Analysis of High-Dimensional Environmental Health Data

描述(由申请人提供)：本提案的目标是开发环境健康数据的统计方法，当感兴趣的健康影响复杂时。具体目标的动机是在空气中颗粒物对健康的影响的毒理学和环境流行病学研究中出现的问题。该项目的具体目标是开发(I)基于小波的历史功能数据模型，用于评估暴露与健康之间的高维关联；(Ii)分层隐马尔可夫模型，用于分析动物颗粒物集中器研究产生的多变量功能数据；(Iii)解决颗粒物流行病学研究中因空间和时间错位而产生的暴露测量误差的方法。我们将把建议的方法应用于现有分析方法没有充分利用数据的几个数据集，包括(I)来自敏感子群中的物质的半连续心率变异性数据。在激励性的应用中，这些方法将为环境健康研究中的两个科学紧迫问题提供洞察力：识别空气颗粒致病和死亡的生物机制，以及识别对观察到的健康影响负有责任的污染源。更一般地，所提出的方法代表了在可应用于涉及高维数据的各种生物医学研究环境中的函数数据分析、隐马尔可夫建模和测量误差建模领域的进步。

DESCRIPTION (provided by applicant): The goal of this proposal is to develop statistical methods for environmental health data when the health effects of interest are complex. The Specific Aims are motivated by problems arising in toxicological and environmental epidemiological studies of the health effects of airborne particulate matter. Specific aims of the project are the development of (i) Wavelet-based historical functional data models for assessing high-dimensional associations between exposure and health; (ii) Hierarchical hidden Markov models for analyzing multivariate functional data arising from animal particulate matter concentrator studies; (iii) Methods to address exposure measurement error arising from spatial and temporal misalignment in particulate matter epidemiology studies. We will apply the proposed methods to several data sets for which existing analysis methods do not make full use of the data, including (i) semi-continuous heart- rate variability data from matter in sensitive subpopulations. In the motivating applications, the methods will provide insight into two scientifically pressing issues in environmental health research: the identification of biologic mechanisms of morbidity and mortality of air particles, and identification of pollution sources responsible for observed health effects. More generally, the proposed methods represent advancements in the areas of functional data analysis, hidden Markov modeling, and measurement error modeling that are applicable in a variety of biomedical research settings involving high-dimensional data.