Bayesian Methods for Comparative Effectiveness Research with Observational Data

 描述（由申请人提供）：比较有效性研究（CER）依赖于对快速扩展的观察数据的分析，这些数据是由于医疗保健服务的日益整合、电子病历系统的传播和临床登记数据的开发而成为可能的。这些数据既为旨在提高医疗保健价值的研究提供了非凡的机会，也为有意义的调查带来了新的挑战。一个关键障碍是缺乏健全的统计方法和工具，无法解决观察性研究中估计治疗效果的多个方面 当治疗有效性可能在亚群之间变化时，定义这些亚群的协变量信息是高维的，有时无法测量。目标1开发了用于大规模观测数据因果推断的新贝叶斯方法，以1）估计平均因果效应，解释选择测量混杂因素时的模型不确定性，2）估计亚群中的平均因果效应，解释选择亚群时的不确定性。这种新提出的方法概括了现有的方法，因为它不依赖于单个模型的规格，而是通过对多个模型进行平均来估计参数。目标2：开发新的贝叶斯方法，用于在存在未测量的混杂因素的情况下评估治疗效果，这些混杂因素在大型观察数据中调节治疗效果。新方法使用工具变量来1）确定基本因果参数的分布，而不是平均值，2）通过系统地放松选择偏差假设将因果参数与亚组联系起来。目标3将新方法应用于观察性研究，以提供医疗器械，外科手术和药物治疗领域的新的和完全可重复的知识。目标4：开发灵活、高效、健壮、文档齐全、用户友好的R库和SAS宏，促进我们新开发方法的传播。我们的新方法，它们在大型管理和临床登记数据中的应用，以及它们的传播，将使整个研究界能够以最高的方法学严谨性来解决现代CER问题。

 DESCRIPTION (provided by applicant): Comparative effectiveness research (CER) relies upon the analysis of a rapidly expanding universe of observational data made possible by the growing integration of health care delivery, the dissemination of electronic medical records systems, and the development of clinical registries data. These data present both extraordinary opportunities for research aimed at improving value in health care as well as new challenges for meaningful investigation. A critical barrier relates to the lack of sound statistical methods and tools that can address the multiple facets of estimating treatment effects in observational studies when treatment effectiveness may vary across subpopulations and covariate information defining these subgroups is high dimensional and sometimes unmeasured. Aim 1 develops new Bayesian methods for causal inference in large observational data to 1) estimate average causal effects accounting for model uncertainty in the selection of measured confounders and 2) estimate average causal effects in sub-populations accounting for uncertainty in the selection of the subgroups. This newly proposed approach generalizes existing methods because it will not rely on the specification of a single model, but instead will estimate parameters by averaging across several models. Aim 2 develops new Bayesian methods for assessing treatment effects in the presence of unmeasured confounders that moderate treatment effects in large observational data. The new approach uses instrumental variables to 1) identify the distributions, rather than means, of essential causal parameters and 2) link causal parameters to subgroups by systematically relaxing selection bias assumptions. Aim 3 applies the new methods to observational studies to provide new and fully reproducible knowledge in the areas of medical devices, surgical procedures, and pharmaceutical treatments. Aim 4 develops flexible, efficient, robust, well documented, user-friendly R libraries and SAS macros, facilitatin dissemination of our newly developed methods. Our new methods, their applications to large administrative and clinical registry data, and their dissemination will allow the entire research community to address modern CER questions with the highest methodological rigor.