The NIH-AARP cohort was established in 1995–1996 by the National Cancer Institute to investigate the roles of diet and lifestyle in cancer etiology (7
). Cohort participants included 566,402 AARP members aged 50–71 years in 1995–1996 from six states (California, Florida, North Carolina, Pennsylvania, New Jersey, and Louisiana) and two metropolitan areas (Atlanta and Detroit). All study participants completed a comprehensive dietary survey that included a 124-item food-frequency questionnaire and a short survey on demographics, medications, and lifestyle (7
). A follow-up questionnaire was mailed out to surviving participants of the original cohort in 2004–2006 to update exposures and to ascertain the occurrences of major chronic diseases, including diabetes. A total of 318,261 participants responded to the follow-up survey and were therefore eligible for the present study. We excluded 21,632 participants who provided no or inconsistent information on multivitamin use and 31,383 participants with missing values on diabetes diagnosis. Because vitamin supplement use was assessed in 1995–1996, to reduce the possibility that diabetes itself might have affected health behaviors, we further excluded 33,239 diabetic patients who reported a diabetes diagnosis before 2000. Therefore, the final analytic sample included 232,007 participants, 217,877 without diabetes and 14,130 diabetic case subjects diagnosed after 2000.
Assessment of vitamin supplement use
As part of the baseline dietary survey, participants were asked whether they took any vitamins or minerals in the past 12 months with three categorical choices (no, less than once per month, and once or more per month). For those who took vitamins once or more per month, we further asked for the frequencies of use for three types of multivitamins (stress-tab type, therapeutic or theragran type, and one-a-day type). Five categorical answers were allowed for each of these multivitamins (times/week): never, less than one, one to three, four to six, and daily. In addition to multivitamins, we also asked for the use of vitamin A, β-carotene, vitamin C, vitamin E, and calcium (including Tums) with the same categorical frequencies. Finally, participants were asked whether they took several other individual vitamins and minerals more than once per month, including iron, zinc, selenium, and folic acid. For individual vitamins and minerals, we instructed participants not to include vitamins and minerals contained in multivitamins that they had reported in the multivitamin section.
In addition, the baseline survey also collected data on basic demographics and lifestyle such as date of birth, sex, race, education level, marital status, smoking habit, and physical activity. Consumption of coffee and alcohol and dietary intakes of calories and micronutrients were derived from the food-frequency questionnaire. Finally, participants were asked to self-evaluate their health status as excellent, very good, good, fair, or poor and to report weight in pounds (0.45 kg) and height in inches (2.54 cm). BMI was calculated as weight in kilograms divided by the square of height in meters (kg/m2).
Ascertainment of diabetes
A question on lifetime occurrence of physician-diagnosed diabetes was first asked at the dietary survey in 1995–1996. This information was updated on the follow-up questionnaire in 2004–2006, along with the year of first diagnosis in the following categories: before 1985, 1985–1994, 1995–1999, or 2000 to present. These questions did not differentiate type 2 from type 1 diabetes; however, in adults, ~90–95% of all diagnosed diabetes is type 2 diabetes. Because the current study included only older adults and adults with incident cases of diabetes diagnosed after 2000, we believe that most of diabetes cases in the current analysis should be type 2 diabetes.
To minimize the possibility that diabetic patients modified their lifestyle around or after the diagnosis, the current analysis only included incident diabetes cases diagnosed after 2000, at least 4–5 years after the exposure assessment. The frequency for overall multivitamin use was calculated by adding up the use of individual types of multivitamins. In the analysis, the frequency of supplement use was defined as (times/week) never, less than one, one to three, four to six, and daily. No frequency data were collected for supplemental uses of iron, zinc, selenium, and folic acid, and, thus, only yes/no variables were included in the analysis. We performed t tests or ANOVA for continuous covariates and χ2 tests for categorical covariates to formally test the differences of covariates between nonusers and all users as well as among four categories of users only. Multivariate odd ratios (ORs) and 95% CIs were derived from logistic regression models. Potential confounders included age in year, sex, race (whites versus nonwhites), education level (<8 years, 8–11 years, 12 years or completed high school, post–high school or some college, college and postgraduate), marital status (married or living as married, widowed, divorced, separated, or never married), smoking status (never smokers; past smokers with years since last smoking: ≤35, 30–34, 20–29, 10–19, or 1–9; or current smokers with the numbers of cigarettes/day: 1–10, 11–20, or >20), coffee consumption (cups/day: none, less than one, one, two to three, or more than three), alcohol consumption (drinks/day: 0, <1, 1–1.9, 2–2.9, or ≥3), general health status (excellent, very good, good, fair, or poor), BMI (in kg/m2: 12.0–24.9, 25.0–29.9, or ≥30.0), physical activity (never/rarely, one to three times per month, one to two times per week, three to four times per week, and five or more times per week), and total energy intake (quintiles). Multivitamin users were also likely to use individual vitamin supplements, and considerable amounts of micronutrients form individual supplements may affect our primary analyses on multivitamin use. We therefore conducted further analysis adjusting for the supplemental uses of individual vitamins and minerals. The statistical significance for a linear trend was tested by assigning a value to each exposure category (0 for nonusers, 0.5 for less than one time per week, two for one to three times per week, five for four to six times per week, and seven for daily users) and including it as a continuous variable in the logistic regression model. The analysis was first conducted for all study participants and then stratified by age (years: <60, 60–64, and ≥65), sex (men versus women), general health status (excellent or very good, good, and fair or poor), smoking status (never versus ever), BMI (in kg/m2: 12.0–24.9, 25.0–29.9, or ≥30.0), and physical activity (never/rarely, more than once per month). When possible, detailed values of the stratifying variables were adjusted along with all other confounders to minimize the possibility of residual confounding.
Our primary analysis focused on multivitamin supplements. However, we also examined the use of individual vitamins and minerals in relation to diabetes. These analyses showed that vitamin C and calcium supplement use was associated with a lower risk of diabetes. Therefore, we conducted post hoc analysis to examine whether these associations were modified by the use of multivitamins (users versus nonusers) or dietary intakes of corresponding nutrients from foods (above versus below medians). All statistical analysis was performed by using SAS release 9.1 (SAS Institute, Cary, NC), and the significance tests were two tailed with α = 0.05.