Physical Activity Patterns via New Dimension-Informative Cluster Models.

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Physical activity is known to be a modifiable risk factor for various health outcomes and an effective trial could have significant effect on public health. Physical activity is a component of the American Heart Association (AHA) guidelines for ideal cardiovascular health, which advise at least 150 minutes per week of moderate intensity, or 75 minutes of vigorous intensity activity. A physical activity program is a critical component o primary and secondary prevention strategies for cardiovascular disease, and yet it may not be easy to follow these recommendations due to time and space constraints, or concomitant medical comorbities. Within the time duration guidelines, no further specific recommendations are available. Few studies defined physical activity variable detail enough to distinguish differen profiles or patterns of physical activity. Recognizing existing patterns of physical activity and patterns of changes in physical activity can help to design an effective trial. Goals of this proposal are to develop new cluster analysis methods to accommodate special features of physical activity data arising from questionnaire and accelerometry, apply the proposed cluster analysis to physical activity data from the Northern Manhattan Stroke Study (NOMAS) and the Endotoxin, Obesity, and Asthma in NYC Head Start (OEAHS) study, and validate utility of the identified patterns via proposed methods as predictors of cardiovascular outcome and obesity, respectively. Cluster analysis partitions subjects into meaningful subgroups, when the number of subgroups and other information about their composition may be unknown. Existing literature on cluster analysis of physical activity data are based on summary measures such as calorie consumed or duration spent on fixed number of categories of activities. Physical activity data are composed of variable, not fixed, number and type of activities and furthermore the number of activities is random and informative. State-of-the-art existing model-based cluster analysis has limitations to accommodate complexity of physical activity data. We propose several new model-based cluster analyses incorporating special features of physical activity data that existing cluster analysis cannot accommodate. The proposed model will handle (i) variable length of outcomes; (ii) the case when the dimension of outcome is informative; (iii) strictly positive outcomes without transformation; and (iv) repeatedly measured physical activity data. We will also apply the proposed method to accelerometry data. We will test utility of the identified clusters or patterns as predictors of cardiovascular outcomes using NOMAS questionnaire data, and predictors of obesity using OEAHS accelerometry data.