Study area and study design
This cross-sectional study was conducted in the Kattankulathur (latitude: 12° 81′ 10”N, 80° 03′ 05″E) block of Kancheepuram district in Tamil Nadu, India. The block comprises 39 villages with 133 habitations with a total population of 213,850 as enumerated in 2012–2013. Study participants were apparently healthy adult men and women living in these villages. The inclusion criteria for the study were that individuals should be healthy, aged 18 years or more, residing in the study area for at least 1 year prior to the study, and willing to provide written consent to participate in the study. Those with self-reported coronary artery disease, arthritis, or cancer were excluded. The study was undertaken over a 14 month period from June 2015 to July 2016. A modified 30 × 7 sampling method was used [20, 21]. Assuming a community VDD prevalence of 50% (a conservative estimate derived from published data from India [19]) a sample size of 420 was calculated. The sample size calculation formula was z2*p*q/d2 where the standardized score for 95% confidence level (Z) =1.96, p = 0.5, q = 1-p = 0.5 & d = 0.5, to estimate a VDD prevalence of 50% with a ± 5% error margin. Using a 30-cluster sampling, the sample size was calculated as 14 adults per cluster or a total of 420 adults. Thirty of the 133 habitations in Kattankulathur were selected, and the houses to be sampled were chosen accordingly.
Data collection and sampling
Data collection was undertaken between June to September 2015. Participants were selected in the study area as follows. The first participating household was selected using the Expanded Programme on Immunization (EPI) recommendation [20], i.e., the centre of the area corresponding to the selected habitation was reached and a direction was selected after spinning a bottle and following the direction in which the cap pointed. The first house in that direction was selected as the first household and the first eligible adult encountered in that house was recruited. Subsequent households were selected by following the EPI strategy of going to the household whose door was nearest to the current household until 14 eligible participants accrued in that habitation.
An interview schedule was created. Informed written consent was obtained from each of the participants. Each participant was interviewed directly by the first author using a predesigned form (Additional file 1). Data was collected on demographic profile and socioeconomic status (age, sex, marital status, education, income, occupation, and the following life style factors: Duration of sun exposure, body surface area exposed to sunlight, use of sunscreen while going outdoors, intake of fatty fish, beef, liver and milk. Participants were asked to recall the average duration spent outdoors during weekdays and the weekend in the past 1 month and this was categorized as 15–30 min, 30–60 min, 1–2 h and > 2 h per day [22]. The body surface exposed to sunlight when going outdoors was categorized as fully covered (only hands and feet exposed), wearing short sleeves (exposing hands, arms and forearms), wearing short sleeved T-shirt and shorts or dhoti (a single cloth tied around the waist and hitched up to expose the legs) or only wearing shorts or dhoti (which also exposed the torso) [22]. Data on average intake of fatty fish, beef, liver (from poultry or cattle) and milk were obtained by asking participants to recall their intake in the 2 weeks prior to interview. Data were also obtained on use of multivitamin supplements containing vitamin D, and general awareness about vitamin D. The latter was verified by asking whether the participant had heard of vitamin D, whether he or she had learnt anything about vitamin D in school, through television, or through promotional literature distributed by health organizations or in advertisements. In addition, we asked questions on birth (urban versus rural), source of drinking water for the household, use of boiled or bottled water at home, availability of tap water at home, and availability of a close toilet system at home. This latter set of questions as aimed at elucidating information related to hygiene factors that are believed to influence the composition of the gut microbiome of an individual. Data were assigned into categories for analysis. Socioeconomic status was assigned using a modified Kuppuswamy score which was calculated based on the level of education, occupation and income of the head of the household [16].
Sample collection and vitamin D assay
Venous blood samples were collected at home from consenting participants using Vacutainers and were transported to the laboratory within an hour. Samples were immediately centrifuged to separate the serum which was stored at -20 °C. Total 25-OH vitamin D was measured in duplicate by ELISA using the DIAsource 25-OH Vitamin D Total kit (Catalog No. KAP1971, DIAsource Immunoassays SA, Louvain-la Neuve, Belgium), which uses monoclonal antibodies to measure 25-OH D2 and 25-OH D3 and has been certified by the Vitamin D External Quality Assessment Scheme. The intra-assay coefficient of variation (range 2.7–7.8%) and the inter-assay coefficient of variation (range 4.7–9.4%) for this assay were both less 10%. Appropriate controls and calibrators (provided with each kit) were used with each ELISA plate to generate standard curves. Based on the level, an individual was classified as vitamin D deficient if serum level was < 12 ng/mL, vitamin D insufficient if 12–20 ng/mL, and vitamin D replete or sufficient if > 20 [13]. The vitamin D estimation was done at the GI Research Lab, SIMS Hospitals, Vadapalani, Chennai, by an investigator who was blinded to the participant characteristics.
Statistical analysis
Data were entered in MS-Excel and analyzed using IBM SPSS version 20. In univariate analysis, Pearson’s chi-square test was used to determine association of the variables with vitamin D status. P values < 0.05 were considered as statistically significant. Variables that were associated with vitamin D status on univariate analysis with P values less than 0.10 (this P value was selected in order to limit the number of variables in the model) were entered into an ordinal logistic regression analysis in which vitamin D status was the dependent variable with the ordinal categories ordered as deficient, insufficient, and sufficient. The independent variables introduced into the model as factors were sun exposure time per day, SES group, gender and age group, consumption of fish and consumption of milk; educational qualification, occupation, birth place, awareness of VDD, source of drinking water, boiled drinking water, tap water available at home and closed toilet system at home were included as covariates. There were many cells with small observed and predicted counts, and the goodness-of-fit statistic showed a Pearson Chi-square of 848.087 and Deviance Chi-square of 766.370 (df 807. Since there were cells with small observed counts, the overall model fitting test showed a − 2 log likelihood of 841.124 for intercept only compared to 772.151 for the final model, leading to a Chi-square of 68.973 (df 11) and a significance < 0.0001, indicating that the model with the predictors was valid. The proportional odds assumption of ordinal logistic regression analysis was tested using the test of parallel lines. The test showed a − 2 log likelihood of 752.48 for the null hypothesis model compared to 728.942 for the general model, leading to a Chi-square of 23.538 (df 19) and a significance of 0.214 thus affirming the null hypothesis that the proportional odds were the same across response categories. Odds ratios and 95% confidence intervals are presented for the variables that showed independent association with vitamin D status.