A Statistical Framework for the Spectral Analysis of Electrophysiology

描述(申请人提供)：大量研究使用多导睡眠图连续和非侵入性地记录睡眠期间的电生理信号，目的是利用这些数据来阐明睡眠影响健康和功能的生物学途径。多导睡眠图信号，如脑电图和心率变异性，测量复杂的动态过程，其频域特性提供了有价值的和可解释的信息。缺乏易于处理的统计模型和方法来量化非平稳时间序列集合的频域特性与其他研究变量(如临床结果和实验条件)之间的关联，这限制了通过分析多体描记数据可以准确解决的问题的范围。本研究项目的目标是开发一个基于随机半参数进化传递函数的框架，用于对睡眠研究中收集的电生理时间序列进行频谱分析。这一框架的三个具体方面将被探索：(1)混合效应模型的时变频谱模拟，该模型将允许对进化功率谱和其他变量之间的关联进行半参数分析，同时考虑到相关信号之间的相关性。(2)一种区分非平稳时间序列总体的程序，例如区分对治疗有积极反应的参与者和对治疗无效的参与者。(3)用于获得高维时间序列与由大量相关变量测量的过程之间关联的低维度量的时频典型相关分析。对于框架的每个方面，将开发估计器，建立理论和经验属性，并创建有效的算法和计算机程序。这些新方法将被用来分析现有的两项研究的数据：一项是关于老年人睡眠的临床实验研究，另一项是关于更年期妇女睡眠的多文化流行病学研究。

DESCRIPTION (provided by applicant): A large number of studies use polysomnography to continuously and non-invasively record electrophysiological signals during sleep with the goal of using these data to elucidate the biological pathways through which sleep affects health and functioning. Polysomnographic signals such as electroencenphograms and heart rate variability measure complex and dynamic processes whose frequency domain properties provided valuable and interpretable information. A dearth of tractable statistical models and methods for quantifying associations between frequency domain properties of collections of nonstationary time series with other study variables, such as clinical outcomes and experimental conditions, has limited the scope of questions that can be accurately addressed by analyzing polysomographic data. The goal of this research project is to develop a framework based on stochastic semiparametric evolutionary transfer functions for the spectral analysis of electrophysiological time series collected during sleep studies. Three specific aspects of this framework will be explored: (1) A time-varying spectral analogue of mixed effects models that will allow for the semiparametric analysis of the association between evolutionary power spectra and other variables while accounting for dependencies among correlated signals. (2) A procedure for discriminating between populations of nonstationary time series, such as between participants that respond positively to a treatment from those who do not. (3) A time-frequency canonical correlation analysis for obtaining low-dimensional measures of association between high-dimensional time series and processes measured by large collections of correlated variables. For each aspect of the framework, estimators will be developed, theoretical and empirical properties will be established, and efficient algorithms and computer programs will be created. These new methods will be used to analyze data from two existing studies: a clinical experimental study of sleep in older adults and a multi-cultural epidemiological study of sleep in women during the menopausal transition.