Joint modeling of continuous and binary data in meta-analysis

点击翻译按钮获取中文摘要

Joint Modeling of Continuous and Binary Data in Meta-Analysis Principal Investigator: Lifeng Lin, Ph.D. Summary Systematic reviews and meta-analyses are critical tools for comparative effectiveness research and evidence- based medicine. They combine and contrast research findings from individual studies and derive a form of evi- dence to underpin guidelines and aid medical decision making. In meta-analysis practice, researchers fre- quently face studies that report the same outcome differently, such as a continuous variable (e.g., scores for rating depression) or a binary variable (e.g., counts of patients diagnosed with depression). To combine these two types of studies in the same analysis, a simple conversion method has been widely used to handle stand- ardized mean differences and odds ratios. However, this may be inaccurate when effect sizes are large or cut- off values for dichotomizing binary events are extreme (leading to rare events). In addition, this conversion method is built under the conventional framework of meta-analysis, where study-specific effect sizes (e.g., log odds ratios) are approximated to normal distributions and within-study variances are treated as fixed, known values. These assumptions may not be appropriate in some situations (e.g., small sample sizes) and could produce misleading meta-analysis conclusions. With advances in statistical computing, the exact distributions of effect measures could be properly analyzed, and the uncertainties in within-study variances could be fully incorporated in meta-results. In response to PA-20-200, this proposal aims at developing cutting-edge statistical methods for combining con- tinuous and binary outcome data and thus improving the efficiency and generalizability of meta-analysis. In this project, we will: develop Bayesian hierarchical models to jointly synthesize continuous and binary effect measures; use extensive simulation studies and high-quality real-world datasets to evaluate the performance of the proposed methods; and develop user-friendly, open-source software (including R packages and SAS macros) to implement the proposed methods. Specifically, this project will evaluate the strengths and weak- nesses of the developed models using meta-analyses on various outcomes such as depression. The proposed methods are also broadly applicable for many other diseases, including cancers, infectious diseases, among others. The simulation studies will be carefully designed and conducted so that they cover a wide range of set- tings, with respect to the number of studies with continuous and binary outcomes, sample sizes, event rates, heterogeneity, etc. Our developed user-friendly, open-source software will include detailed instructions and worked examples, so that practitioners can easily and accurately apply the proposed methods to clinical prac- tice. The output of this project will directly improve comparative effectiveness research and evidence-based medicine on diverse medical topics.