Assessment of Treatment Effects in High-Dimensional, Routine Care Claims Data

描述（由申请人提供）：关于药物比较有效性的数据有限，因为几乎没有头对头试验，而且大多数试验并不代表一般人群或现实世界的实践。使用非随机医疗数据进行的比较有效性研究可以为常规护理中药物和程序的有效性和安全性提供关键证据。虽然在大型电子医疗保健数据库中进行的此类研究可以为常规护理中的药物效应提供快速且成本较低的证据，但依赖于少量干扰素指定混杂因素的传统混杂调整方法通常无法产生无偏倚的结果。近年来，有三种新的方法有望克服这些限制。然而，在真实的世界比较有效性研究中，从未对它们的性能进行过比较：1）将联合收割机声明数据研究与亚群中的详细临床调查相结合，并通过倾向评分校准（PSC）使用此信息改善混杂控制。 2)根据经验识别并优先考虑大量潜在混杂因素，并使用倾向评分方法进行调整，称为高维倾向评分调整（hd-PS）。 3)识别医疗保健系统中影响患者特征（例如处方者偏好）以外的治疗选择的准随机过程，并应用工具变量分析（IVA）。 虽然经济学家早就知道，IVA是相当新的比较有效性研究。 三分之一的美国成年人患有心血管疾病，这种疾病的住院总费用约占美国医院护理总费用的四分之一。缺乏良好的比较有效性信息是改善护理的一个重大限制。将在三项心血管示例研究中测试3种新方法的性能，包括（a）Vytorin与单独使用他汀类药物，（B）MI后高强度与低强度他汀类药物治疗，以及（c）经皮冠状动脉介入治疗期间抗凝治疗的短期和中期有效性。具体而言，我们将：目标1：使用相关心血管示例研究在比较有效性研究中实施三种新型方法来改善混杂控制，目标2：比较三种方法的性能并改进其实施，目标3：传播方法并为免费分析软件和结果库提供互联网支持。 该项目将有意义地改进方法，使用广泛的医疗保健数据库进行心血管医学的比较有效性研究。该项目完成后，将在一个交互式门户网站上提供一个经验证的算法库，该门户网站支持应用程序，并保存补充结果。 公共卫生相关性：纵向医疗保健数据库的使用是一个潜在的强大工具，以评估心血管药物的比较有效性，在常规护理中使用。然而，传统的混杂因素调整方法，依赖于有限数量的特定的协变量，往往无法产生无偏的结果。我们将实施并严格评估三种新的分析方法，以增强对常用药物有效性和安全性的因果解释。

DESCRIPTION (provided by applicant): Data on the comparative effectiveness of medications are limited because few head-to-head trials are available and most of them do not represent the general population or real-world practice. Comparative effectiveness research using non-randomized healthcare data can provide critical evidence on the effectiveness and safety of medications and procedures in routine care. While such studies in large electronic healthcare databases can provide expedited and less costly evidence on drug effects in routine care, the conventional confounding adjustment methods that rely on a small number of investigator-specified confounders often fail to produce unbiased results. In recent years three novel methodologies have shown promise to overcome these limitations. Their performance, however, has never been compared with each other in real world comparative effectiveness studies: 1) Combine claims data studies with detailed clinical surveys in subpopulations and use this information for improved confounding control through Propensity Score Calibration (PSC). 2) Empirically identify and prioritize very large numbers of potential confounders and adjusting for them using propensity score methods, called high-dimensional propensity score adjustment (hd-PS). 3) Identify a quasi-random process in the healthcare system that influences treatment choice beyond patient characteristics, e.g. prescriber preference, and apply instrumental variable analysis (IVA). Although long know to economists, IVA is fairly new to comparative effectiveness research. One in three American adults has a cardiovascular condition and the total inpatient cost for such conditions approximates one fourth of the total cost of hospital care in the US. The lack of good comparative effectiveness information is a significant limitation for improving care. The performance of the 3 novel approaches will be tested in three cardiovascular example studies, including (a) Vytorin vs. statin use alone, (b) high vs. low intensity statin therapy after MI, and (c) short and medium-term effectiveness of anticoagulation therapies during percutaneous coronary interventions. Specifically, we will: Aim 1: Implement three novel approaches to improved confounding control in comparative effectiveness research using relevant cardiovascular example studies, Aim 2: Compare performance of the three approaches and improve their implementation, Aim 3: Disseminate methods and provide internet support for free analysis software and result libraries. This project will meaningfully improve methodologies for comparative effectiveness research in cardiovascular medicine using a wide array of healthcare databases. After completion of this project a library of validated algorithms will be available on an interactive web-portal that supports applications and is a depository of supplemental results. PUBLIC HEALTH RELEVANCE: The use of longitudinal healthcare databases is a potentially powerful tool to evaluate the comparative effectiveness of cardiovascular medications as used in routine care. However, conventional confounder adjustment methods that rely on a limited number of investigator-specific covariates often fail to produce unbiased results. We will implement and rigorously evaluate three novel analytic methods to enhance causal interpretation of the effectiveness and safety of commonly used medications.