Monitoring real-world driver behavior for classification and early prediction of Alzheimer’s disease

这项研究项目解决了NIH/NIA使用个人自己的车辆作为被动探测的巨大挑战 用于标记可能发出预警信号的潜在年龄和疾病相关异常驾驶的系统 功能衰退或早期阿尔茨海默病(AD)的迹象。早期识别和治疗是 减轻AD日益增长的成本和负担的基本步骤。我们在以下方面的基础性进展 从车载系统(黑匣子)和可穿戴传感器量化驾驶员行为，并制定战略 使用统计和机器学习方法的分析方法和管道，直接与 迎接NIH/NIA的挑战。该提案战略性地建立在当前项目的发现和成功的基础上 它全面表征了136名年龄较大的司机的真实驾驶暴露和风险模式 开了500,000英里。在该提案的概念框架下，职能能力决定了具体的驱动因素 行为模式和错误。NIA-阿尔茨海默病患者的行为、行为、驾驶功能指标和临床特征 轻度认知障碍(MCI)和阿尔茨海默病(AD)的协会(AA)核心临床标准(操作性 由阿尔茨海默氏症临床综合征[ACS])。睡眠和活动能力在人际关系中起着关键的中介作用 驾驶员行为和功能障碍之间的关系。因此，我们的具体目标(SA)是：SA1)提取密钥 在连续的3个月基准期内，真实的驾驶员行为特征将通常老化、 MCI，以及NIA-AA核心临床标准的ACS驱动程序。SA2)确定真实世界的驾驶员在多大程度上 在连续3个月的基准期内收集的睡眠和活动因素调节了这种关系 提取的驾驶员行为和临床特征之间的关系(SA1)。Sa3)开发模型(统计和监督 机器学习)，它结合了驾驶员行为(SA 1)和真实世界睡眠和移动(SA2)的特征，以 检测MCI和AD的临床特征和严重程度，并预测疾病进展。为了达到这些目标，我们的 专家团队-医学、AD、衰老和疾病中的驾驶、认知神经科学、交通运输 工程学、机器学习、计算机视觉和纵向生物统计学--将把我们的方法应用到 从老化到AD范围内具有更广泛损伤的司机。共有180名司机，年龄65岁- 90岁，根据NIA-AA临床标准，患有ACS(N=40)、MCI(N=80)或正常老龄化(N=60) 将在为期3个月的基准期内研究真实世界中自然驾驶行为、睡眠和机动性 监控。两次纵向评估，每一次相隔一年，将全面评估每个司机的风险 功能衰退的原因和严重程度。通过提取驾驶员异常行为的“数字指纹” AD的风险，该项目补充了临床前AD生物学诊断的地震进展和进展 NIH的首要任务是提高老年司机的安全性、机动性和生活质量，并提供前所未有的LO诊断 关心。被动监测真实世界行为以直接预测AD高危个体的临床状态 促进旨在早期治疗和防止阿尔茨海默病临床前期进展的干预措施。

This research project tackles the NIH/NIA grand challenge of using a person's own vehicle as a passive-detection system for flagging potential age- and disease-related aberrant driving that may signal early warning signs of functional decline or incipient Alzheimer's disease (AD). Early identification and treatment are essential steps to mitigating the growing costs and burden of AD. Our foundational advancements in quantifying driver behavior from in-vehicle systems ("Black Boxes") and wearable sensors, and strategic analytic methods and pipelines using statistical and machine learning approaches, are directly relevant to meeting this NIH/NIA challenge. The proposal builds strategically on current project discoveries and successes that comprehensively characterized patterns of real-world driving exposure and risk in 136 older drivers across 500,000 miles driven. Under the proposal's conceptual framework, functional abilities determine specific driver behavior patterns and errors. Behaviors, in tum, index driver functional abilities and clinical features of NIA-Alzheimer's Association (AA) core clinical criteria of mild cognitive impairment (MCI) and AD (operationalized by Alzheimer's clinical syndrome [ACS]). Sleep and mobility play roles as key mediators of relationships between driver behavior and functional impairment. Accordingly, our Specific Aims (SA) are: SA1) Extract key real-world driver behavior features over a continuous, 3-month, baseline period that classify normally aging, MCI, and ACS drivers by NIA-AA core clinical criteria. SA2) Determine the extent to which real-world driver sleep and mobility factors, collected over a continuous, 3-month baseline period, mediate the relationship between extracted driver behavior and clinical features (SA1). SA3) Develop models (statistical and supervised machine learning) that combine features of driver behavior (SA 1) and real-world sleep and mobility (SA2) to detect clinical feature severity of MCI and AD and predict disease progression. To address these aims, our team of experts-in medicine, AD, driving in aging and disease, cognitive neuroscience, transportation engineering, machine learning, computer vision, and longitudinal biostatistics--will apply our approach to drivers with a broader range of impairments across the aging to AD spectrum. A total of 180 drivers, ages 65- 90 years, who have ACS (N=40), MCI (N=80), or are normally aging (N=60) based on NIA-AA clinical criteria will be studied across a 3-month baseline period of real-world naturalistic driver behavior, sleep, and mobility monitoring. Two longitudinal assessments, each 1 year apart, will comprehensively assess each driver's risk for and severity of functional decline. By extracting "digital fingerprints" of aberrant driver behavior in drivers al risk for AD, this project complements seismic advances in biologic diagnosis of preclinical AD and advances NIH priorities lo improve older driver safely, mobility, quality of life, with unprecedented access lo diagnostic care. Passive monitoring of real-world behavior to predict clinical status in individuals at risk for AD directly promotes interventions aimed at early treatment of and preventing progression of AD in its preclinical stages.