Towards Better Understanding of ALS using a Multi-Marker Discovery Approach from a Multi-Modal Database (ALS4M)

项目概要/摘要 本研究的总体目标是使用新的大型多模态数据资源和基于机器学习的 数据挖掘算法，以更好地了解风险因素，并改善肌萎缩症患者的诊断 侧索硬化症（ALS）。肌萎缩侧索硬化症（ALS）是一种罕见的，致命的神经退行性疾病， 90%的散发病例没有遗传原因，其影响风险因素在很大程度上是未知的。最 关于ALS风险因素的已知信息来自使用登记数据的流行病学研究， 历史上形成了主要的标准化大数据源，以帮助描述自然史，流行病学， 疾病负担;然而，这些研究得出的证据的强度差异很大。一个潜在 登记研究数据的主要局限性是收集的字段基于已知的潜在风险因素， 限制了它在探索新的关联和因果关系方面的可用性。此外，ALS是一种罕见的疾病， 患病率，因此无法使用传统的观察性研究设计研究其病因， 统计功率限制。医疗保健记录的数字化以及与其他相关信息的链接能力 数据来源现在使研究人群更具代表性，丰富和统计学上强大;以及 非常适合利用机器学习驱动的假设生成模型来识别新的风险因素， 模式识别新的危险因素，对理解，诊断或治疗ALS患者很重要。建筑 基于已建立的集成良好的真实的世界大数据源和已建立的集成嵌入特征 选择框架，已建立的多标记物（生物标记物、临床标记物、地理标记物、社会标记物） 将开发发现算法，以发现新的、可推广的风险因素（目标1）;新的症状 早期诊断模式（目标2）和ALS的有效临床护理途径（目标3）。最好的翻译 为了将研究结果转化为临床见解，组建了一个多学科和多利益相关者的团队，其中包括 只有在统计学，机器学习，临床研究信息学，神经学， 计算机科学，流行病学，但也有一个迷人的病人咨询委员会与不同的社会背景。 拟议的工作将是应用基于AI/ML的假设生成算法的首批试点研究之一 基于统计学上强大的真实世界数据，以弥合ALS风险因素的知识差距。这项工作不仅将 为CDC有毒物质和疾病登记机构（ATSDR）提供经验证据， 优先考虑未来扩大ALS登记风险因素调查的决定，但有助于更好地设计 对未来ALS的病因学研究和靶向试验提出建议。这项研究还将提供一个范例， 该框架可以推广到其他罕见和复杂疾病领域的研究， 利用真实的世界证据。

PROJECT SUMMARY / ABSTRACT The overarching goal of this study is to use new large multi-modal data resources and machine-learning-based data mining algorithm to better understand risk factors and improve diagnosis for people with Amyotrophic lateral sclerosis (ALS). Amyotrophic lateral sclerosis (ALS) is a rare, fatal neurodegenerative disorder, with 90% sporadic cases do not have genetic causes and their contributing risk factors are largely unknown. Most of what is known about ALS risk factors comes from epidemiological studies using registry data, which historically forms the main standardized big data source to help describe the natural history, epidemiology, and burden of disease; however, the strength of evidence resulting from these studies varies greatly. One potential major limitation to registry data are the fields collected are based upon known potential risk factors, which have restricted its usability for exploring novel associations and causalities. Moreover, ALS is a rare disease with low prevalence, thus making it infeasible to study its etiology using traditional observational study design due to statistical power constraints. The digitization of healthcare records and the capacity to link to other relevant data sources now enables a more representative, enriched and statistically powerful study population; and ideal for leveraging machine-learning-driven, hypothesis-generating models to identify new risk factors and patterns identify new risk factors important for understanding, diagnosing, or treating people with ALS. Building on established well-integrated real world big data source and established ensemble embedded feature selection framework, an established multi-marker (biomarker, clinical marker, geo-marker, socio-marker) discovery algorithm will be developed to discover novel, generalizable risk factors (Aim 1); new symptomatic patterns for early diagnosis (Aim 2), and effective clinical care pathways for ALS (Aim 3). To best translate findings into clinical insights, a multi-disciplinary and multi-stakeholder team has been assembled, including not only investigators with diverse expertise in statistics, machine learning, clinical research informatics, neurology, computer science, epidemiology, but also an engaging patient advisory board with diverse social background. The proposed work will be one of the first pilot studies applying AI/ML-based, hypothesis-generating algorithms on statistically powerful real-world data to bridge the knowledge gap on ALS risk factors. The work will not only provide CDC agency of toxic substance and disease registry (ATSDR) with empirical evidence to better prioritize future decisions on expanding the ALS registry risk factor survey but serve to inform better designed proposals for future etiological studies and targeted trials for ALS. This study will also provide an exemplar framework which can be generalizable to advance research of other rare and complex disease domains by leveraging real world evidence.