Determinants of Vitamin D Levels

All statistical analyses were performed using SAS V.9.4 (SAS Institute, Cary, NC, USA). The variables were assessed for normality and log transformed where relevant. Mean differences between sexes for continuous variables were measured by independent samples t-test and analysis of variance; and Pearson χ² test for categorical variables. We performed univariable linear regression analysis to explore the association between explanatory variables and serum 25(OH)D₂, 25(OH)D₃ and total 25(OH)D concentrations. We log transformed 25(OH)D₂, 25(OH)D₃ and 25(OH)D, and expressed these on standardised scales (z-scores). To examine whether sex was an effect modifier of associations, an interaction term (sex × explanatory variable) was additionally included in univariable analyses. We conducted multivariable analyses aiming to examine mutually adjusted associations of different exposures with 25(OH)D₂, 25(OH)D₃ and 25(OH)D measures, namely season of blood sampling (low and high sunlight period), latitude, BMI, waist circumference, SEP, smoking status, alcohol consumption, leisure time computer use, physical activity, diet score and contraception status. In addition, we examined serum 25(OH)D₂, 25(OH)D₃ and 25(OH)D concentrations by excluding women using OCPs.
Following examination of the determinants associated with 25(OH)D₂, 25(OH)D₃ and 25(OH)D concentrations, we performed multinomial ordinal logistic regression analysis to assess the risk factors associated with being in the lower tertile (reference: tertile III) of vitamin D. Owing to equivocal definitions of cut-off values for vitamin D status in the general population, we categorised the analysis sample into tertiles of 25(OH)D. Statistical significance was set at global p<0.05 using two-tailed test.