Genetic Epidemiology of Sleep Apnea and Comorbidities in Biobanks

摘要 睡眠呼吸暂停(SA)和失眠是两种最常见的睡眠障碍，两者都单独和 共同应对心肺、代谢和精神疾病的风险。尽管它们的流行率很高， 对SA和失眠的治疗仍然是次优的。失眠和失眠被认为是由不同的 亚型的定义仍然不明确，可能会对健康结果产生不同的风险。我们的目标是 使用机器学习将精确的表型分析应用于生物库，以确定SA和 失眠和发现基于遗传学和共病的SA和失眠亚型，以减少 表型异质性，指导患者分层，帮助发现更个性化的治疗方法。 我们的方法是将医疗系统生物库数据与研究多导睡眠图(PSG)相结合，以实现 发现SA和失眠相关表型的遗传变异并表征其特征的统计能力 相关的临床结果和内表型(生理机制)。我们将使用先进的自然 语言处理(NLP)方法，大幅提高SA和失眠表型的准确性。 我们预计的样本量将是先前SA基因研究的11倍，提供必要的 基因发现的统计学力量。根据我们的结果得出的多基因风险分数可以用来量化 睡眠障碍的风险，即使在那些没有睡眠表型的人中也是如此。机器学习方法可以识别 诊断的预测因素-病历中包含的群集性患者组。精确度极高- 表型PSG数据(如低氧负荷)可以在相关遗传基因座上表征内表型 基因定位。我们将得出高级SA和失眠的表型，这些表型对人口统计差异很强 在生物库网站上，对经过验证的SA和失眠表型进行迄今最大规模的遗传分析， 确定新的基因座特征，并研究与临床诊断数据的相关性，以改善患者的分类 三个生物库。我们将探索性别特定的关联，并在两个生物库中验证铅基因关联。 我们的具体目标是：1)构建先进的SA和失眠表观算法 人口统计群体和地点；2)识别和表征与SA和失眠的遗传联系； 3)根据相关的合并症识别和表征不同的SA和失眠患者亚组 配置文件。拟议的项目的目标是改善心肺、血液和睡眠障碍的治疗。 通过潜在地解决疾病的异质性，发现与睡眠障碍有关的新的基因，以及 帮助澄清SA和失眠与心肺、代谢和精神疾病的重叠。

ABSTRACT Sleep apnea (SA) and insomnia are the two most common sleep disorders, and both contribute individually and jointly to the risk of cardiopulmonary, metabolic, and psychiatric diseases. Despite their high prevalence, treatments for SA and insomnia remain suboptimal. SA and insomnia are thought to be comprised of distinct subtypes, which remain poorly defined and may contribute to differing risks for health outcomes. Our goal is to use machine learning to apply precise phenotyping to biobanks to identify the genetic bases of SA and insomnia and discover SA and insomnia subtypes based on genetics and comorbidities in order to reduce phenotype heterogeneity, guide patient stratification and aid in the discovery of more personalized treatments. Our approach is to combine health care system biobank data with research polysomnography (PSG) to achieve statistical power to discover genetic variants for SA and insomnia-related phenotypes and characterize their associated clinical outcomes and endophenotypes (physiological mechanisms). We will use advanced natural language processing (NLP) methods to substantially improve the accuracy of SA and insomnia phenotyping. Our anticipated sample size will be >11-fold larger than prior genetic studies of SA, providing the necessary statistical power for genetic discovery. Polygenic risk scores derived from our results can be used to quantify sleep disorder risk, even among those without sleep phenotypes. Machine learning methods can identify predictors of diagnosis-clustered patient groups contained within the medical record. Precision deeply- phenotyped PSG data (eg hypoxic burden) can characterize endophenotypes at associated genetic loci using genetic localization. We will derive advanced SA and insomnia phenotypes robust to demographic differences across biobank sites, perform the largest genetic analysis of validated SA and insomnia phenotypes to date, characterize novel loci, and study associations with clinical diagnosis data to improve patient classification in three biobanks. We will explore sex-specific associations and validate lead genetic associations in two biobanks. Our specific aims are: 1) to construct advanced SA and insomnia phenotying algorithms across diverse demographic groups and sites; 2) to identify and characterize the genetic associations with SA and insomnia; and 3) to identify and characterize distinct SA and insomnia patient subgroups based on related comorbidity profiles. The proposed project has a goal of improving the treatment of heart, lung, blood, and sleep disorders by potentially resolving disease heterogeneity, discovering novel genetic associations with sleep disorders, and helping to clarify the overlap of SA and insomnia with cardiopulmonary, metabolic, and psychiatric disease.