Developing and Validating a Veterans Affairs Cardiac Risk Score

描述(由申请人提供)： 以患者为中心的医疗保健的中心目标是针对个别患者的风险、利益和偏好定制治疗协议。然而，如果不能准确预测单个患者罹患重大疾病的风险，这是不可能的。虽然使用多变量风险/收益预测工具来单独定制治疗可以提高治疗精度，但缺乏促进VA这种改善的工具。对于心脑血管(CCV)疾病，这是美国发病率和死亡率的主要原因，目前的风险预测工具有很大的缺点，包括要求手动输入风险因素信息，针对与VHA服务的患者人群截然不同的患者群体进行开发和校准，未能利用新的数据挖掘技术，以及未能利用 退伍军人管理局的电子病历(VA EMR)中提供了全面的临床数据。该项目将重点放在退伍军人的初级心血管预防上。我们将开发一个基于退伍军人管理局的风险预测评分，并使用该评分，开发新的临床算法，为退伍军人量身定做临床决策和风险/利益沟通。我们提出了一个为期3.5年的项目，使用10年的可集中获得的退伍军人事务部EMR数据(2001年至2010年)，并辅之以国家死亡指数、图表审查、质量和绩效办公室数据以及非退伍军人事务队列数据，以开发和验证退伍军人事务风险评分(VARS)。我们的研究有两个具体的目标：1.开发和评估两种相互竞争的方法来开发基于VA EMR的CCV风险预测工具，使用标准回归(REG)模型和机器学习(ML)模型。2.比较这些VA EMR衍生的CCV风险预测工具(REG和ML模型)的准确性和临床影响，以及与VA外部开发的常用风险预测模型，如Framingham和Euro Score Risk工具的准确性和临床影响。除了传统的鉴别措施(如C统计量)外，使用VAR的临床决策和患者结果的潜在改善将通过重新分类分析和以患者为基础的临床决策分析的发展来评估。我们的方法将使用国家VA数据来创建一个10年的纵向队列，以开发VA特有的CCV风险工具。我们将从DSS国家数据摘录中提取实验室和药房数据；从SAS医疗数据集中提取有关门诊、住院使用、ICD-9代码和CPT代码的数据；从企业数据仓库中提取临床测量数据；从国家死亡指数中提取死亡原因数据。REG模型将使用威布尔生存分析来开发，而ML模型将主要使用随机森林ML方法(目标1)。这些模型将用VA-CMS数据集和国家死亡指数数据进行验证。这些数据将用自然语言处理工具的结果加以扩充。数据质量将通过图表回顾和来自患者健康护理体验调查(SHEP)和社区动脉粥样硬化风险(ARIC)研究的数据进行评估。AIM 2将使用现代验证技术，包括风险重分类分析，以评估新分数的可靠性和准确性。 我们还将开发以患者为基础的临床决策分析，评估决策的风险和好处。这项工作是开发自动化工具、指南和质量评估所需的先决条件研究，这些工具、指南和质量评估可以集成到VA EMR或基于Web的界面(如MyHealtheVet)中，帮助临床医生和患者在门诊环境中优化和个性化CCV风险降低治疗决策。

DESCRIPTION (provided by applicant): The central goal of patient-centered medical care is to tailor treatment agreements toward individual patient risks, benefits, and preferences. This, however, cannot be possible without the ability to accurately predict an individual patient's risk of developing major medical illness. Whie the use of multivariable risk/benefit prediction tools to individually tailor treatments could grealy increase treatment precision, tools to facilitate such improvements in VA are lacking. For cardio- and cerebrovascular (CCV) disease, the leading cause of morbidity and mortality in the US, current risk prediction tools have substantial shortcomings, including requiring manual entry of risk factor information, being developed on and calibrated to patient populations quite different from those served by VHA, failing to utilize new data-mining techniques, and failing to utilize the full spectrum of clinical data available in VA's electronic medical record (VA EMR). This project will focus on primary cardiovascular prevention in Veterans. We will develop a VA-based risk prediction score and, using that score, novel clinical algorithms to tailor clinical decision-makin and risk/benefit communication to individual Veterans. We propose a 3.5-year project using 10-years of centrally available VA EMR data (2001 thru 2010) supplemented by the National Death Index, chart review, Office of Quality and Performance data, and non-VA cohort data to develop and validate the Veterans Affairs Risk Score (VARS). Our study has two specific aims: 1. To develop and assess two competing approaches to developing VA EMR- derived CCV risk prediction tools, using standard regression (REG) models and machine learning (ML) models. 2. To compare the accuracy and clinical impact of these VA EMR-derived CCV risk prediction tools (the REG and ML models) to each other and to commonly-used risk prediction models developed outside of VA, such as the Framingham and Euro SCORE risk tools. In addition to traditional measures of discrimination (such as the C statistic), potential improvements in clinica decision-making and patient outcomes using VARS will be assessed using reclassification analysis and the development of patient-based clinical decision analyses. Our methods will use national VA data to create a 10-year longitudinal cohort to develop the VA-specific CCV risk tool. We will extract laboratory and pharmacy data from the DSS National Data Extracts; data about outpatient visits, inpatient use, ICD-9 codes, and CPT codes from the SAS Medical Datasets; clinical measures from the Corporate Data Warehouse; and cause of death from the National Death Index. The REG models will be developed using Weibull survival analysis and the ML models will primarily use random forest ML methods (Aim 1). The models will be validated with VA-CMS datasets and National Death Index data. The data will be augmented with results from natural language processing tools. Data quality will be assessed with chart review and data from the Survey on Health Care Experiences of Patients (SHEP) and the Atherosclerosis Risk in Communities (ARIC) study. Aim 2 will use modern validation techniques, including risk reclassification analysis, to assess the reliability and accuracy of the new scores. We will also develop patient-based clinical decision analyses that will assess the risks and benefits of decisions. This work is the prerequisite research needed in order to develop automated tools, guidelines, and quality assessments that can be integrated into the VA EMR or a web-based interface (such as MyHealtheVet), helping clinicians and patients to optimize and personalize CCV risk reduction treatment decisions in the outpatient setting.