Novel Analytic Techniques to Assess Physical Activity

描述（由申请人提供）：在开发和使用基于加速度计的运动传感器进行身体活动研究方面取得了进展。然而，处理活动监测数据的传统方法不能提供足够的准确性，以满足目前在研究领域使用客观体育活动数据的趋势。本提案的目的是解决基于加速度计的PA评估方法中的这一弱点：具体目标是：1)开发和验证处理Actigraph加速度计数据的新方法，以使用强大的现代分类方法（分类树，判别分析，隐马尔可夫模型，神经网络，回归样条和支持向量机）改进PA估计；2)比较这些分类方法和传统方法在受控环境下评估PA；3)比较自由生活PA条件下PA的分类方法和传统量化方法，选择推荐方法；4)从新的分类方法和传统的PA量化方法中，修正习惯PA汇总估计中的测量误差。我们唯一合格的多学科研究小组将通过首先开发创新的分类方法来识别实验室环境中的特定活动，然后使用从受控实验室环境和自由生活环境中进行的已知活动收集的数据来验证模型，从而实现这些目标。在这些研究结果的基础上，将对分类方法进行改进，并使用统计测量误差方法调整对PA行为的估计，以获得更准确的PA估计。我们选择的分类方法包括其他人可以轻松使用的公开可用的“现成的”分类方法。由此产生的数据处理程序将在流行的商业软件包中实现，并免费提供。拟议的调查结果将通过提供创新的方法来获得更准确和详细的PA评估，从而推动PA评估领域向前发展，使用一种流行的基于加速度计的PA监测器。这种系统的方法将提供信息，使人们更清楚地了解PA与健康之间的剂量-反应关系以及这种关系的生理基础。

DESCRIPTION (provided by applicant): Progress has been made in developing and using accelerometer-based motion sensors for physical activity research. However, traditional methods of processing activity monitor data do not provide sufficient accuracy to satisfy current trends in the use of objective physical activity data in the research arena. The aims of this proposal address this weakness in accelerometer- based PA assessment methodologies: The specific aims are: 1) To develop and validate novel methods to process Actigraph accelerometer data to improve estimates of PA using powerful modern classification methods (classification trees, discriminant analyses, hidden Markov models, neural networks, regression splines, and support vector machines); 2) To compare these classification methods and traditional approaches for assessing PA in a controlled setting; 3) To compare the classification methods and traditional approaches for quantifying PA in free living PA conditions and to select a recommended method; and 4) To correct for measurement error in summary estimates of habitual PA from the novel classification methods and traditional approaches for quantifying PA. Our uniquely qualified multidisciplinary research group will address these aims by first developing innovative classification methods to identify specific activities in a laboratory setting, and then validating the models using data collected from known activities performed in both controlled laboratory environments and free- living situations. Based on the results of these studies, the classification methods will be refined, and estimates of PA behavior will be adjusted using statistical measurement error methods to derive more accurate estimates of PA. We have chosen the classification methods to include publicly available "off-the shelf" classification methods that others can easily use. The resulting data processing programs will be implemented in popular commercial software packages and made freely available. The results of the proposed investigations will move the field of PA assessment forward by providing innovative approaches to derive more accurate and detailed estimates of PA using a popular accelerometer-based PA monitor. This systematic approach will provide information leading to a clearer understanding of the dose-response relationship between PA and health and the physiological basis of this relationship.