Multilevel Longitudinal Models and Causal Networks for Childhood Obesity

描述（由申请人提供）：1980年至2004年间，美国儿童和青少年的肥胖患病率增加了一倍多。肥胖让医疗保健提供者感到沮丧，因为它很难逆转。确定肥胖高危儿童一直是个问题，而且很难在任何层面上确定可采取预防性干预措施的致病因素。我们建议在Fels纵向研究（FLS）数据库中开发和应用创新的统计方法，以分析过去30年来儿童肥胖患病率增加的多变量和多层次决定因素。从出生开始的身体尺寸和组成的重复测量，以及在几十年里对FLS中同一个人的生活方式、饮食、运动和社会地位的频繁问卷调查，允许推断出无法通过横断面数据提供的因果关系。我们计划使用多层次纵向层次模型和格兰杰因果关系，这应该是最容易受到干预，以预防或延迟儿童肥胖的发生。我们将使用从1929年开始在FLS收集的2076个个体的序列数据，开发多变量和多层次模型和格兰杰网络来推断因果关系。我们将通过自举方法和交叉验证以及使用模拟FLS数据库的模拟数据来验证这些模型和网络的鲁棒性。我们的多层次模型和与肥胖流行有关的因素的格兰杰因果关系网络应该描绘出解释和跟踪肥胖流行的因素之间的合理途径和相互作用。这些途径和相互作用的发现和验证将揭示同时进行多层次干预的最佳目标，以预防儿童肥胖和/或改变相关因果变量的时间进程。我们认为，在减少社会、经济、饮食和其他环境变量方面，这种多层次干预将比目前应用的单层面干预更成功。该研究还允许发现和分析1929年至2008年社会环境变化的群体效应。虽然将对肥胖的危险因素进行调查，但这些方法将适用于其他变量集。我们的新方法应该有助于其他研究者计划纵向研究和分析纵向数据。

DESCRIPTION (provided by applicant): The prevalence of obesity among US children and adolescents more than doubled between 1980 and 2004. Obesity frustrates health care providers because of its difficulty to reverse. Identifying children at high risk of obesity has been problematic, and it is difficult to identify causative factors at any level that are amenable to preventive interventions. We propose to develop and apply innovative statistical methods to the Fels Longitudinal Study (FLS) database to analyze multivariate and multilevel determinants of the increase in prevalence of childhood obesity over the past three decades. Repeated measurements of body size and composition from birth and frequent administration of questionnaires on lifestyle, diet, exercise and SES on the same individuals in the FLS over decades permits inferences of causality that cannot be supplied by cross-sectional data. We plan to use the multilevel longitudinal hierarchical models and Granger causality s that should be most susceptible to interventions to prevent or delay the onset of obesity in childhood. We will use serial data collected in 2076 individuals in the FLS, beginning in 1929, to develop multivariate and multilevel models and Granger networks to infer causality. We will validate these models and networks for their robustness by bootstrap methods and cross-validation and by using simulated data that mimics the FLS database. Our multilevel modeling and Granger causality networks of factors involved in the obesity epidemic should delineate plausible pathways and interactions among factors that explain and track the epidemic. Discovery and validation of these pathways and interactions should reveal optimal targets for simultaneous multilevel interventions to prevent obesity in childhood and/or to alter the time course of relevant causal variables. We assume that such multilevel interventions will be more successful than currently applied single level interventions in reducsocial, economic, dietary and other environmental variables. The study also permits discovery and analysis of cohort effects of social-environmental changes from 1929 through 2008. While investigations will be performed on risk factors for obesity, these methods will be applicable to other sets of variables. Our new methods should assist other investigators in planning longitudinal studies and in analyzing longitudinal data.