Nutrition and Functional Status

Bias from unmeasured confounding is a challenge in observational gerontology research. Vitamin D research exemplifies the problem: results on extra-skeletal outcomes from observational studies of serum 25-hydroxyvitamin D [25(OH)D] concentrations are often not replicated in randomized vitamin D supplementation trials. Unmeasured factors such as sunlight exposure and skin color are the presumed culprit, motivating the National Institute on Aging to issue a request for applications for vitamin D supplement trials of functional outcomes. However, determining health effects of 25(OH)D and those from supplements are fundamentally different research questions. Since 25(OH)D concentrations cannot be randomized, observational studies are still key to answering this question. To address unmeasured confounding, two statistical methods to examine associations of 25(OH)D with depressive symptoms are described and applied. The first method uses season of blood draw in a modified instrumental variable analysis that overcomes the survival bias present in Mendelian randomization studies. Using data from 3,147 person-visits from the InCHIANTI study, a prospective cohort of older Italian adults, low 25(OH)D (< 20 ng/mL) was associated with 4.62 (95% confidence interval[CI] 3.25, 6.56) times greater risk of depressed mood (Center for Epidemiologic Studies Depression [CES-D] scale >= 16), as compared with 25(OH)D >= 20 ng/mL. The second method is a joint mixed-effects model that leverages repeated 25(OH)D measurements in longitudinal studies. Low 25(OH)D at three visits was associated with 1.80 (95% CI 0.18, 3.42) lower CES-D scores than low 25(OH)D at no visits.