STATISTICAL METHODS IN LONGITUDINAL STUDIES OF DEMENTIA

描述：（改编自研究者摘要）纵向 阿尔茨海默病相关痴呆症的流行病学研究使用了 两阶段抽样设计，以估计人口特征， 疾病发病率以及疾病与假定风险之间的关联 因素对这些研究数据的统计分析因以下因素而变得复杂： 纵向设计，复杂的抽样方案，各种磨损模式，以及 多个感兴趣的结果。特别是，目前的统计方法 都未能解决因老年人死亡而导致结果缺失的问题 受试者，这可能是不可否认的基于可能性的推理。它 众所周知，忽略缺失数据问题的分析可能会导致 有偏见的结果。拟议的研究项目是发展统计方法 对于在存在不可解释性的情况下分析纵向痴呆数据， 由于死亡而缺失数据。具体而言，提出了三种方法。第一 是一种选择模型方法，其中由于死亡机制而缺失的数据是 根据死亡前的疾病状态进行建模。二是混合型 模型方法，其中疾病结局和死亡过程被建模以共享 一组共同的随机效应参数，被认为代表了受试者的 对疾病和死亡的力量（或易感性）。三是要建模 疾病发病率和死亡率同时使用疾病-死亡 随机模型申请人还提出了一些设计问题 纵向痴呆研究，并考虑调整错误分类 估计疾病与风险之间的关联时的风险暴露误差 因素申请人希望拟议的研究能够导致统一和 有效的统计方法在纵向痴呆研究，从而减少 研究结果中的偏差、低效率和差异， 分析方法申请人指出，本申请中开发的方法 建议广泛适用于其他流行病学研究， 纵向两相抽样设计。

DESCRIPTION: (Adapted from the Investigator's Abstract) Longitudinal epidemiological studies in Alzheimer's disease related dementia have used two-phase sampling designs to estimate population characteristics such as disease incidence and the associations between disease and putative risk factors. Statistical analyses on data from these studies are complicated by the longitudinal design, complex sampling schemes, various attrition patterns, and multiple outcomes of interest. In particular, current statistical methods have all failed to deal with the problem of missing outcomes due to death of elderly subjects, which is likely to be non-ignorable to likelihood based inference. It is well known that analyses ignoring the missing data problem may lead to biased results. The proposed research project is to develop statistical methods for the analysis of longitudinal dementia data in the presence of non-ignorable missing data due to death. Specifically, three methods are proposed. The first is a selection model approach where the missing data due to death mechanism is modeled to depend on disease status prior to death. The second is a mixture model approach where the disease outcome and death process are modeled to share a common set of random effect parameters thought to represent the subjects' strength (or susceptibility) for both disease and death. The third is to model disease incidence and mortality simultaneously using an illness-death stochastic model. The applicants also address some questions for designing longitudinal dementia studies and consider adjustment for misclassification error on risk exposure when estimating association between disease and risk factors. The applicants hope that the proposed research can lead to unified and efficient statistical methods in longitudinal dementia studies and hence reduce the bias, inefficiency, and variation in study results due to statistical analysis approach. The applicant states that the methods developed in this proposal are widely applicable to other epidemiological studies using longitudinal two-phase sampling designs.