Electrocardiogram-based deep learning and decision analysis to improve atrial fibrillation risk estimation

项目摘要/摘要 房颤(房颤)是一个主要的公共卫生问题，导致可预防的中风和增加的发病率 心力衰竭和早期认知功能衰退。房颤预计将影响美国近1200万人 到2030年。口服抗凝(OAC)在降低房颤相关中风的风险以及其他预防措施方面非常有效 诸如减肥、锻炼和戒酒等干预措施可能会降低房颤和相关疾病的风险 并发症。然而，房颤通常是无症状的，经常是间歇性的，因此可能很困难。 去诊断。尽管筛查可以发现未确诊的房颤，但大规模筛查方法并未导致 临床结果有意义的改善。当前筛查方法固有的主要低效 是否对许多房颤风险相对较低的个体进行筛查，从而导致低效和低收益的筛查 干预。因此，确定房颤高危个体的迫切需要尚未得到满足。 首先，为了优化房颤筛查和预防干预的效率。在本提案的目标1中， 我们将开发和比较基于深度学习的新方法，以自动方式评估房颤风险 使用移动单导联心电图机。在目标2中，我们将进行个人级别的模拟以量化 房颤筛查中基于风险的方法的比较和成本效益 以65岁为单纯年龄界值的房颤筛查临床标准。在目标3中，我们将进行一次试验 量化房颤风险评估的用户可接受性和量化两者之间的关联的研究 在18个月时评估房颤风险和真实房颤发生率。这项提案的总体目标是建立 自动房颤风险评估指导预防性干预的可行性和潜在临床价值 目的：减少房颤及其相关并发症的发生率。这些目标将在 制定全面的职业发展计划，旨在为库尔希德博士提供早期职业生涯 调查员，具备成为专注于以下领域的独立临床医生调查员所需的技能和经验 通过疾病风险预测改善心律失常的预后。这项建议 推动一个由机器学习、决策科学和前瞻性专家组成的多学科团队 临床研究，他将指导库尔希德博士向科学独立过渡。

Project Summary/Abstract Atrial fibrillation (AF) is a major public health problem resulting in preventable strokes and increased incidence of heart failure and early cognitive decline. AF is expected to affect nearly 12 million people in the United States by 2030. Oral anticoagulation (OAC) is highly effective in reducing risk of AF-related stroke, and other preventive interventions such as weight loss, exercise, and alcohol cessation may reduce risk of AF and associated complications. However, AF is commonly asymptomatic and is frequently episodic, and therefore may be difficult to diagnose. Although screening can detect undiagnosed AF, mass screening approaches have not resulted in meaningful improvements in clinical outcomes. A major inefficiency inherent within current screening approaches is the screening of many individuals at relatively low risk for AF, leading to an inefficient and low-yield screening intervention. Therefore, there is a critical unmet need to identify individuals at elevated risk of developing AF upfront, in order to optimize the efficiency of AF screening and preventive interventions. In Aim 1 of this proposal, we will develop and compare novel deep learning-based methods to estimate AF risk in an automated fashion using mobile single-lead electrocardiograms. In Aim 2, we will conduct an individual-level simulation to quantify the comparative and cost-effectiveness of a risk-based approach to AF screening, as compared to the current clinical standard of AF screening based on the simple age cutoff of ³65 years. In Aim 3, we will perform a pilot study to quantify the user acceptability of prospective AF risk estimation and quantify associations between estimated AF risk and true AF incidence at 18 months. The overall goal of this proposal is to establish the feasibility and potential clinical value of automated AF risk estimation to guide preventive interventions designed to reduce the morbidity resulting from AF and its associated complications. The aims will be executed in the setting of a comprehensive career development program designed to provide Dr. Khurshid, an early career investigator, with the skills and experience required to become an independent clinician investigator focused on the improvement of outcomes in cardiac arrhythmias through the use of disease risk prediction. This proposal impanels a multi-disciplinary team comprising experts in machine learning, decision science, and prospective clinical studies, who will guide Dr. Khurshid in his transition to scientific independence.