Investigating the association of vitamin D seasonality on inflammatory and hemostatic markers

Abstract

Seasonal variations in health outcomes are commonly used to hypothesize a link with nutritional vitamin D (25-hydroxyvitamin D, 25(OH)D) status. The majority of vitamin D intake is from skin exposure to sunlight and varies seasonally in countries at a distance away from the Equator. However, despite the strong seasonality of vitamin D intake, no statistical method using cyclical patterns has been proposed to deduce an association between 25(OH)D and health indicators. Our motivation was to overcome the influence of related confounders, such as obesity, between 25(OH)D and health indicators: obesity would be expected to have little or no effect on the seasonal variations in 25(OH)D and in five inflammatory/hemostatic health outcomes (fibrinogen, tissue plasminogen activator [tPA], von Willebrand factor [vWF], C-reactive protein [CRP], and D-dimer). The data analyzed was from the 1958 British birth cohort biomedical survey (n = 6195) and the biomarkers were ascertained from blood drawn over an 18-mo period. We used mediation analysis to determine whether the seasonal variations of the outcomes were mediated by 25(OH)D to infer an association. The assumptions of mediation analysis fit naturally into the study’s cross-sectional setting, where day of year of blood collection is the independent variable transformed by the harmonic function, and 25(OH)D is the mediator of the seasonal variation of the outcomes. The harmonic terms were tested to establish the presence of seasonal variation in the outcomes and 25(OH)D in order to determine whether the statistical mediation test could be applied. The data were collected over an 18-mo period and assayed in multiple batches to measure the serum biomarkers. When the assay batches were modeled as fixed effects, significant correlation was found with date of when blood was drawn. Thus, variation in assay batches was accounted for as random effects terms on the intercept in linear mixed-effects models. Inferences were based on tests from mediation analysis defined by the product of regression coefficients; we extended this test to allow for harmonic functions with multiple frequencies in order to statistically test the mediated effect through 25(OH)D. This was done using parametric bootstrap when the models were run in the Frequentist setting. We also replicated the analyses in the Bayesian setting to ascertain the change in amplitude of the seasonal variation that was due to 25(OH)D. Out of the five health outcomes, three (tPA, D-dimer, and fibrinogen) had significant seasonal associations that were partially mediated through 25(OH)D, one (vWF) had a seasonal pattern not mediated through 25(OH)D, and finally another (CRP) had no significant seasonal pattern. The association of 25(OH)D was strongest for tPA, and less so for D-dimer and fibrinogen. Our results and adaption of the mediation test show that there is broad potential in using seasonal variations of health indicators to deduce an association that may have not been affected by nonseasonal confounding.

• 25(OH)D
• bayesian analysis
• harmonic function
• hemostatic
• inflammation
• mediation analysis
• product of coefficients test
• seasonal variations

Acknowledgements

We are grateful to all the participants in the 1958BC. We acknowledge Professor Ian Gibb, Dr. Steve Turner, and Marie-Claude Fawcett (Royal Victoria Infirmary, Newcastle-upon-Tyne, UK) for carrying out the laboratory assays and the Centre for Longitudinal studies, Institute of Education (original data producers), for providing the data.