Bayesian nonparametric hierarchical meta-regression: a flexible approach for modeling multiple biases when combining studies of varying quality and different types

本项目的目标是提供一种灵活的方法，用于在荟萃分析中联合建模有效性和偏倚，该荟萃分析结合了不同类型研究的结果（例如，随机对照试验，观察性研究）和不同的质量。我们调查如果贝叶斯非参数方法的分层荟萃回归提供了一个更数据驱动和强大的方法对模型的错误设定的荟萃分析。分层元回归模型明确区分了两个子模型：用于处理数据收集过程的子模型（例如，建模内部和外部有效性偏差），以及用于回答研究问题的子模型（例如，有效性、预后）。贝叶斯非参数方法的潜在优势将在这些子模型中的一个或两个进行研究。基于这些方法将开发一个R包。该项目旨在为卫生技术评估做出重大贡献。

The objective of this project is to provide a flexible approach for jointly modeling effectiveness and biases in a meta-analysis that combines results from studies of different types (e.g., RCTs, observational studies) and varying quality. We investigate if Bayesian nonparametric methods of the Hierarchical Meta-Regression provide a more data-driven and robust approach against model misspecification in meta-analysis. The hierarchical meta-regression model explicitly distinguishes two sub-models: a sub-model used to handle the data collection process (e.g., modeling internal and external validity bias), and a sub-model used to answer the research questions (e.g., effectiveness, prognosis). The potential advantage of Bayesian nonparametric methos will be investigated in one or both of these sub-models. An R package will be developed based on these methods. The project aims to make a significant contribution to the assessment of health technologies.