Artificial Intelligence for early Detection of Peripheral Artery Disease (AID-PAD)

项目摘要/摘要 外周动脉疾病是一种典型的下肢动脉粥样硬化性疾病，是一种危及生命和 影响数以百万计的美国人的衰弱状况。一旦确诊，包括启动在内的医疗管理 抗血小板治疗、降脂药物和行为治疗，如监督锻炼和 戒烟已被证明能显著改善PAD患者的健康状况。然而， PAD的诊断可能很困难，因为患者和提供者对这种疾病的认识很差，患病率很高 非典型症状和相互矛盾的筛查指南建议。此外，尽管有 与较高的疾病患病率类似，黑人、女性和社会经济较低群体中的个人 在疾病过程中较晚诊断，导致较差的结果。为了解决诊断率低的问题，我们 开发了一个基于人工智能(AI)的模型，用于在临床医生诊断之前使用Vavast检测PAD 大量的电子健康记录(EHR)数据和先进的机器学习算法。然而，对于我们的 技术要对现实世界产生影响，显然需要：1)验证我们基于AI的PAD的性能 跨不同临床环境和人群的检测模型(目标1)，2)，评估使用 基于AI的PAD筛查工具和设计有效的临床工作流程，以提高净收益和采用率(目标 2)和3)评估基于人工智能的PAD筛查工具对PAD诊断率和医疗效率的影响 管理模式(目标3)。目标1将使用来自3个临床站点的电子病历数据进行， 不同的患者群体。我们最终的模型将使用独特的美国家庭队列进行验证 注册表，这是一个丰富的基于门诊患者的EHR数据集，由来自所有50个州的患者组成，包括近 100万农村居民和60多万少数民族。我们将对AI进行严格的评估 使用算法公平度量的模型偏差。使用决策分析，我们将评估模型效用，以确保 我们的模型在部署前显示了正的净收益，我们还将采用独特的质量 改进和混合方法方法，与提供商合作开发临床工作流，以促进 使用人工智能进行PAD检测，并最大限度地提高模型效益。最后，采用阶梯式楔形临床试验设计。 我将对基于人工智能的PAD筛查工具对PAD诊断率的影响进行务实分析 和治疗。在这项研究的结论中，我们将对基于AI的 PAD筛查工具可用于改进PAD检测，减少诊断率差异，并提高 医疗管理。

PROJECT SUMMARY / ABSTRACT Peripheral artery disease, an atherosclerotic disorder typically of the lower extremities, is a life threatening and debilitating condition affecting millions of Americans. Once diagnosed, medical management including initiation of antiplatelet therapy, lipid lowering medications, and behavioral therapy such as supervised exercise and smoking cessation have all been shown to significantly improve health outcomes for those with PAD. However, diagnosis of PAD can be difficult due to poor patient and provider awareness of the disease, a high prevalence of atypical symptoms and conflicting guideline recommendations on screening. Furthermore, despite having similar to higher prevalence of disease, Blacks, females and individuals in lower socioeconomic groups are diagnosed later in the disease process, contributing to poorer outcomes. To address low diagnosis rates we developed an artificial intelligence (AI)-based model to detect PAD prior to clinician diagnosis using vast amounts of electronic health record (EHR) data and advanced machine learning algorithms. However, for our technology to have real-world impact, there is a clear need to: 1) Validate performance of our AI-based PAD detection model across diverse clinical settings and populations (Aim 1), 2), Evaluate clinical utility of using an AI-based PAD screening tool and design effective clinical workflows to enhance net benefit and adoption (Aim 2), and 3) Evaluate the effect of an AI-based PAD screening tool on rates of PAD diagnosis and medical management patterns (Aim 3). Aim 1 will be conducted using EHR data from 3 clinical sites with distinctly different patient populations. Our final model will be validated using the unique American Family Cohort registry, a rich outpatient-based EHR dataset made up of patients from all 50 states, including nearly 1,000,000 rural residents and over 600,000 racial/ethnic minorities. We will perform rigorous evaluation of AI model bias using algorithmic fairness metrics. Using decision analysis we will evaluate model utility to ensure our models demonstrate positive net benefit prior to deployment and we will also employ a unique quality improvement and mixed methods approach to work with providers to develop clinical workflows that foster the use of AI for PAD detection and maximize model benefit. Lastly, using a stepped wedge clinical trial design we will perform a pragmatic analysis of the effect of an AI-based PAD screening tool on rates of PAD diagnosis and treatment. At the conclusion of this study, we will have developed an understanding of how an AI-based PAD screening tool can be used to improve PAD detection, reduce disparities in diagnosis rates, and improve medical management.