Prediction of Chronic Kidney Disease by Simulation Modeling to Improve the Health of Minority Populations

项目摘要/摘要 在慢性肾脏疾病(CKD)、CKD进展和终末期肾脏病中存在显著的健康差异 疾病(ESRD)在不同种族的人群中。非裔美国人(AA)的患病率高出约25% 慢性肾脏病，终末期肾病的发生率高出3倍，在估计患有肾小球疾病的患者中，死亡风险最高 滤过率(EGFR)45~95ml/min/1.73m2。慢性肾脏病和终末期肾病最重要的传统风险因素是 糖尿病和高血压分别占新发终末期肾病病例的60%和70%。非- CKD的传统危险因素，如环境、文化行为因素、地理、教育、 保险覆盖面、社会经济地位和不平等获得最佳医疗保健的机会不成比例地影响 少数民族的慢性肾脏病健康状况。这些因素在现实世界中对慢性肾脏病进展的独特组合 仍然没有明确的定义。确定可减少慢性肾脏病差异的可修改风险因素将是 对提高生活质量、延长预期寿命、减轻经济负担具有不可估量的作用。仿真模型具有 已成功地应用于其他临床领域，但在CKD的发生和进展方面受到限制， 由于数据集较小，且缺乏使用纵向观察健康数据的建模技术。 此外，还没有任何模型在现实世界的少数族裔人群中进行测试，以揭示干预的可能性 将在更大范围内减少CKD差距的研究。据我们所知，我们创造了最大的， 从2006-2016年间查看的1000万人的电子健康记录中获得全面的数据库 加州大学洛杉矶分校(180万)和普罗维登斯·圣约瑟夫健康(PSJH；920万)系统公司之间为期两年的合作伙伴关系。 从加州大学洛杉矶分校注册人群中，我们确定了AA之间在EGFR轨迹下降方面的显著差异 和根据基线EGFR的非AAs，表明模式从较高的EGFR向较低的、较陡峭的EGFR转变 这一轨迹表明，可能存在关键的干预窗口，以减少AAA中的CKD差异。 所有民族队列的线性混合模型的种族/民族差异仍然存在，即使在控制 已知的影响EGFR轨迹的人口统计学和临床变量。我们假设，使用 在加州大学洛杉矶分校PSJH CKD/高危CKD联合登记中的种族多样化人群可以识别出一种新的组合 CKD危险因素；并改进现有模拟模型的性能，以预测CKD进展。这个 具体目标是：1)开发和测试基于机器学习的CKD和EGFR仿真模型 使用UCLA PSJH CKD/At-Risk CKD注册进行轨迹分析；并对模型和 与现有CKD风险模型的比较，2)基于不同种族/民族的分层和检验模拟模型 小组，包括基于跨机构比较的外部验证，以及3)通过以下方式开展重点小组 加州大学洛杉矶分校的初级保健医生，管理种族/民族患者，以诱导他们对现有和 设计了模拟模型，以减少CKD健康差距。这些创新的方法将促进我们的 长期目标是为临床决策支持方法提供信息，以减少/消除CKD健康差异。

Project Summary/Abstract Significant health disparities exist in chronic kidney disease (CKD), CKD progression, and end stage renal disease (ESRD) in ethnically diverse populations. African Americans (AAs) have ~25% higher prevalence of CKD, 3-fold higher rate of ESRD, and the highest risk of mortality among those with estimated glomerular filtration rate (eGFR) 45-95mL/min/1.73m2. The most significant traditional risk factors for CKD and ESRD are diabetes and hypertension accounting for >60% CKD and >70% of new ESRD cases, respectively. Non- traditional risk factors for CKD such as environmental, cultural-behavioral factors, geographic, education, insurance coverage, socioeconomic status and unequal access to optimal healthcare, disproportionately affect CKD health in ethnic minorities. The unique combination of these factors on CKD progression in the real world remains poorly defined. Identification of modifiable risk factors that may reduce CKD disparities would be invaluable to improve quality of life, life expectancy, and decrease economic burden. Simulation models have been successfully applied in other clinical domains, but are limited in CKD development and CKD progression, due to small datasets and the absence of modeling techniques using longitudinal observational health data. Further, no models have been tested in a real-world minority population to uncover the potential for interventional studies that would reduce CKD disparities on a larger scale. To our knowledge, we have created the largest, comprehensive database from electronic health records of >10 million individuals seen between 2006-2016 from a 2-year partnership between UCLA (1.8 million) and Providence St. Joseph Health (PSJH; 9.2 million) systems. From the UCLA Registry population, we identified significant differences in eGFR trajectory decline between AAs and non-AAs according to baseline eGFR, indicating a pattern shift from a higher to a lower, steeper eGFR trajectory suggesting there may be critical windows for interventions to reduce CKD disparities in AAs. Race/ethnicity differences from linear mixed models of all ethnic cohorts persisted even after controlling for demographic and clinical variables known to influence eGFR trajectories. We hypothesize that the use of ethnically diverse populations in the joint UCLA PSJH CKD/At-risk CKD Registry can identify a novel combination of CKD risk factors; and improve the performance of existing simulation models to predict CKD progression. The specific aims are to: 1) develop and test a machine learning-based simulation model for CKD and eGFR trajectories using the UCLA PSJH CKD/At-risk CKD Registry; and conduct internal validation of the models and comparisons with existing CKD risk models, 2) stratify and test simulation models based on different racial/ethnic groups, including external validation based on cross-institution comparisons, and 3) conduct focus groups with UCLA primary care physicians, who manage racial/ethnic patients, to elicit their perspectives on existing and designed simulation models to reduce CKD health disparities. These innovative approaches will facilitate our long-term goal to inform clinical decision support methods to reduce/eliminate CKD health disparities.