Low-cost detection of dementia using electronic health records data: validation and testing of the eRADAR algorithm in a pragmatic, patient-centered trial.

目前，近一半的痴呆症患者没有得到诊断，延误了治疗 以及对病人和家属的教育和支持。因此，NIA要求应用程序支持 低成本工具的实用临床试验，以改善临床环境中认知能力下降的检测（RFA-AG- 20-051）。在NIA的试点资助下，我们利用机器学习开发了一种名为eRADAR的低成本工具 （电子健康记录阿尔茨海默氏症和痴呆症风险评估规则），该规则使用易于访问的 电子健康记录（EHR）中的信息，以帮助识别未确诊的痴呆症患者。在 此外，我们还采访了患者、护理人员、临床医生和医疗系统领导， 在临床环境中实施eRADAR。利益攸关方强烈认为，这一工具应 在现有临床关系的背景下，通过初级保健实施，并需要 并为患者和临床医生提供额外的支持。我们目前的建议在很大程度上是基于此 发展工作。在目标1中，我们将使用EHR数据来评估eRADAR在不同环境中的性能。 患者亚组，包括按人种/种族，在两个医疗保健系统中，以告知切割的选择， 在临床环境中使用的要点。我们将选择一个最佳的临界点，用于针对性痴呆症 评估与利益相关者的投入，平衡灵敏度，特异性和阳性预测值。在目标2中， 将进行一项实用的临床试验，以确定是否将eRADAR作为 对高风险患者的支持外展进程改善了痴呆症的检测。该设置将 Kaiser Permanente华盛顿（KPWA）内的初级保健诊所，这是一个综合医疗保健提供系统 在华盛顿州，和加州大学，旧金山弗朗西斯科（UCSF），一个城市，学术医疗保健 一个具有不同患者群体的系统。该研究包括6家诊所，约24，000例年龄≥65岁的患者。内 每个诊所，初级保健提供者（PCP）将被随机分配，让他们的患者具有高eRADAR 针对外展（干预）或常规护理（对照）的评分。我们的临床研究人员， 旨在反映这些医疗保健系统中的现有角色，以最大限度地发挥实用主义的作用， 对eRADAR评分高的患者进行认知障碍评估， 建议给PCP，并在诊断后支持患者。eRADAR评分较高的患者， 将对治疗组进行随访，以确定eRADAR对新诊断的痴呆（原发性）的影响。 结果）从EHR评估（再次，最大限度地实用主义）。在目标3中，我们将探讨 eRADAR实施对次要结局的影响，包括医疗保健利用和 患者和家属。如果这一务实的试验取得成功，eRADAR工具和流程可能会 传播到其他医疗保健系统，可能改善认知衰退的检测，患者护理， 生活质量

Nearly half of people currently living with dementia have not received a diagnosis, delaying access to treatment as well as education and support for the patient and family. Thus, NIA has requested applications to support pragmatic clinical trials of low-cost tools to improve detection of cognitive decline in clinical settings (RFA-AG- 20-051). With pilot funding from NIA, we used machine learning to develop a low-cost tool called eRADAR (electronic health record Risk of Alzheimer's and Dementia Assessment Rule), which uses easily accessible information in the electronic health record (EHR) to help identify patients with undiagnosed dementia. In addition, we interviewed patients, caregivers, clinicians, and healthcare system leaders to inform pragmatic implementation of eRADAR in clinical settings. Stakeholders felt strongly that such a tool should be implemented through primary care, in the context of existing clinical relationships, and would need to be accompanied by additional support for patients and clinicians. Our current proposal is heavily informed by this development work. In Aim 1, we will use EHR data to evaluate eRADAR's performance in different patient subgroups, including by race/ethnicity, in two healthcare systems to inform selection of cut- points for use in clinical settings. We will select an optimal cut-point to use for targeted dementia assessment with stakeholder input, balancing sensitivity, specificity, and positive predictive value. In Aim 2, we will perform a pragmatic clinical trial to determine whether implementing eRADAR as part of a supported outreach process to high-risk patients improves dementia detection. The setting will be primary care clinics within Kaiser Permanente Washington (KPWA), an integrated healthcare delivery system in Washington State, and the University of California, San Francisco (UCSF), an urban, academic healthcare system with a diverse patient population. The study includes 6 clinics with ~24,000 patients age ≥65. Within each clinic, primary care providers (PCPs) will be randomly assigned to have their patients with high eRADAR scores targeted for outreach (intervention) or to usual care (control). Our clinical research staff—whose roles were designed to reflect existing roles within these healthcare systems to maximize pragmatism—will reach out to patients with high eRADAR scores, conduct an assessment for cognitive impairment, make follow-up recommendations to PCPs, and support patients after diagnosis. Patients with high eRADAR scores in both treatment arms will be followed to determine the impact of eRADAR on new diagnoses of dementia (primary outcome) as assessed from the EHR (again, to maximize pragmatism). In Aim 3, we will explore the impact of eRADAR implementation on secondary outcomes including healthcare utilization and experience of patients and family members. If this pragmatic trial is successful, the eRADAR tool and process could be spread to other healthcare systems, potentially improving detection of cognitive decline, patient care, and quality of life.