Estimating Long-Term Disease Trajectories from Short-Term Data

 描述(由申请人提供)：阿尔茨海默氏症的临床试验设计主要以对来自预先指定的组的短期数据的零散分析为指导。我们通过对评估分布的横断面检查来构想这些群体。然后我们对这些组的纵向变化进行表征，以计算功率和样本量。这种方法在为分期试验建立实用的纳入标准方面是有效的，但结果标准可能是不必要的过程。捕捉疾病全过程的新分析方法将更有效地利用多变量纵向数据，允许更复杂和更具包容性的纳入标准，并为针对特定干预措施量身定做的最佳人群提供新的见解。更好地了解疾病动态的长期连续体将使我们能够发现我们目前理解中的差距，并提供一种全面的、以数据为导向的方法来设计临床试验。支持AD临床试验设计这一新方法的分析方法还不发达，在技术上具有挑战性。我们将开发、实施和解释新的统计方法，以准确和有效地表征从临床前到痴呆的阿尔茨海默病标记物的长期动态。我们将把我们的统计方法应用于现有的短期生物标记物数据集，以优化阿尔茨海默氏症临床试验设计，并改善个性化的预后预测。该项目将通过更准确地描述长期疾病动态，识别风险因素和理想的临床试验人群和结果指标，并提供个性化的预后预测，对我们在阿尔茨海默氏症方面的进展产生重大影响。具体目标1：开发用于模拟长期AD动态的分析框架。我们将开发一种新的分层贝叶斯潜变量模型，从短期采样数据中识别长期多变量疾病轨迹。基本模型将被扩展到使用动态贝叶斯信念网来模拟样本变化、疾病异质性和复杂的网络中介结构。具体目标2：开发优化的临床试验研究设计和个性化预后。根据目标1开发的模型将被用来为特定的机械性生物标记物效应确定研究人群和结果测量的最佳组合，并获得对病程和治疗效益的个性化预测。我们将开发和验证使用我们的模型进行相关临床试验和个性化预后预测所需的机制。具体目标3：开发软件和Web应用程序。将开发实施AIMS 1-2中开发的方法的计算机软件，并作为免费R包分发。我们将构建一个交互式Web应用程序，以便临床医生和临床试验人员可以探索、可视化和询问我们的分析结果，以预测预后并模拟临床试验情景。该应用程序将使临床试验人员能够基于我们的模型和他们对干预的生物标记物效应的先验知识来优化他们的试验设计。

 DESCRIPTION (provided by applicant): Alzheimer's clinical trial design is predominantly guided by fragmented analyses on short-term data from pre- specified groups. We conceive these groups by cross-sectional examination of the distribution of assessments. We then characterize longitudinal changes in these groups to calculate power and sample size. This approach is effective in establishing practical inclusion criteria for staging trials, but the resuting criteria may be unnecessarily course. Novel analytical methods that capture the full continuum of the disease will make more efficient use of multivariate longitudinal data, allow more sophisticated and inclusive inclusion criteria, and provide new insights into optimal populations tailored to specific interventions. A better understanding of the long-term continuum of disease dynamics will allow us to discover gaps in our current understanding and provide a comprehensive, data-driven approach to designing clinical trials. Analytic methods to support this new approach to AD clinical trial design are underdeveloped and technically challenging. We will develop, implement, and interpret novel statistical methods to accurately and efficiently characterize the long-term dynamics of Alzheimer's disease markers from preclinical to dementia. We will apply our statistical methods to existing short-term biomarker datasets to optimize Alzheimer's clinical trial design and improve personalized prognostic predictions. This project will have a substantial impact on our progress with Alzheimer's by more accurately characterizing long-term disease dynamics; identifying risk factors and ideal clinical trial populations and outcome measures; and providing personalized prognostic predictions. Specific Aim 1: To Develop an Analytic Framework for Modeling Long-Term AD Dynamics. We will develop a novel hierarchical Bayesian latent variable model to discern long-term multivariate disease trajectories from short-term sampled data. The base model will be extended to model sampling variation, disease heterogeneity, and complex network mediation structure using Dynamic Bayesian Belief Nets. Specific Aim 2: To Develop Optimized Clinical Trial Study Designs and Personalized Prognoses. The models developed under Aim 1 will be leveraged to determine the optimal combinations of study population and outcome measure for specific mechanistic biomarker effects, and to obtain personalized predictions of disease course and therapeutic benefit. We will develop and validate the machinery necessary to use our models to make relevant clinical trial and personalized prognostic predictions. Specific Aim 3: To Develop Software and Web Application. Computer software implementing the methods developed in Aims 1-2 will be developed and distributed as a free R package. We will build an interactive web application so that clinicians and clinical trialists can explore, visualize, and interrogate our analysis results to predict prognoses and simulate clinical trial scenarios. The application will enable clinical trialists to optimize their trial designs based on our model and their prior knowledge of their intervention's biomarker effects.