INTRODUCTION Cognitive impairment, a principal cause for functional disability among the elderly, can lead to dementing illness over time mainly in the forms of Alzheimer's Disease (AD) and vascular dementia (VaD). In fact, AD prevalence is expected to rise, reaching 100 million worldwide by 2050, with 1 in 85 persons potentially living with AD.(1) Thus, uncovering modifiable risk factors that would prevent or delay cognitive impairment is important.
The neuroprotective effects of antioxidant nutrients (e.g. vitamins E) and B-vitamins (e.g. folate) have been at the forefront of cognitive aging and nutritional epidemiology research over the past two decades.(2) Vitamin D's role in preserving cognitive function with aging has recently gained attention in epidemiological investigations.(3) Its public health significance lies in the fact that vitamin D deficiency \[25-hydroxyvitamin D3 (25(OH)D)\<11 ng/ml (\<27.5 nmol/L)\] is a highly prevalent condition, particularly among the poor and among African-Americans.(4, 5) Vitamin D is a steroid hormone with its primary function being to regulate body levels of calcium, phosphorus and bone mineralization. While sunlight exposure is its primary source through skin synthesis from 7-dehydrocholesterol, dietary and supplemental intakes of vitamin D play a key role in its overall status. (3) The active form of vitamin D3, namely 1,25-dihydroxy vitamin D3 influences many metabolic pathways through genomic and nongenomic actions which help maintain and stabilize intracellular signaling pathways involved in memory and cognitive function.. (3, 6, 7) The neuroprotective role of vitamin D may be mediated through vasculo-protection and preservation of neurons partly through the expression of neurotrophins and other neurotransmitters which help protect against cognitive dysfunction, through the suppression of inflammatory cytokines.(3, 5, 8) Vitamin D can also down-regulate receptors in memory-relevant regions and enhance amyloid phagocytosis and clearance. (8) Serum 25(OH)D and dietary vitamin D were shown to influence cognitive outcomes in large epidemiological studies. (9-34) This study will examine associations between vitamin D status and intake with longitudinal change in various domains of cognition among a large sample of ethnically and socio-economically diverse US urban adults. It also explores those associations systematically across sex/age groups and race. We hypothesize that both vitamin D status and intake are associated with slower decline in domain-specific cognitive performance over time, but perhaps differentially by age/sex and race.
MATERIALS AND METHODS Database The Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study is a prospective cohort study initiated in 2004 that focused on the cardiovascular and cognitive health of an ethnically and socio-economically diverse urban population. Specifically, it used area probability sampling to recruit a socioeconomically diverse sample of African American and White urban adults (30-64 years old) residing in thirteen neighborhoods of Baltimore, Maryland.(35) Written informed consent was obtained from participants who were also provided with a protocol booklet and a video that explains study procedures. The study protocol was approved by the National Institute on Environmental Health Sciences Institutional Review Board of the National Institutes of Health. Data for the present study were derived from baseline visit 1 (2004-2009) and the first follow-up examination (visit 2; 2009-2013). Follow-up time ranged from \<1y to \~8y, with a mean of 4.64±0.93y.
Study sample HANDLS initially recruited 3,720 participants (Phase I, visit 1). Given that only Phase II had in-depth data including biochemical indices and cognitive performance measures, 25(OH)D was available for 1,981 participants at baseline. The corresponding sample size for dietary and supplemental vitamin D were N=2,177 and N=2,159, respectively. Complete and reliable cognitive tests at each visit varied in sample size as well. Further, the final analytic sample was determined based on exposure and covariate non-missingness at baseline and outcome non-missingness at either visit. Figure 1 describes sample selection for all exposures. The final analytic sizes ranged between N=1,231 and N=1,803 with k=1.5-1.9 observation/participant.
Cognitive assessment Cognitive performance was assessed with 7 tests yielding 11 test scores and covering 7 domains (Global, attention, learning/memory, executive function, visuo-spatial/visuo-construction ability, psychomotor speed, language/verbal): the Mini-Mental State Examination (MMSE), the California Verbal Learning Test (CVLT) immediate (List A) and Delayed Free Recall (DFR), Digit Span Forward and Backwards tests (DS-F and DS-B), the Benton Visual Retention Test (BVRT), Animal Fluency test (AF), Brief Test of Attention (BTA), Trails A and B and the Clock Drawing Test (CDT) (Supplemental method 1). All participants were judged capable of informed consent and were probed for their understanding of the protocol. Although no formal dementia diagnosis was conducted, all participants were given mental status tests, which they completed successfully. In every case, low mental status performance was due to low literacy level without any sign of dementia.
Vitamin D status Total 25(OH)D (in ng/ml) was measured using immunoassay at baseline and follow-up visits. The collected sample was \~0.8 ml of preferably fasting serum which was refrigerated and transported to the lab for analysis. Visit 1 analyses were conducted at the Massachusetts General Hospital.(36) Visit 2 analyses were conducted by Quest Diagnostics, Chantilly, VA.
Dietary vitamin D Dietary factors included in our analyses were measured at the baseline visit. Baseline 24-hour dietary recalls were obtained using the US Department of Agriculture (USDA) Automated Multiple Pass Method, a computerized structured interview.(37) Measurement aids used included measuring cups, spoons, ruler and an illustrated Food Model Booklet. Two recalls were administered in-person by trained interviewers, 4-10 days apart. Trained nutrition professionals used Survey Net, matching foods consumed with 8-digit codes from the Food and Nutrient Database for Dietary Studies (FNDDS) version 3.0,(38) and My pyramid equivalents database for food groups (MPED 2: http://www.ars.usda.gov/SP2UserFiles/Place/80400530/pdf/mped/mped2\_doc.pdf). Dietary vitamin D was among the nutrients made available by the FNDDS, from which daily values could be estimated and expressed in µg/d, using the average from the two 24-hr recalls conducted at baseline.
Supplemental vitamin D The HANDLS dietary supplement questionnaire was adapted from the 2007 NHANES instrument.(39) HANDLS participants provided supplement bottles during their dietary interview at the follow-up visit only (i.e. visit 2). Information on Over-The-Counter (OTC) vitamin and mineral supplements, antacids, prescription supplements, and botanicals were reported, and supplement users were asked about dose strength, dose amount consumed, length of supplement use (converted to days), frequency of use (daily, monthly, seasonally, annually), and if each supplement was taken the day prior to interview.
A HANDLS dietary supplement database was developed by trained nutritionists and registered dietitians. This database consisted of 4 files integrated to generate daily intake of each nutrient consumed by a dietary supplement user. \[See detailed description at the HANDLS study website: https://handls.nih.gov/\]. Vitamin D supplemental intake was ascertained for visit 1 if the daily amount (IU/d) was non-zero at visit 2 and the length of time for intake was greater or equal than the length of time (days) between the two visits, per individual. Thus, HANDLS participant was either 0: non-vitamin D containing supplement user at baseline or follow-up, 1: vitamin-D containing supplement user at baseline and during follow-up, 2: vitamin-D containing supplement user during follow-up only.
Covariates Covariates included age, sex, race (White vs. African American), marital status, educational attainment (\<High School (HS); HS, \>HS), poverty income ratio (PIR\<125% for "poor"), measured body mass index (BMI, kg/m2), opiate, marijuana or cocaine use ("current" vs. "never or former"), smoking status ("current" vs. "never or former") and the Wide Range Achievement Test (WRAT) letter and word reading subtotal scores to measure literacy. (See Supplemental Method 1) To assess depressive symptoms with focus on affective, depressed mood, the 20-item CES-D was used. Baseline CES-D total score was included in the analysis as a potential confounder in the association between vitamin D exposures and cognitive change or baseline performance. (See Supplemental Method 1) The Healthy Eating Index (HEI-2010) total score, based on two 24-hr recalls administered at baseline, was used as a measure of overall dietary quality. See steps for calculating HEI-2010 at http://appliedresearch.cancer.gov/tools/hei/tools.html and http://handls.nih.gov/06Coll-dataDoc.html. Further, season of baseline MRV exam was used as proxy for sunshine exposure and was included as covariate in all models. Finally, self-reported history of type 2 diabetes, hypertension, cardiovascular disease (stroke, congestive heart failure, non-fatal myocardial infarction or atrial fibrillation) and dyslipidemia at first-visit were considered as covariates.(40) Statistical analysis Using Stata 15.0 (41) and accounting for sampling weights, population estimates of means and proportions were obtained. Means across stratifying variables (e.g. age/sex or race) were compared using svy:reg, relationship between categorical variables using svy:tab and design-based F-tests. Further, mixed-effects regression models with 11 continuous cognitive test score(s) as alternative outcomes were conducted. In these models the time variable was interacted with several covariates including the main exposure variables, namely VITDserum, VITDdiet and VITDsuppl. The models assume missingness at random with time points ranging between \~1.5-2.0 visits/person. (42) Predictive margins were estimated and plotted across TIME, stratifying by exposure group, from selected mixed-effects regression models, particularly those showing significant associations in the total population.
Moderating effect of sex and age groups was tested by adding interaction terms to separate multivariable mixed-effects regressions (3-way and 4-way interaction terms between Time, exposure, Age group and sex) and stratifying by sex and age group to examine relationships among the following groups: (1) Younger men, (2) Older men, (3) Younger women, (4) Older women, whenever at least one 4-way interaction was deemed statistically significant. Further, moderating effects by race were also examined using a similar approach \[(1) Whites, (2) African-Americans\] (Supplemental method 2), given the well-known higher prevalence of vitamin D deficiency among African-Americans compared with Whites and the differential rates of increases in vitamin D status recently shown by age, sex and race groups.(43) Variable time of follow-up is accounted for in the mixed-effects regression model as annual rate of change in the outcome was of primary interest.
Moreover, selection bias may occur due to non-random selection of participants with complete data from the target study population. Thus, in each mixed-effect regression model, a 2-stage Heckman selection process was conducted, by running a probit model to compute an inverse mills ratio at the first stage (derived from the predicted probability of being selected, conditional on the covariates in the probit model, mainly baseline age, sex, race, poverty status and education). At the second stage, this inverse mills ratio was then entered as a covariate in the final mixed-effects regression model, as was done in a previous study.(44) In all analyses, a type I error of 0.05 was considered for main effects whereas a p\<0.10 was deemed significant for interaction terms,(45), prior to correcting for multiple testing. A familywise Bonferroni procedure was used to correct for multiple testing by accounting only for cognitive tests and assuming that exposures related to separate substantive hypotheses.(46) Therefore, for main effects, p\<0.004 (0.05/11) was considered significant. Due to their lower statistical power compared to main effects, 2-way interaction terms had their critical p-values reduced to (0.10/11=0.009), while 3-way and 4-way interaction terms had their critical p-value reduced to 0.05. A similar approach was adopted in two other studies. (47, 48)
