STATISTICAL ANALYSIS OF BODY COMPOSITION RISK FACTORS

总体目标是弥合最近统计进展之间的差距 并改进统计方法， 它们在人体成分研究中的应用， 心血管及相关疾病的危险因素。的现有数据 身体成分和心血管疾病风险的纵向研究 因子将被用来提供数值例子，并证明 采用先进的统计方法。具体而言， 为： (i)随机效应方法将通过将时间 一系列的过程中，受试者内的错误，以确保充分 描述随时间变化的模式，并准确预测 未来价值和未来风险。未来高水平的风险 将研究以先前值为条件的值。 (ii)将进一步研究核估计器以包括： (i)实施交叉验证选择最佳组合 的平滑参数和权重函数，（ii）推导 估计量和预测值的置信区间，（iii） 拟合曲线的速度和加速度的发展，以及（iv） 标志点的特征，如最大速度，年龄， 最大值和最大速度下的预测值。 这些拟议的统计发展的动机是以下问题， 发生在建模不规则的身体组成和危险因素的数据， 纵向设计。身体成分的变化模式和风险 个体之间的因素是不规则的， 描述这些模式的一个或多个模型的错误说明。 因此，估计系数和预测值是 有偏见随机效应方法的推广及进一步发展 核回归将改善对未来值的预测， 以前的价值观准确预测未来价值的能力， 早期值在心血管和相关疾病的研究中是重要的 儿童期早期干预可防止发生的疾病 在成年期的异常水平。 虽然调查将根据身体成分数据进行， 心血管和相关疾病的风险因素，这些统计 方法将推广到任何性别或少数民族的任何序列数据 人口

The overall goal is to bridge the gap between recent statistical advances and current practice and to improve statistical methods so as to increase the applicability of their use in studies of human body composition and risk factors for cardiovascular and related diseases. Existing data from a longitudinal study of body composition and cardiovascular disease risk factors will be used to provide numerical examples and to demonstrate the application of the developed statistical methods. Specifically the aims are: (i) A random effects method will be extended by incorporating a time series process for the within-subject errors to ensure an adequate description of patterns of change over time and the accurate prediction of future values and future risk. The risk of elevated levels of future values conditioned on earlier values will be investigated. (ii) A kernel estimator will be investigated further to include: (i) the implementation of cross-validation to select the best combination of a smoothing parameter and a weight function, (ii) derivation of confidence intervals for the estimator and the predicted values, (iii) development of the velocity and acceleration of the fitted curve, and (iv) characterization of the landmark points such as maximum velocity, age at maximum, and predicted value at maximum velocity. These proposed statistical developments are motivated by the problems that occur in modeling irregular body composition and risk factors data from longitudinal designs. Patterns of change in body composition and risk factors are irregular among individuals which leads to the misspecification of the model or models describing these patterns. Consequently, the estimated coefficients and the predicted values are biased. The extension of random effects method and the further development of the kernel regression will improve the prediction of future values from earlier values. The ability to accurately predict future values from earlier values is important in studies of cardiovascular and related diseases where early intervention in childhood may prevent the occurrence of aberrant levels in adulthood. Although the investigation will be performed on body composition data and risk factors for cardiovascular and related diseases, these statistical methods will be generalizable to any serial data in any gender or minority population.