The proposed research develops statistical models that describe longitudinal patterns of functional status and clinical symptoms. Data used for these analyses were collected as part of 8 completed studies of depression and treatment for depression in a primary care setting. These individual studies used overlapping measures of depressive symptoms and functional status, and assessed patients for up to two years following treatment initiation. The proposed research will increase our understanding of the time course of depression outcomes and the longitudinal relationship between clinical and functional outcomes. These analyses will also provide important information about variation in individuals' longitudinal patterns of response.
Aim 1 : Develop models that describe longitudinal clinical and functional outcomes: The primary focus of the proposed work is parametric modeling of longitudinal outcomes using hierarchical modeling approaches. Multivariate models that simultaneously describe longitudinal functional and clinical outcomes and associations between these outcomes will build on models derived for univariate longitudinal outcomes. The structure of our hierarchical models will be based on naturally occurring clusters in the data. The Level I model describes longitudinal outcomes within individuals. The Level II model describes variability in longitudinal patterns across individuals, within studies. Multivariate models will incorporate correlation between clinical and functional outcomes within the Level II model. The Level III model describes variability in individuals' expected longitudinal patterns across studies, capturing the metaanalytic component of the analyses. Model development will require careful attention to choice of distributions across all levels of the model hierarchy.
Aim 2 : Derive informative model summaries: The hierarchical models will be used to answer clinically meaningful questions about the course of depression following treatment. We will estimate the expected time to stabilization of functional and clinical outcomes and the level of depressive symptoms and functional impairment once stabilization has occurred. Finally, we will examine synchrony of change in clinical and functional outcomes. Methods: We will analyze primary data across studies. Preliminary univariate modeling using clustered data methods will guide development of hierarchical models. Final hierarchical models will be estimated using simulation-based methods implemented using SAS-IML. A key component of univariate modeling is Multivariate models will build on these univariate models. Population-average estimates will be calculated using Monte-Carlo integration.
