Sleep Health Profiles and Mortality Risk in Older Adults: A Multi-Cohort Application of Novel Statistical Methods

项目总结/摘要 已知个人睡眠障碍和特征与不良健康结果相关。 然而，很少有研究在多变量背景下考虑睡眠。因此，我们提出了一个范式转变， 睡眠健康的睡眠研究。通过睡眠健康透镜进行研究可以改变健康状况 通过澄清哪些睡眠特征是最重要的治疗，加强筛查工作， 促进将研究结果推广到所有人，而不仅仅是那些患有特定睡眠障碍的人， 投诉然而，在实现这些长期目标之前，睡眠健康研究的关键第一步是 确定哪些睡眠概况预测健康结果（目标1），并开发预测算法， 可以根据睡眠和其他风险因素来识别有不良健康后果风险的个人（目标2）。 这些目标只有在认识到与固有的统计挑战有关的统计挑战的情况下才能实现。 睡眠健康的多维性（例如，睡眠的多个域、表示每个域的多个特征 域、多个数据源等）。这种多层面性产生了巨大的方法障碍 关于变量和模型选择，并威胁到研究人员的能力，以发展透明和 可复制的模型在本R 01中，研究人员将应用严格、复杂的统计方法， 60岁以上成年人的自我报告和多导睡眠图特征的汇总样本，以及10岁以上 三项母研究的纵向随访年：男性骨质疏松性骨折睡眠研究（MrOS）， 骨质疏松性骨折研究（SOF）和睡眠心脏健康研究（SHHS）。在目标1中，研究人员将 使用Cox比例确定哪些多维睡眠曲线可以预测老年人的死亡时间 风险模型、树结构生存分析和聚类。在目标2中，研究人员将开发强大的 机器学习算法，结合睡眠和其他风险因素，以确定哪些老年人患有 早期死亡的风险最大。在每一项目标中，将严格调查性别和种族的影响， 模型将在独立样本中进行内部验证，以确保重现性。次要目标是 调查不同人口统计学特征和睡眠方案研究的可推广性（威斯康星州 睡眠队列研究，檀香山亚洲睡眠美国老龄化研究）和不同的数据类型（睡眠日记， 活动记录法）。所有模型将在其他重要已知风险因素的背景下开发，包括但不限于 仅限于年龄、性别、种族、睡眠呼吸暂停、心血管疾病、吸烟和BMI。目标1的结果将 随后的机械研究集中在与睡眠相关的死亡原因上， 确定针对睡眠问题的有针对性的治疗方法，以降低死亡率和发病率。 Aim 2的发现将为增强筛选工具提供重要的初步证据， 准确识别哪些人有可能出现不良健康结果。

PROJECT SUMMARY / ABSTRACT Individual sleep disorders and characteristics are known to be associated with adverse health outcomes. However, few studies consider sleep in a multivariate context. Therefore, we propose a paradigmatic shift in sleep research towards sleep health. Conducting research through a sleep health lens can change health practice by clarifying which sleep characteristics are most important to treat, enhancing screening efforts, and facilitating the generalization of findings to all individuals, not just those with specific sleep disorders or complaints. Before achieving these long-term goals, however, the crucial first steps in the study of sleep health are to determine which sleep profiles predict health outcomes (Aim 1), and develop predictive algorithms that can identify individuals at risk of adverse health outcomes based on their sleep and other risk factors (Aim 2). These aims can only be achieved with an awareness of the statistical challenges related to the inherent multidimensionality of sleep health (e.g. multiple domains of sleep, multiple characteristics to represent each domain, multiple data sources, etc.). This multidimensionality generates substantial methodological barriers regarding variable and model selection and threatens researchers' abilities to develop transparent and reproducible models. In this R01, investigators will apply rigorous, sophisticated statistical methods to a large aggregated sample of adults aged >60 with self-report and polysomnographic characterization and over 10 years of longitudinal follow-up from three parent studies: Osteoporotic Fractures in Men Sleep Study (MrOS), Study of Osteoporotic Fractures (SOF), and the Sleep Heart Health Study (SHHS). In Aim 1, investigators will determine which multidimensional sleep profiles predict time to mortality in older adults using Cox-proportional hazards models, tree-structured survival analysis, and clustering. In Aim 2, investigators will develop powerful machine learning algorithms that incorporate sleep and other risk factors to identify which older adults have the greatest risk for early mortality. Within each aim, the impact of sex and race will be rigorously investigated, and models will be internally validated in independent samples to ensure reproducibility. The Secondary Aim is to investigate generalizability to studies with different demographic profiles and sleep protocols (Wisconsin Sleep Cohort Study, Honolulu Asia Aging Study of Sleep America) and different data types (sleep diary, actigraphy). All models will be developed in the context of other important known risk factors, including but not limited to age, gender, race, sleep apnea, cardiovascular disease, smoking, and BMI. Findings from Aim 1 will jumpstart subsequent mechanistic research focused on sleep-related causal factors of mortality, leading to identification of targeted treatments for sleep problems that could reduce risk of mortality and morbidity. Findings from Aim 2 will provide important preliminary evidence for enhanced screening tools that can more accurately identify which individuals are at risk for adverse health outcomes.