Analysis of Multiple Informant Data in Psychiatry

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): We will develop statistical methodology to address several key issues in studies designed to assess treatment effects on mental health outcomes in service-based or trial-based samples. In our past work (MH54693) we developed a general approach to the analysis of multiple informant data arising from a wide variety of mental health studies. This research led to the development of more efficient methods for the analysis of multiple informant outcomes or multiple informant predictors in both cross-sectional and longitudinal studies. These methods have also been extended to handle partially observed informant reports and cases where informant outcomes are non-commensurate, i.e., out- comes measured on different scales or representing more than one construct. A logical next step in this research endeavor is the development of more powerful tests of treatment effects and improved statistical methods for understanding the causes of treatments on multiple non-commensurate mental health outcomes. The two most common approaches to this problem are single testing of a composite outcome or separate testing of each constituent outcome. However these approaches are unbiased and efficient only in the rare situation of complete data. We propose to develop and illustrate a framework for joint testing of multiple non-commensurate outcomes. This will include developing and evaluating methods applicable to cross-sectional and longitudinal studies allowing for incomplete observations. We will compare these methods to approaches using separate models for each outcome, composite endpoints and global tests. These methods will be extended to observational settings using causal inference methodologies. We will apply these methods to cohorts of patients with depression, schizophrenia, and bipolar disorder. These diseases exert significant social, personal, and economic costs, and multiple outcomes are often used to assess treatment effectiveness. The methods that we will develop will permit a more precise understanding of the causal effects of treatments and more powerful tests of treatment effects. While this application focuses on treatments, our methods are broadly applicable to other settings involving multiple outcomes including genetic-based association studies. Major depression, bipolar disorder and schizophrenia exert significant personal, social, and economic costs, and there is a pressing need to improve methodology to assess treatment effects for prevention and intervention studies. This proposal will develop statistical methodology to improve the analysis of psychiatric clinical trials and observational studies with multiple outcomes. These novel and innovative methods will improve the assessment of treatment effects and will contribute to improving the health of the population.