models suggest that higher intakes of saturated fats and simple carbohydrates are associated with neurophysiologic changes (i.e., insulin signaling, synaptic plasticity, and neurogenesis) in the hippocampus and hippocampal-dependent learning and memory (95).

Our investigation has many strengths, which include the use of a large and long-term prospective cohort study with repeated measurements on dietary intake and a comprehensive battery of cognitive performance, allowing us to assess effects of baseline diet on baseline cognitive performance and on cognitive change over time. Specifically, associations of caffeine and alcohol intake and NAS with cognitive performance over time were examined while controlling for key potential confounders, including each of those 3 exposures and socio-demographic and lifestyle factors. Moreover, use of advanced statistical techniques such as time-interval, mixed-effects linear regression models is a major study strength.

Our findings, however, should be interpreted with caution in light of several limitations. First, the BLSA is an open-cohort study of participants selected as a convenience sample, with continuous recruitment and dropout throughout the follow-up. Second, sample selectivity was noted whereby the final analytic sample differed from the original eligible BLSA cohort. To reduce selection biases, we used a 2-stage Heckman selection model (80). Third, although observation frequency for dietary intakes and cognitive function was adequate, data structure was largely unbalanced, given that first-visit age and duration between visits varied across participants. Consequently, we used time-interval, mixed-effects linear regression models, assuming missingness at random (79). Fourth, other covariates such as cardiovascular risk factors were not considered given their potential mediating effects between diet and cognition. Additionally, chance findings may be caused by the number of hypotheses being tested and the subgroup. However, for the most part, we considered those hypotheses to be independent, given that cognitive domains included in each test were distinctive. Our analyses adjusted for multiple testing accounting for multiplicity of exposures. Finally, residual confounding and selection bias could explain some positive findings and low power may underlie some negative findings.

In conclusion, associations of caffeine and alcohol intake and nutrient adequacy with longitudinal cognitive performance are mixed in this sample of older adults. Consistent with prior studies, potential beneficial effects were found for some measures, but not others, with moderation by baseline age and sex. Although findings for alcohol consumption were mixed, the findings were generally supportive of the idea that a high-quality diet and higher caffeine intake may benefit cognition acutely and even prevent age-related declines in certain cognitive domains, including global cognition, verbal memory, and attention. Further longitudinal studies conducted on larger samples of adults are needed to determine whether dietary factors individually or in combination are modifiers of cognitive trajectories among adults.