Development and Validation of Cognitive Status Assessments in Older Adults Utilizing Conversational Agent Technology (CAT) Delivered on Smart Displays - the CAUCAT Study

项目概要： 轻度认知障碍（MCI），以及阿尔茨海默氏痴呆症和相关痴呆症（ADRD）仍然存在 不幸地被低估了尽管有许多科学验证的筛查方法， 工具和最近对医生的财政激励措施。这种“证据-实践”的差距至少在很大程度上是 因为可用的筛查技术管理起来麻烦、昂贵或太耗时， 在日常临床护理中经常使用。此外，目前的仪器往往需要训练有素的讲解员管理， 和/或护理人员确认，并且通常在人工临床环境中施用，而不是在人工临床环境中施用。 在损害影响日常生活的环境中进行监测，引入潜在的回忆偏差。相反地， 尝试重复评估典型地不是针对目标人口统计的最佳形式和功能， 需要额外的计算机知识和技能或广泛的用户技术培训，并限制用户 验收我们已经开发并展示了会话代理技术的潜在效用 (CAT)在智能扬声器和智能显示器上提供，用于连续监测认知和日常生活 老年人在家中的地位。对760名老年人（60-95岁）的初步数据表明， 部署由语音和屏幕管理支持的认知筛查工具，而不受以下潜在限制： 不熟悉的技术界面或对临床管理员的依赖，所有这些都在受试者的舒适环境中进行。 生活环境配套的智能手机应用程序允许护理人员在需要时进行评估。以下 文献回顾和专家小组的投入，我们已经确定，认知评估测试从井- 经过验证的NIH认知测验（NIHTCB）是适应和潜在广泛规模的最佳目标 临床和商业用途。本研究有两个目的：1）创建一个家庭会话代理 管理的综合认知评估平台，利用NIHTCB的评估测试;以及 2)定义CAT适应性认知评估的心理测量学特性，并将其与标准进行比较 评估。在我们的1A阶段研究中，具体目标是在30人的队列中证明可行性和可用性 从我们现有的商业用户群中招募的患者。在有令人满意的技术可行性证据后， 我们将进行第1B阶段的研究，具体目标是确定适应性的心理测量特性， 认知评估工具在100病人照顾者的二人组。CAT管理的评估将 与当前标准评估相比，评价了并行标准有效性。我们假设 与当前标准评估相比，组内相关系数（ICC）>0.9。评定者内可靠性 将采用重测方法进行评估，我们假设r >0.8。这项研究的结果将告知 第二阶段项目评估纵向评估和机器学习算法的临床效用， 预测认知能力下降的轨迹。 1

PROJECT SUMMARY: Mild Cognitive Impairment (MCI), as well as Alzheimer’s Dementia and Related Dementias (ADRD) remain woefully under diagnosed. This is despite the availability of numerous scientifically validated screening instruments and recent financial incentives to physicians. This “evidence-practice” gap is at least in large part because available screening techniques are cumbersome to administer, expensive or too time consuming for regular use in daily clinical care. Additionally, current instruments often require trained docent administration, and/or caregiver confirmation, and typically are administered in the artificial clinical setting rather than monitored in the setting where impairment affects daily living, introducing potential recall bias. Conversely, attempts to digitize assessments typically are not form and function optimized for the target demographic, requiring additional computer literacy and skills or extensive user technology training, and limiting user acceptance. We have developed and demonstrated the potential utility of conversational agent technology (CAT) delivered on smart speakers and smart displays for serially monitoring the cognitive and daily living status of older adults in the home. Preliminary data on 760 older adults (age 60-95) suggests the ability to deploy cognitive screening tools enabled by voice and screen administration, without the potential limitations of unfamiliar technology interfaces or dependency on a clinical administrator, all in the comfort of the subject’s living environment. A companion smart phone app allows caregiver assessment where needed. Following literature review and expert panel input, we have determined that cognitive assessment tests from the well- validated NIH Toolbox Cognitive Battery (NIHTCB) are the best target for adaptation and potential broad scale clinical and commercial use. The present study has two objectives: 1) Create an in-home conversational agent administered comprehensive cognitive assessment platform utilizing assessment tests from the NIHTCB; and 2) Define the psychometric properties of CAT adapted cognitive assessments and compare them to standard assessments. In our Stage 1A study, the specific aim is proving feasibility and usability in a cohort of 30 patients recruited from our current commercial user base. Upon satisfactory evidence of technology viability, we will undertake a Stage 1B study with the specific aim of defining the psychometric properties of the adapted cognitive assessment instruments in 100 patient-caregiver dyads. CAT administered assessments will be evaluated for concurrent criterion validity compared to current standard assessments. We hypothesize Intraclass Correlation Coefficient (ICC) >0.9 compared to current standard assessments. Intra-rater reliability will be assessed with test-retest methods and we hypothesize r >0.8. Results from this study will inform a Phase II project assessing the clinical utility of longitudinal assessments and machine learning algorithms as predictors of the trajectory of cognitive decline. 1