Improving the Accuracy of ASCVD Risk Estimation Using Population-Level EHR and Genetic Data

总结 心血管疾病（CVD）是美国发病率和死亡率的主要原因。 动脉粥样硬化性心血管疾病（ASVD）是CVD发展的主要机制， 疾病控制和预防中心认为这在很大程度上是可以预防的。降脂治疗是 目前ASCVD预防性治疗的主流和药物治疗指南依赖于2013年 用于估计10年风险的合并队列方程（PCE）。虽然这些方程已经过验证， 在人群水平上，它们具有影响现实世界患者水平有效性的显著缺点。这些 包括实现（即，临床医生将患者数据输入到电话或基于Web的 计算器），治疗改变对高度可变输入（例如单个时间点血压）的敏感性， 某些患者亚组的矛盾风险估计是线性建模的伪影（例如，女性 吸烟者），对种族的生硬处理（即黑人患者单独推导的方程），以及 现代队列（即导致高估风险）。本项目将尝试解决这些问题 缺点首先，便携式工具，用于分析罗得岛卫生部内发现的电子健康记录 信息交换（HIE）将被开发用于提取PCE风险因素，以实现自动化 计算ASCVD风险。PCE风险因素提取排列（例如，末次血压与平均血压）将 优化，并将为人口校准方程。接下来，EHR系统不可知的工具， 提取医疗记录中可用的额外风险因素，包括症状发展、社会风险、 健康的决定因素和家族史将得到发展。PCE和非PCE风险因素将用于 人工神经网络和动态贝叶斯网络建模ASCVD风险表型集群， 增加PCE风险预测。最后，单核苷酸多态性（SNP）基因型数据推导ASCVD 遗传风险评分将与HIE衍生的风险因素相结合，以证明潜在的临床 实施omics集成学习医疗保健系统的影响。该项目将作为 为主要研究者提供基础培训，使其成为该领域的医生-科学家 生物医学信息学。 假设：动脉粥样硬化性心血管疾病风险评估是当前降脂治疗的核心 指南这个项目将检验一个假设，即人口一级的数据驱动的方法将改善 风险计算器的准确性。 目的1：确定来自纵向HIE数据的PCE危险因素的预测性能 目的2：定义基于人群的ASCVD风险表型集群 目标3：展示HIE-Omics-集成学习医疗保健系统，直接面向消费者排序

SUMMARY Cardiovascular diseases (CVD) are the leading causes of morbidity and mortality in the United States. Atherosclerotic cardiovascular disease (ASVD) is the primary mechanism for the development of CVD and is largely considered preventable by the Center for Disease Control and Prevention. Lipid-lowering therapy is the current mainstay of preventative treatment for ASCVD and guidelines for pharmacotherapy rely on the 2013 Pooled Cohort Equations (PCE) for estimating 10-year risk. While these equations have been validated at a population level they have significant shortcomings that impact real-world patient-level effectiveness. These include implementation (i.e. time and effort for clinicians to enter patient data into a phone or web-based calculator), therapy changing sensitivity to highly variable inputs (e.g single time point blood pressure), paradoxical risk estimation for some patient subgroups that are an artifact of linear modeling (e.g. women smokers), blunt treatment of race (i.e. separately derived equations for black patients), and poor calibration for modern cohorts (i.e. resulting in the overestimation of risk). This project will attempt to address these shortcomings. First, portable tools for analyzing electronic health records found within the Rhode Island Health Information Exchange (HIE) will be developed for the extraction of PCE risk factors to enable the automated calculation of ASCVD risk. PCE risk factor extraction permutations (e.g. last vs mean blood pressure) will be optimized and the equations will be calibrated for the population. Next, EHR-system agnostic tools for extracting additional risk factors available within the medical record including symptom development, social determinants of health, and family history will be developed. PCE and non-PCE risk factors will be used for artificial neural network and dynamic Bayesian network modeling of ASCVD risk phenotype clusters to augment PCE risk prediction. Finally, a single nucleotide polymorphism (SNP) genotype data derived ASCVD genetic risk score will be integrated with the HIE derived risk factors to demonstrate the potential clinical implications of implementing an omics-integrated learning healthcare system. The project will serve as foundational training for the principal investigator towards pursuing a career as a physician-scientist in the field of biomedical informatics. Hypothesis: Atherosclerotic cardiovascular disease risk estimation is central to current lipid-lowering therapy guidelines. This project will test the hypothesis that population-level data-driven methods will improve the accuracy of risk calculators. Aim 1: Determine the Predictive Performance of PCE Risk Factors Derived from Longitudinal HIE Data Aim 2: Define Population-Based ASCVD Risk Phenotype Clusters Aim 3: Demonstrate HIE-Omics-Integrated Learning Healthcare System with Direct-to-Consumer Sequencing