System construction for population pharmacokinetic and pharmacodynamic modeling using electronic health records: toward precision medicine

项目摘要 为了改善患者的治疗效果，必须确定能够减少治疗变异性的因素 和量化。对于药物治疗，药代动力学（PK）和 药效学（PD）是变异性的主要来源。已证明群体PK/PD研究 这对确定这些因素非常有用，但未得到充分利用。这种方法的主要障碍是 难以从大量患者中获得数据，这需要精确的时间信息， 给药和药物血浆浓度或反应测量的准确值 用于PK/PD研究。电子健康记录（EHR）可能是一个很好的来源， 这些数据，但用于PK/PD建模的数据提取和预处理的标准化方法 缺乏我们将致力于开发EHR数据的抽象、验证和预处理方法， 构建PK/PD研究数据。在目标1中，我们将开发数据预处理和构建程序 PK/PD数据集，并使用测试数据集验证开发的程序。在目标2中，我们将评估 通过执行临界点分析，使用EHR数据构建系统的稳健性， 故意将错误引入测试数据集。这种方法得到了食品和药物管理局的高度推荐， 药品管理局确定分析对处理缺失数据方法的敏感性。在目标3中， 我们将基于贝叶斯PK/PD，开发他克莫司的初步剂量优化算法， 预测模型，这将作为未来研究的临床支持决策工具的基础。一个更 数据构建的高效和标准化系统将促进基于人群的PK/PD研究， 提供PK/PD数据管道。这项工作的最终产品将是一个可推广和验证的 使用EHR数据分析药物暴露和反应的数据管道，这将广泛 适用于一系列药物的许多人群研究。我们经过验证的系统将扩展 为更广泛的研究群体提供机会，以进行基于人群的PK/PD研究，并促进 发现影响PK/PD特征的因素的机会。这项研究将用于提高精度 一系列的治疗方法。

Project Summary   In order to improve patient outcomes, factors that can reduce variability in treatment must be identified and quantified. For drug therapies, individual differences in pharmacokinetics (PK) and pharmacodynamics (PD) are major sources of variability. Population PK/PD studies have been proven to be very useful to identify these factors, but are underutilized. A major barrier to this approach is the difficulty of obtaining data from a large number of patients, which requires precise time information and accurate values for dosing and drug plasma concentration or response measurements appropriately formatted for PK/PD studies. Electronic health records (EHRs) are potentially an excellent source for such data, but standardized methods for data extraction and preprocessing for PK/PD modeling are lacking. We will work to develop methods for abstraction, validation and preprocessing of EHR data to construct data for PK/PD studies. In Aim 1, we will develop programs for data preprocessing and building PK/PD datasets, and validate the developed programs using the test datasets. In Aim 2, we will evaluate the robustness of data construction system using EHRs by performing tipping point analyses by intentionally introducing errors into test datasets. This approach is highly recommended by the Food and Drug Administration to determine the sensitivity of analysis to methods of handling missing data. In Aim 3, we will develop a preliminary dose optimization algorithm for tacrolimus based on a Bayesian PK/PD prediction model, which will serve as a basis for a clinical support decision tool for future study. A more efficient and standardized system for data construction will promote population based PK/PD studies by providing a PK/PD data pipeline. The ultimate product of this work will be a generalizable and validated data pipeline for analysis of drug exposures and responses using EHR data, which will be widely applicable to many population studies for an array of medications. Our validated system will extend opportunities to a wider research community to perform population based PK/PD studies and facilitate the chance of finding factors affecting PK/PD profiles. This research will be utilized to advance precision medicine across a wide array of therapies.