Functional Regression and Subgroup Analysis for Asynchronous Longitudinal Data.

项目总结/摘要 异步纵向设计已广泛应用于衰老和阿尔茨海默病（AD）的研究。在 这些长达数十年的研究，作为纵向变量，通常从不同的平台（如食品）收集 频率问卷、实验室结果和临床或心理测试），获得的数据不同步 即使是同一个主题。作为激励规范老化研究（NAS）的一部分， 有兴趣解决烟酸的饮食摄入是否与AD和认知能力下降有关。在 在这项研究中，在参与后的6年内，对参与者进行了四次认知测试， 营养素摄入量是通过在基线和相邻基线之间分发的食物频率问卷来确定的。 认知测试因此，暴露变量，包括营养摄入量和认知结果， 测试，异步测量。忽视时机错误问题的天真分析将导致有偏见的结果 错误的结论。此外，衰老相关疾病的发展往往在各种（潜伏） 亚群，其中疾病进展与生物标志物和环境暴露之间的关联各不相同 基本上跨越了亚群。老龄化人群中的亚组分析对于更好地开发 预防、治疗和管理衰老相关疾病的策略，如AD。 考虑到NAS数据的特殊性，所提出的方法将有助于解决以下关键问题： 饮食中烟酸摄入量与NAS中AD事件之间的关系。我们还将检查其他生物标志物 和与AD相关的环境暴露。拟议的亚组分析也将有助于确定潜在的 在老龄化人口中的亚组，并辨别亚组特定的风险因素，导致认知障碍。 与现有的依赖于限制性模型假设的工作不同，所提出的方法是完全非参数的， 数据自适应，可以与最先进的模型选择方法相结合，以选择相关的风险因素 从数百个可能随时间变化且与响应异步的变量中提取AD。方法和 本研究所发展之理论将可广泛应用于其他纵向研究。我们还将开发新的 计算工具和软件包，可扩展的大型纵向数据集与长期随访。

Project Summary / Abstract Asynchronous longitudinal designs have been widely used in studies on aging and Alzheimer's disease (AD). In these decades-long studies, as longitudinal variables were routinely collected from different platforms (e.g. food frequency questionnaires, lab results and clinical or psychological tests), the obtained data were not synchronized in time even for the same subject. As part of the motivating Normative Aging Study (NAS), it was of substantial interest to address whether dietary intake of niacin was associated with incident AD and cognitive decline. In the study, four cognitive tests were administered to participants in a span of 6 years after participation, and nutrient intake was determined by food frequency questionnaires distributed at the baseline and between adjacent cognitive tests. As a result, the exposure variables, including nutrient intake, and the outcomes of cognitive tests, were measured asynchronously. Naive analyses ignoring the mistiming issue will lead to biased results and wrong conclusion. In addition, the development of aging-associated diseases tends to differ in various (latent) subpopulations, where associations between disease progression and biomarkers and environmental exposures vary substantially across subpopulations. Subgroup analyses in aging populations are crucial for developing better strategies of preventing, treating and managing aging-associated diseases, such as AD. Accounting for the peculiar nature of the NAS data, the proposed methods will help address the key question on the association between dietary intake of niacin and incident AD in the NAS. We will also examine other biomarkers and environmental exposures associated with AD. The proposed subgroup analysis will also help identify latent subgroups within the aging population and discern subgroup-specific risk factors leading to cognitive impairment. Unlike the existing work relying on restrictive model assumptions, the proposed methods are fully nonparametric, data adaptive and can be combined with state-of-the-art model selection methods to choose relevant risk factors for AD from hundreds of variables that might be time varying and asynchronous with the response. Methods and theory developed in this project will be widely applicable to other longitudinal studies. We will also develop new computation tools and software packages that are scalable for large longitudinal data sets with long followups.