Statistical Methods and Software for Multivariate Meta-analysis

多元Meta分析的统计方法与软件 主要研究者：Haitao Chu，M.D.，博士 总结 比较有效性研究（CER）的目的是告知有关利益的医疗保健决策， 不同的预防策略、诊断工具和治疗方案的风险。荟萃分析（MA）是一种 一种统计方法，将多项独立研究的结果结合起来，以提高统计功效， 减少个别研究中的某些偏差。MA也有能力对比不同研究的结果 并确定这些结果之间的模式和分歧来源。虽然MA的许多统计方法 虽然已经提出并进行了调查，但仍然存在重要的研究空白。越来越多的预防 战略、评估工具和治疗方案，以及快速 成本的上升，产生了在临床实践中同时比较多种选择的需要， 创新和严格的多元MA方法。 根据NIH数据科学战略计划和国家医学图书馆的优先领域， “异构数据类型的集成”，响应PA-18-484，本提案的总体目标是开发 尖端的统计方法，以提高MA的可重复性、效率和可推广性，以及 开发易于使用的软件。具体而言，在本建议中，我们将：（1）检查偏度的性能 的标准化偏差的量化发表偏倚在单变量MA，并制定方法量化 多变量MA中的发表偏倚;（2）开发贝叶斯分层汇总接收器操作 特征（HSROC）网络荟萃分析框架，用于同时比较多种诊断 （3）建立一个解释多变量MA随机化后变量的因果推理框架; 以及（4）开发开源、跨平台、公开可用和易于使用的软件（包括R软件包 和SAS宏）来实现所提出的MA方法。 我们将评估这些建议的方法与现有的MA方法的优点和缺点 使用许多真实的案例研究和广泛的模拟研究。拟议的统计方法将广泛 适用于Meta分析。完成这四个目标将直接受益于CER证据库， 在用户友好的软件中实现的最先进的方法，包括R包和SAS宏， 将免费提供给公众。它将通过促进预防、诊断和 治疗癌症和心血管疾病、传染病和其他疾病。

Statistical Methods and Software for Multivariate Meta-analysis Principal Investigator: Haitao Chu, M.D., Ph.D. Summary Comparative effectiveness research (CER) aims to inform health care decisions concerning the benefits and risks of different prevention strategies, diagnostic instruments and treatment options. A meta-analysis (MA) is a statistical method that combines results of multiple independent studies to improve statistical power and to reduce certain biases within individual studies. MA also has the capacity to contrast results from different studies and identify patterns and sources of disagreement among those results. While many statistical methods for MA have been proposed and investigated, important research gaps remain. The increasing number of prevention strategies, assessment instruments and treatment options for a given disease condition, as well as the rapid escalation in costs, have generated a need to simultaneously compare multiple options in clinical practice using innovative and rigorous multivariate MA methods. Following the NIH strategic plan for data science and the National Library of Medicine priority area on “integration of heterogeneous data types”, in response to PA-18-484, this proposal's overall goal is to develop cutting-edge statistical methods to enhance the reproducibility, efficiency and generalizability of MA, as well as to develop easy-to-use software. Specifically, in this proposal, we will: (1) examine the performance of skewness of the standardized deviates for quantifying publication bias in univariate MA, and develop methods quantifying publication bias in multivariate MA; (2) develop a Bayesian hierarchical summary receiver operating characteristic (HSROC) network meta-analysis framework for simultaneously comparing multiple diagnostic tests; (3) develop a causal inference framework accounting for post-randomization variables in multivariate MA; and (4) develop open-source, cross-platform, publicly available and easy-to-use software (including R packages and SAS macros) to implement the proposed MA methods. We will evaluate the strengths and weaknesses of these proposed methods versus existing MA methods using many real case studies and extensive simulation studies. The proposed statistical methods will be broadly applicable to meta-analysis. Completing these four aims will directly benefit the CER evidence base by providing state-of-the-art methods implemented in user-friendly software including R packages and SAS macros, which will be made freely available to the public. It will improve public health by facilitating prevention, diagnosis, and treatment of cancers and cardiovascular, infectious, and other diseases.