Advancing personalized medicine in PD using harmonized multi-site clinical data

项目摘要 在神经系统疾病中，增长最快的是帕金森病（PD），超过了阿尔茨海默病。 放松。PD表现为异质性临床综合征，临床表型的这种变异性突出了 需要调整治疗的类型和/或剂量，以适应个人生活的具体和不断变化的需求， PD的。个体化或精确医学的主要目标是利用患者的特征来确定 个性化治疗策略（ITS），以促进健康。由于PD的复杂性质， 表型异质性，制定成功的个性化医疗方法是一个复杂的问题， 这可能会从更多的数据驱动方法中贝内。开发可靠的信息技术系统的挑战之一是 分析需要相当大的样本量和纵向评估的主题超过一个 相对较长的时间。数据集还必须包括各种处方模式，以允许分析 学习不同治疗顺序（策略）效果的方法。这些重要要求排除了 研究人员使用来自单个临床研究的数据来构建数据驱动的ITS。 现有的帕金森病对症药物治疗指南最好被描述为“允许的”。的相对 由于缺乏不同类别药物的比较证据， 遵循任何特定的治疗策略。我们通过提出两阶段研究来填补这一重要空白。第一个 阶段（R61）的重点是通过整合来自六项临床试验的数据， PPMI观察性研究共涉及4，705名患者，随访时间为23.5至96个月。到 据我们所知，这种全面的数据协调以前在PD中没有做过，它可以 也为今后的研究提供了很好的信息来源。在第二阶段（R33），我们将利用 协调数据集，以针对多个临床结局开发高质量的PD ITS，包括 在24个月和48个月随访时，测量了ADL评分、生活质量和Schwab和England（SE）ADL。 具体而言，R33阶段的目标是（目标1）比较常用的药物类别序列， PD;（目的2）确定最佳个体化治疗策略，以告知药物类别的最佳顺序， 警局在追求这些目标的过程中，我们将提出一个强大的、严格的和计算效率高的统计机器。 学习方法，用于构建PD的数据驱动的最佳ITS。该提案扩大了现有的 通过放松某些不切实际的假设和使用灵活的建模来开发ITS的方法 技术（例如，机器学习方法），同时保持有效的统计推断。这些新方法 将被集成到易于使用的R语言公开软件中（目标3）。这将最大化 其他研究人员采用所提出的方法，并允许研究人员分析其他PD 数据集，目标是构建PD的ITS。此外，由于这些方法不是疾病特异性的， 我们的方法和软件将使其他疾病的类似探索成为可能。

Project Summary Among neurological disorders, the fastest growing is now Parkinson's disease (PD), surpassing Alzheimer's dis- ease. PD manifests as a heterogeneous clinical syndrome and this variability in the clinical phenotype highlights the need to tailor the type and/or the dosage of treatment to the specific and changing needs of individuals living with PD. The main goal of individualized, or precision, medicine is to use patient characteristics to determine an individualized treatment strategy (ITS) to promote wellness. Due to the complex nature of PD coupled with phenotypic heterogeneity, formulating successful individualized approaches to medical care is a complex prob- lem that may benefit from a more data-driven approach. One of the challenges in developing reliable ITSs is that the analyses require studies with fairly large sample sizes and longitudinal assessment of subjects over a relatively long period of time. The data set must also include various prescribing patterns to allow the analytic method to learn the effects of different treatment sequences (strategies). These important requirements preclude investigators from using data from a single clinical study to construct data-driven ITSs. Existing guidelines for symptomatic drug therapy for PD can best be described as "permissive". The relative lack of comparative evidence for different classes of drugs has created challenges in devising recommendations to follow any specific therapeutic strategy. We fill this important gap by proposing a two phase study. The first phase (R61) focuses on creating a harmonized and curated dataset by integrating data from six clinical trials and the PPMI observational study that, in aggregate, involved 4,705 patients followed from 23.5 to 96 months. To the best of our knowledge, such comprehensive data harmonization has not been done before in PD and it can provide an excellent source of information for future studies as well. In the second phase (R33), we will leverage the harmonized data set to develop high quality ITSs for PD with respect to several clinical outcomes including UPDRS score, quality of life, and Schwab and England (SE) ADL measured at 24 and 48 months of follow-up. Specifically, the goals of the R33 phase are to (Aim 1) compare commonly used sequences of drug classes for PD; (Aim 2) identify the best individualized treatment strategies to inform optimal sequences of drug classes for PD. In pursuit of these aims, we will propose robust, rigorous and computationally efficient statistical machine learning methods for constructing data-driven optimal ITSs for PD. The proposal expands the scope of existing methods in developing ITSs by relaxing certain unrealistic assumptions and through the use of flexible modeling techniques (e.g., machine learning methods) while maintaining valid statistical inference. These new methods will be integrated into easy-to-use, publicly available software in the R language (Aim 3). This will maximize the adoption of the proposed methodology by other investigators and allow researchers to analyze other PD datasets with a goal of constructing an ITS for PD. Furthermore, because the methods are not disease-specific, our methods and software will enable similar exploration for other diseases.