Modeling Time-Dynamic Multilevel Outcomes in Patients on Dialysis

 描述(申请人提供)：终末期肾病与加速死亡有关，心血管疾病是导致死亡的主要原因。与更有效地护理透析患者相关的是描述结果轨迹如何随着时间的推移而演变，并确定其相关的危险因素。从透析开始，阐明患者水平的风险因素(如感染)和设施水平的特征(如设施的患者护理人员组成)随时间的变化对患者的CV结果的影响是重要的。我们的长期目标是在确定可修改的 患者水平和设施水平的危险因素和提高透析护理提供者质量的途径。为了实现这一目标，我们将开发一个通用框架来估计和推断患者结果的多级别时间动态建模，该框架适应多级别数据结构(例如，嵌套在透析设施或护理提供者中的患者和嵌套在患者体内的一段时间的观察)。我们提出的对透析患者心血管事件和感染风险因素的时间动态影响的新模型，对于设计有效的疾病管理和预防方法非常重要，因为它允许识别风险增加的特定时间段。此外，拟议的框架与设施级别的决策相关，包括预测透析机构的患者护理策略的变化是否会随着时间的推移而导致患者结果的改善，以及动态时间绩效评估。创新。到目前为止，在适应多层纵向数据结构的多层变系数建模(MVCM)中，还没有一个可行的框架来估计和双重推理(患者和设施层)。我们的工作将是第一次同时研究MVCM中患者和设施水平的推理，并灵活地模拟跨越一系列模型的设施水平效应，包括设施(I)。 固定效应，(Ii)恒定随机效应，以及(Iii)时间的随机效应函数(随机系数函数)。这也将是第一项研究，以检查持续透析设施的性能评估从透析开始，并允许确定特定的时间段，以改善目标患者的结果。目标。建议的框架将通过以下具体目标实现：1)(主题级推理)开发和应用MVCM用于多级别纵向响应(结果)，具有一般主题级协变量和设施级效应的灵活建模；2)(设施级推理)开发和应用MVCM用于设施级推理；3)(MVCM性能特征)描述MVCM的操作特征，包括跨信息稀疏水平和推理目标的相对效率和对建模假设的敏感性。

 DESCRIPTION (provided by applicant): End stage renal disease is associated with accelerated mortality, and cardiovascular (CV) disease is the leading cause of death. Of relevance to more effective care of patients on dialysis is characterizing how outcome trajectories evolve over time and identifying their associated risk factors. Elucidating the time-varying effects of patient-level risk factors, such as infection, and facility-level characteristic, such as facilities' patient care staffing composition, on patients' CV outcome over time, from the start of dialysis is important. Our long-term goal is to provide guidance in identifying modifiable patient-level and facility-level risk factors and approaches to quality improvement of dialysis care providers. Towards this goal, we will develop a general framework to estimation and inference for multilevel time-dynamic modeling of patient outcomes that accommodates multilevel data structures (e.g., patients nested within dialysis facilities or care providers and observations over time nested within patients). Our proposed novel modeling of time- dynamic effects of risk factors of CV events and infection in patients on dialysis, is important for designing effective approaches to disease management and prevention because it allows identification of specific time periods of increased risk. In addition, the proposed framework is o relevant to facility-level decision making, including prediction of whether changes in a dialysis facility's patient care strategy would lead to improved patient outcome over time, as well as time-dynamic performance evaluation. Innovation. To date, there does not exist a feasible framework for estimation and dual inference (patient- and facility-level) in multilevel varying coefficient modeling (MVCM) that accommodates multilevel longitudinal data structures. Our work will be the first to study both patient- and facility-level inference in MVCM simultaneously and to flexibly model facility-level effects that span a spectrum of models, including facility (i) fixed effects, (ii) constant random effects, and (iii) random effects functions of time (random coefficient functions). This will also be the first study to examine continuous dialysis facility performance assessment from initiation of dialysis and allow for identification of specific time periods for targeted patient outcome improvement. Aims. The proposed framework will be achieved through the following specific aims: 1) (Subject-level Inference) To develop and apply MVCM for multilevel longitudinal response (outcome) with general subject- level covariates and flexible modeling of facility-level effects; 2) (Facility-level Inference) To develop and apply MVCM for facility-level inference; 3) (MVCM Performance Characteristics) To characterize the operational characteristics of MVCM, including relative efficiency and sensitivity to modeling assumptions, across information sparsity levels and inferential goals.