Developing Explainable AI for Equitable Risk Stratification of Atrial Fibrillation and Stroke

项目总结 房颤(AF)会导致严重的发病率、死亡率和每年超过60亿美元的住院费用 在受其影响的近600万美国成年人中。房颤是一种心律失常，可导致血液在 心房和形成的血栓进入大脑，导致中风。努力降低与以下疾病相关的中风发生率 房颤受到基本的卒中风险分层工具和护理差异的限制。迫切需要 对房颤患者进行个性化、社会感知、公平的中风风险预测，以实现最佳 现代中风预防疗法的实施。 这项提议的目标是使用人工智能(AI)和机器学习方法来捕获和 量化已知和新发现的房颤危险因素在社会经济背景下的协同效应。我的中央 假设中风的预防可以通过利用计算方法的方法来改进 在大数据集上增加了健康的社会决定因素(SDoH)。初步研究由 我们的团队和其他人已经揭示了SDoH因素对准确风险至关重要的患者亚组 分层。目标1是发现房颤患者新的危险因素关系，包括SDoH数据， 使用创新的共病发现框架(泊松二项共病发现)。目标2 重点是建立模型，将目标1中确定的变量与已建立的风险因素相结合以进行预测 使用人工智能方法的结果。为了做到这一点，我将构建新的概率图形模型(PGMS)来理解 SDoH和新发现的因素对房颤相关卒中风险的影响。 这一建议的主要创新是使用新的分析方法来理解和还原 房颤风险预测模型的差异。该提案旨在为改善全国各地的医疗保健提供手段 房颤患者的范围和解决目前护理标准中的差异。创建的人工智能工具将 临床医生和患者易于接触和理解，以帮助指导个人的治疗决定。 这项提议的完成将产生一种个性化和公平的方法来预防中风 自动对讲机。 该项目提供多学科计算和临床培训，并由 这两个领域的专家。概述的培训将为我提供计算和翻译 作为一名独立的研究人员和内科科学家，成功所需的心脏病学经验。

PROJECT SUMMARY Atrial fibrillation (AF) leads to significant morbidity, mortality, and over $6B in annual hospitalization costs among the nearly 6 million US adults it affects. AF is a cardiac arrhythmia which can cause blood to collect in the atria and form clots that travel to the brain resulting in a stroke. Efforts to reduce rates of stroke related to AF are limited by rudimentary stroke risk stratification tools and disparities in care. There is a critical need for personalized, socially aware, equitable stroke risk prediction among patients with AF to enable optimal implementation of contemporary stroke-prevention therapies. The objective of this proposal is to use artificial intelligence (AI) and machine learning methods to capture and quantify synergies among known and newly discovered AF risk factors in socioeconomic contexts. My central hypothesis is that stroke prevention can be improved through methods that leverage computational methods on large datasets augmented with information on social determinants of health (SDoH). Preliminary studies by our group and others have revealed subgroups of patients for whom SDoH factors are critical for accurate risk stratification. Aim 1 is to discover new risk-factor relationships for patients with AF that include SDoH data, using an innovative comorbidity discovery framework (Poisson Binomial Comorbidity Discovery). Aim 2 focuses on building models that combine the variables identified in Aim 1 with established risk factors to predict outcomes using AI methods. To do so, I will build novel Probabilistic Graphical Models (PGMs) to understand the impact of SDoH and newly identified factors on AF-related stroke risk. The primary innovation in this proposal is employing novel analytic approaches to understand and reduce disparities in AF risk prediction models. The proposal aims to provide means for improved care across the spectrum of patients with AF and address disparities in the present standard of care. The AI tools created will be readily accessible and interpretable by clinicians and patients to help guide individual treatment decisions. Completion of this proposal will yield a personalized and equitable approach to stroke prevention in the context of AF. This project provides multidisciplinary computational and clinical training augmented with mentorship from experts in both domains. The outlined training will provide me with the computational and translational cardiology experiences required to succeed as an independent investigator and physician-scientist.