Cutting Edge Survival Methods for Epidemiological Data

研究的主题是开发解决统计问题的新方法 我们团队在老龄化队列研究中的长期合作中出现的问题 人群和肾病患者。我们专注于开发强大的， 延迟输入数据回归参数的有效估计方法 流行的队列研究，当 协变量易受删失和测量误差的影响， 最佳模型的影响，终端事件的纵向测量的策略。 我们还计划开发公开的统计软件， 传播和推广。 所提出的延迟进入方法是基于一个增广的条件 从截断时间构造的似然性，而不指定截断分布， 这导致更有效的估计器。回归模型的建议方法 对于有删失协变量的纵向数据， 常用的非参数和参数方法。例如，非参数 方法无法恢复删失协变量的任何尾部信息， 检测，而参数方法可能会产生有偏见的结果，由于模型 错误说明我们的方法将有助于调查受限制的协变量 检测和测量误差，反映了许多实验室测量的真实性。 提出了具有终端发生的纵向数据的统计模型， 事件反映了响应变量和协变量之间的关系 与通常的边际模型大致相同（不考虑终端 事件），当数据被观察到远离终端事件，但变得严重 当数据采集时间接近终端时， 活动这些方法的动机是，并将适用于广泛的 数据集，包括肾脏疾病进展数据、肾移植登记 数据，从密歇根州骨骼健康中心收集的女性健康纵向数据， 代谢研究和全国妇女健康研究， 纵向终末期肾病医疗费用数据。这些方法将广泛应用于 适用于许多其他生物医学研究领域的问题。

The main theme of the research is to develop new methodologies for resolving statistical issues emerging from our team’s long-term collaborations in cohort studies of aging populations and patients with kidney disease. We focus on developing robust and efficient estimating procedures for regression parameters from data with delayed entry in prevalent cohort studies, making appropriate and efficient statistical inference when covariates are subject to censoring and measurement error, and developing new strategies that best model the effects of terminal events on longitudinal measurements. We also plan to develop publicly available statistical software with the goal of dissemination and generalization. The proposed approach for delayed entry is based on an augmented conditional likelihood constructed from truncation times without specifying the truncation distribution, which leads to a more efficient estimator. The proposed approach for regression models for longitudinal data with censored covariates addresses some serious issues with the common nonparametric and parametric methods. For example, the nonparametric methods cannot recover any tail information for the censored covariates due to limit of detection, while the parametric methods may yield biased results due to model misspecification. Our method will facilitate investigation of covariates subject to limit of detection together with measurement error, reflecting the reality of many lab measures. The proposed statistical models for longitudinal data with the occurrence of a terminal event reflect the reality that the relationship between a response variable and covariates is approximately the same as the usual marginal model (without considering terminal event) when data are observed far from the terminal event, but becomes heavily dependent on the terminal event time when data collecting time is close to the terminal event. These methods are motivated from and will be applied to a wide range of datasets, including the kidney disease progression data, the kidney transplant registry data, the women’s health longitudinal data collected from the Michigan Bone Health and Metabolism Study and the Study of Women's Health Across the Nation, and the longitudinal end stage renal disease medical cost data. The methods will be widely applicable to problems in many other fields of biomedical research.