Cross Design Synthesis: Combining Evidence About Antidepressants and Suicidality

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Every scientific investigation takes place in the context of uncertain evidence about a question of interest. We believe that the primary goal of research synthesis is to bring differing opinions concerning such questions to consensus. The focus of our application is the question regarding the evidence for an association between antidepressant use and suicidality and builds on our reanalysis of the FDA meta-analysis of antidepressant use and suicidality in youth (Kaizar, Greenhouse, Kelleher, Seltman 2005). Our work is motivated by the recognition that the existing evidence base that can inform this question is available from both experimental and non-experimental studies, and is neither perfect, in the sense that experiments may be rigorous but restricted, and non-experimental studies more general but may be biased nor complete, in the sense that the available outcomes of interest may be indirect (e.g., suicidal ideation/behavior and not completed suicide). The primary work of this application is the conduct of a research synthesis to help advance our understanding of the relationship among antidepressant use, suicide, and suicidality. Specifically, we will use cross-design synthesis, a relatively new research strategy for combining data from studies with complementary designs, to evaluate and combine evidence from both randomized trials and non-experimental studies. Our primary synthesis will include randomized trials of antidepressants both with and without psychotherapy along with data from at least two larger studies that include antidepressant use. We are particularly concerned with the development and application of methods that will facilitate the combining of evidence from multiple data sources, including methods for adjusting studies for factors that affect their external validity, and methods for exploring the effects of biases that affect internal validity. Our approach will be based on the use of Bayesian hierarchical models. We focus on Bayesian hierarchical models as one solution to the problem of making inferences when syntheses require not only modeling of within and between study heterogeneity, but also the qualitative differences of study types. The Bayesian approach will also help facilitate sensitivity analyses for assessing the robustness of inferences to various model inputs. We will also develop accessible and user friendly statistical software programs to implement the methodology that we develop.