Understanding and Reducing Racial Bias in Cardiovascular Risk Prediction Using Novel AI Methods

在这项研究中，我们将研究预测分数和人工智能模型的使用，根据一个人过去的病史、人口统计信息和心电图（心脏活动的简单记录）来预测心脏病。尽管这种风险评分在临床实践中被大量使用，但它们通常是从大多数白人人群中开发出来的，而且人们还不太清楚它们是否适用于所有英国种族。为了回答这个问题，我们将分析现有的分数对不同种族的效果，以及它们是否能准确预测每个人的心脏病。然后，我们将使用机器学习技术来确定最重要的人口统计学和病史因素，这些因素会增加不同种族患心脏病的风险，并为每个种族量身定制风险评分。此外，为了了解使用人工智能和心电图的预测模型中是否存在类似的种族偏见，我们将比较主要来自白人人群的数据训练的模型与来自更多种族多样化人群的数据训练的模型。最后，我们将开发新的基于人工智能的模型，这些模型比目前的模型更容易解释，并且可以更准确地预测心脏病，而且没有种族偏见。

In this study, we will investigate the use of prediction scores and artificial intelligence models to predict heart disease based on a person's past medical history, demographic information, and electrocardiograms (simple recordings of the heart's activity). Although such risk scores are used abundantly in clinical practice, they are commonly developed from mostly white populations, and it is not well understood whether they work accurately for all UK ethnicities. To answer this, we will analyse how well existing scores work for different ethnicities and if they accurately predict heart disease for everyone. We will then use machine learning techniques to determine the most significant demographic and medical history factors that increase one's risk for heart disease in different ethnicities and create risk scores tailored to each ethnicity. Furthermore, to understand if similar ethnic bias exists in prediction models using artificial intelligence and electrocardiograms, we will compare models trained on data from mostly white populations versus models trained on data from more ethnically diverse populations. Finally, we will develop new artificial intelligence-based models that are more easily explainable than the current models, and that can predict heart disease more accurately and without ethnic bias.