The Cache County Study is a prospective study of the elderly residents of Cache County, Utah[6
]. Briefly, in 1995-96 all residents 65 years or older were invited to participate in an examination of cognitive function (Wave I). Surviving participants were then asked to participate in follow-up examinations in 1998-99 (Wave II) and again in 2003-04 (Wave III). At each examination, participants were administered the 3MS and evaluated for dementia with a multistage assessment. Buccal DNA samples were obtained at the baseline, and APOE
genotypes were determined by restriction enzyme analysis[6
]. The institutional review boards of Utah State University, Duke University, and Johns Hopkins University approved all protocols. Informed consent was obtained from all participants at each of the assessments; spouses or next of kin gave consent when participants were unable to provide it.
At the baseline examination, participants were asked to identify all supplements, prescription drugs, and over-the-counter medications they used in the previous two weeks. This information was corroborated by a visual examination of the medication containers. Participants were then asked to provide detailed information about when the medications or supplements were started and the frequency and duration of use. Participants were considered users of vitamin E supplements, vitamin C supplements, or non-aspirin NSAIDs if they reported taking one of these agents four times or more a week for a month or longer. Participants were also considered users of vitamin E or vitamin C supplements if they reported similar use of a multivitamin preparation that contained at least 400 International Units of vitamin E or 500 milligrams of vitamin C.
A total of 5,092 elderly individuals from Cache County (90% of those eligible) participated in the baseline examination. We set aside 356 prevalent cases of dementia at baseline, 1,324 other participants who had only one 3MS evaluation, and another 36 who lacked complete data on medication use. Thus, 3,376 were included in the current analyses. Those not included were older (mean years of age: 78.4 [SD=7.6] vs 74.1 [SD=6.5]; p<0.001), less educated (mean years of education: 12.7 [SD=2.9] vs 13.4 [SD=2.9]; p<0.001) and more likely to be male (proportion: 46.9% vs 41.4%; p<0.001).
We compared changes on 3MS scores over time among five mutually exclusive groups of participants: 1) non-users of vitamin E, vitamin C, and NSAIDs (non-users
), 2) users of NSAIDs but not vitamin E or C (NSAIDs alone
), 3) users of vitamin E and C but not NSAIDs (Vit E and C alone
), 4) users of vitamin E or C (but not both) with or without NSAIDs (Vit E or C +/- NSAIDs
), and 5) users of vitamin E and C and NSAIDs (combined users; Vit E and C + NSAIDs
). First, we calculated change scores between Wave I and Wave III and tested for differences in the means for each group compared to the reference group of non-users using t-tests. Then, we used random effects models to examine the complete data on change over time while controlling for important potential confounders. Such models accommodate fixed and random effects that capture individual differences in 3MS performance over time and account for the correlation in repeated measures[4
]. In these analyses, the mutually exclusive user groups were captured as dummy-coded variables (with non-users as the reference), and time was operationalized as a nominal variable (0, 3, and 8 years for the mean observation points at the three waves) in linear and quadratic forms to account for curvilinear trajectories. Interaction terms between time and the user groups were constructed and models with and without these terms were compared using likelihood ratio tests. Because the interaction terms between the user groups and quadratic time were not significant in the models tested, only the interactions with linear time were retained. Parameterized in this way, the main effect terms for the user groups provide estimates of mean differences in 3MS scores at baseline between each and the reference group of non-users, while the interaction terms provide estimates of the differences in mean rate of change on the 3MS over time.
To assess whether these relationships differed by APOE genotype, we stratified the sample by the presence or absence of one or more ε4 alleles and estimated the models in the sub-strata. All models controlled for other factors found to be significantly associated with baseline 3MS including age, sex, education, APOE status, and history of diabetes and stroke.