An Inter-Method Reliability Study Comparing Interview Information on Sunlight, Tanning Beds, Food, and Supplement Exposures with Serum 25-Hydroxy Vitamin D Levels

Background: We developed a questionnaire designed to capture the vitamin D related exposures of sunlight, tanning bed use, dietary intake, and supplement use, and used an inter-method reliability approach to compare the study method (self-reported information on vitamin D related exposures nearest the blood draw) with serum 25-hydroxy vitamin D (25(OH)D) levels. Methods: This inter-method reliability study included 512 control women from a population-based, case-control study in Alberta, Canada. All women self-reported data on food/supplement intake (average µg/day), sun exposure (cumulative hours/year), and tanning bed exposure (cumulative hours/year) and provided fasting serum samples, measured in duplicate with a DiaSorin immunoassay for 25(OH)D levels. The correlation between participant characteristics and 25(OH)D are described. We used multivariable robust regression to estimate the percent of variation in 25(OH)D explained by our variables of interest. Results: Food intake, sun exposure, and tanning bed exposure had positive and significant correlations with 25(OH)D levels of a similar magnitude (Spearman r= 0.17 to 0.19). Supplement intake (average µg/day, Spearman r= 0.44) had the strongest positive correlation. In both crude and adjusted models (adjusted for age, body mass index, race, smoking, oral contraceptive use, and menopausal status/hormone therapy), we consistently found that food explained 3.1%, supplements 18.9%, sunlight 2.2%, and tanning bed use 3.0% of the variation in 25(OH)D levels, and all variables combined explained 27.5% - 36.0% of the total variation. Conclusions: These results suggest that our comprehensive dietary and light exposure questionnaire may be a reasonable proxy measure of vitamin D status in the recent past when either 25(OH)D measurements or serum samples are not available for study participants. models. These results suggest that our comprehensive dietary and light exposure questionnaire may be a reasonable proxy measure of vitamin D status in the recent past when either 25(OH)D measurements or serum samples are not


Introduction
Vitamin D is an essential nutrient for the human body with reference intakes established by the United States (US) National Academy of Medicine, most recently updated for bone health in 2010(1).
However, unlike other nutrients, it is also synthesized in the skin in response to exposure to ultraviolet B (UVB) radiation (e.g., from sunlight). While vitamin D sufficiency is closely tied with bone 3 health(1), it has also been linked to a number of other disease states from cancer (2) to autoimmune disorders (3) to dementia (4). Along with direct measurement of circulating 25-hydroxy vitamin D (25(OH)D) levels, surveys and questionnaires are often used to estimate vitamin D related exposures such as sunlight exposure, tanning bed use, dietary intake, and supplement use. From both a public health and research standpoint, it is important to consider the relative contribution of these vitamin D related exposures to circulating 25(OH)D levels on an individual level. Such knowledge can be used to develop predictive models for the public concerning vitamin D related exposures that can achieve sufficient vitamin D status while minimizing any harms such as from UVB radiation; many such models have been published (e.g., (5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)). Alternatively, for research purposes, it is useful to evaluate if questionnaire responses, specifically designed to capture vitamin D related exposures, are correlated with circulating 25(OH)D levels.
We developed a questionnaire designed to capture self-reported adult life-time vitamin-D related sun, tanning bed, diet, and supplement exposures for a case-control study (16). Retrospective serum samples from age 20 years forwards for our study participants were not available, but we did have a serum samples as well as self-reported vitamin D related exposures near to the reference date, i.e., an assigned month/year for controls that had the same distribution as the diagnosis dates of ovarian cancer cases. Using an inter-method reliability approach (17) we compared the study method (selfreported information on vitamin D related exposures nearest the blood draw) with serum 25(OH)D levels in the control women only from a population-based, case-control ovarian cancer study in Alberta, Canada.

Study population and data collection
The study population was derived from Alberta control participants who participated in the Ovarian Cancer in Alberta and British Columbia (OVAL-BC) Study, described previously (18). Briefly, in accordance with the Canada Health Act, Alberta has a publicly administered and funded health care systems ensuring eligible residents access to medically necessary hospital and health care services.
All residents register for coverage and the control women were randomly identified through this 4 provincial health roster from September 2005 through June 2011. These participants were: 1) Alberta residents; 2) age 40-79 years; 3) English-speaking; and, 4) able to complete the telephone interview.
Consent was obtained in a two-step process. In the first step, 1,514 eligible women were contacted and 604 (40%) provided active consent for release of their contact information to the study team. In the second step, 519 (86%) of the 604 women provided signed, informed consent, completed the telephone interview and provided a fasting blood sample (two 6 ml red top/serum tubes).
Information on risk factors was ascertained up to the time point of the reference date (month/year) that was assigned for controls as part of the case-control study design (18). The telephone interview was extensive and included personal health history, reproductive and menstrual history, exogenous hormone use, family history of cancer, physical activity patterns (for a subset of women only), adult lifetime caffeine and alcohol consumption, smoking habits, demographic characteristics and adult height and weight at each decade from age 20 years onward. Additionally, four vitamin D related exposures over adult life were ascertained (16) and are described below. For the purposes of this analysis, we used the most recently reported exposures relative to the reference date.

Vitamin D related exposures
For specified ages at ten-year age intervals (20, 30, etc.) through age 70 years women were queried about: calendar year during each age, place of residence, job title or school attended, number of days per week on and off work/school. Although Canada is generally north of 49 0 N, study participants could have lived at any inhabited latitude and were not required to reside in Canada all of their adult lives for this study. Considering both the southern and northern hemispheres, and the sun exposure therein, we used 42 0 N, 42 0 S, the Tropic of Cancer and the Tropic of Capricorn, to split the two hemispheres into five regions: north (> 42 0 N), mid-north (Tropic of Cancer to 42 0 N), equator (Tropic of Cancer to Tropic of Capricorn), mid-south (Tropic of Capricorn to 42 0 S, and south (> 42 0 S). Women were asked about the amount of sunlight exposure received on average on week days and weekend days from 9:00 am to 5:00 pm between April 1 and September 30 when residing in the north region.
Similarly, the amount of sunlight exposure for winter sun holidays/seasonal residences between October 1 and March 31 that occurred in the mid-north, equator or mid-south regions were also recorded. If a woman resided in the mid-north, equator or mid-south regions sunlight exposure was collected for the entire year. If multiple patterns of sunlight exposure occurred at a given age (because of job changes, moves, etc.) each distinct pattern was recorded separately. The cumulative hours of relevant sun exposure per individual was summed over the entire year. For this analysis, we used sunlight exposure for the most recent exposure age (i.e., 40, 50, 60, etc. years of age) that preceded the reference age. For example, for a woman who was 53 years at reference date, her most recent sun exposure was based on the calendar year when she turned 50 years of age. Because blood samples were always collected after the reference date, there was a variable amount of time between the reported sunlight exposures and the date of blood draw. On average this varied by 4.2 years.
Participants also provided information on the consumption of foods/beverages that were vitamin D fortified (all types of milk, margarine) (19) or naturally contain vitamin D (tuna, salmon, oysters, sardines, eggs), and supplements (multivitamins, vitamin D tablets with/without calcium, cod liver oil capsules or liquid) in the last 12 months and for specified ages at ten-year age intervals (20, 30, etc.) through age 70 years. The dietary interview, which we developed specifically to estimate vitamin D intake, was based on the Canadian adaptation of the National Cancer Institute's (NCI) Diet History Questionnaire (20)(21)(22) with slight modifications for our specific foods (e.g., salmon and tuna were asked separately, not as a general fish category). Vitamin D content in food was based on Health Canada nutrient data (23). To estimate average daily dietary vitamin D exposure for each study participant, we summed reported vitamin D intakes (in µg per day) for each food in the 12 months prior to the reference date. For each food item, average daily vitamin D intake was derived as:  (1). Cod liver oil liquid was assigned 100 IU/2.5 µg per gram of oil (24). For each supplement category, average daily vitamin D intake was derived as: (consumption frequency per day) x (vitamin D content) and then summed over all categories of supplements.
To avoid small, sporadic exposures in tanning bed use, women had to have at least 12 exposures in one year since they turned 20 years of age to be considered a tanning bed user. For women who answered affirmatively, lifetime use was ascertained by recording the age started and stopped and hours of exposure per week or month for each pattern of tanning. For this analysis, the cumulative hours of tanning bed exposure per woman was summed during the 12 months prior to the reference date. for absolute agreement using a two-way random effects model. We observed an ICC of 0.94, which is in the excellent range, indicating a high degree of agreement between duplicate measurements. We used the average value of the duplicate serum measurements for each woman in all analyses. All 25(OH)D concentrations are reported in nmol/L (to convert to ng/ml, multiply by 0.4006).

Statistical analysis
The final study population consisted of the 512 women for whom we had interview information and blood samples available. We plotted the distribution of serum 25(OH)D levels, overall and by age and by season. Characteristics of women and the associated 25(OH)D levels are described as well as the p-value for a difference in group means within each characteristic. P-values were calculated using t 7 tests for comparison of two groups and one-way analysis of variance F tests for comparison among three or more groups. Food intake was evaluated both as a continuous variable and grouped in tertiles: low (0.17-2.80 µg/day), medium (2.81-5.07 µg/day) and high (5.08-20.30 µg/day).
Supplement intake was also evaluated as a continuous variable and grouped in tertiles: low (0.0-3.18 µg/day), medium (3.19-15.00 µg/day) and high (15.01-42.50 µg/day). Tanning bed hours and sunlight hours (which included sun holiday hours) were evaluated as continuous variables. We calculated Pearson and Spearman correlation coefficients (r) between serum 25(OH)D levels and food intake, supplement intake, sunlight exposure, and tanning bed exposure. For modeling, we initially used linear least squares regression but outliers violated the assumption of normally distributed residuals. To best accommodate the effects of outliers we performed multivariable robust regression to examine the association between covariates and serum 25(OH)D levels and to estimate the percent of variation in serum 25(OH)D levels that can be explained by food intake, supplement intake, sun exposure, and tanning bed exposure in crude and adjusted models. We implemented Mestimation with Huber weighting in the robust regression (27) to produce estimates of the regression slopes that are robust to outliers by down-weighting the contribution of outliers. Final models were adjusted for age, body mass index (BMI), race, smoking, oral contraceptive (OC) use, and menopausal status/hormone therapy, all which had a borderline to highly statistically significant relationship with serum 25(OH)D levels. We were unable to adjust for physical activity because this was only available for a subset of the women. We also explored interactions by age, BMI, and OC use in white women only using likelihood ratio tests comparing the full model including the interaction term to the reduced model without (Supplemental Table 1). The categories of menopausal status and hormone therapy (HT) use were too sparse to evaluate interactions. All analyses were performed using R Studio version 1.0.136(28). All statistical tests were two-tailed and the significance level was set at α = 0.05. 60.0 nmol/L in all age groups and varied little, with no significant differences between age groups (Table 1). Mean concentrations were over 60 nmol/L in all seasons of blood draw, but was significantly higher in summer (72.7 nmol/L) than in winter (64.3 nmol/L) (p < 0.01). greater sunlight exposure and any tanning bed use (Table 1). Additionally, peri/postmenopausal women who used estrogen plus progesterone and other combinations of hormones had a higher mean serum concentration than the other women classified by menopausal status and hormone use.

Results
Physical activity was only available for a subset of the women, but among this subset, women with an average of 60 or more minutes of physical activity per week had higher mean 25(OH)D concentrations than those with lower levels of activity.
The correlations between dietary and light exposures and 25(OH)D levels are shown in Table 2.
Overall food intake (average µg per day; Pearson r = 0.13 and Spearman r = 0.17), sun exposure    In an exploratory analysis, we also assessed possible interactions between food intake, supplement intake, sunlight exposure, and tanning bed exposure and age, BMI, and OC use in relation to serum 25(OH)D concentrations, in white women only (Supplemental Table 1). Non-whites were excluded from this analysis due to small numbers. While there was some variation across the categories, our We also assessed food intake, sunlight exposure and tanning bed exposure among the 93 women who reported no vitamin D supplement use (Supplemental Table 2). In these women, both food and sunlight exposure explained more of the variation in 25(OH)D levels (~ 5% and ~ 6%, respectively) than in the models that included women who used supplements (~ 3% and ~ 2%, respectively).

Discussion
We found that mean 250H-D concentrations varied significantly across the levels of consumption of Another limitation of this study is that we only had physical activity data for a subset of women (n = 262), but among these women physical activity was significantly related to 25(OH)D levels. We therefore repeated the analysis presented in Table 3 for this subset of women and adjusted for physical activity. The results were very similar to those reported in the final model in Table 3 without adjustment for physical activity, with the partial R 2 for supplements again being the highest (about 18.5%). These results again imply that there is a consistent relationship between 25(OH)D levels and the diet and light variables as measured in our questionnaire regardless of adjustment for other covariates.
There is substantial variation in the ascertainment of vitamin D related foods across studies. We attempted to capture all major sources of vitamin D from food in our study and only included those foods that were consumed by at least 5% of the population in our previous reliability study (16).
Yogurt was not found to be a major source of vitamin D from food but yogurt made with vitamin D fortified milk became more common during our study. In another study in Alberta (35)  Availability of data and materials: The datasets generated and/or analyzed during the current study are not publicly available due to compromising individual privacy, but are available from the corresponding author on reasonable request.
Competing interest:

Not applicable
Funding: