Bayesian Methods and Software for Patient-Centered Network Meta-Analysis of Binar

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Comparative effectiveness research (CER) relies fundamentally on accurate assessment of treatment efficacy and safety that, ideally, can be tailored to specific patients. The growing number of treatment options for a given condition, as well as the rapid escalation in their costs, has generated an increasing need for scientifically rigorous simultaneous comparisons of multiple treatments in clinical practice. Also called mixed or multiple treatments meta-analysis, network meta-analysis (NMA) expands the scope of a conventional pairwise meta-analysis by simultaneously analyzing both direct comparisons of interventions within randomized controlled trials and indirect comparisons across trials .... Compared to traditional meta-analysis of just two treatments, NMA presents many additional statistical challenges. In particular, a typical randomized trial compares only a few (typically tw) treatments, which intrinsically creates a large amount of missing data when, say, a dozen treatments must be compared simultaneously, since the outcomes for treatments not studied in a particular trial are missing by design. Currently available statistical methods, which are based on treatment contrasts, focus only on relative treatment effect estimates and have other serious limitations. The overall goal of this proposal is to develop cutting-edge statistical methods, and to integrate them into publicly available, easy-to-use software, to enhance patient-centered NMA. Specifically, we will develop multivariate Bayesian hierarchical models for binary outcomes from the perspective of missing data methods with the following three specific aims: 1) to extend our preliminary work on estimating patient-centered parameters (e.g., absolute risk, risk difference and relative risk) with a single endpoint to allow non-ignorable missingness; 2) to simultaneously model multiple endpoints (e.g. outcomes for efficacy and safety) with proper consideration of non-ignorable missingness; and 3) to incorporate individual patient characteristics. In addition, we propose new methods to measure and detect inconsistency between the direct and indirect evidence, and to borrow strength cautiously from less reliable data sources. We propose to perform empirical assessment of the strengths and weaknesses of these methods through many real data applications and simulations. Completion of the three aims will substantially advance CER analytical methods for comparing multiple treatments across multiple endpoints and tailored to patient characteristics.