Developing Models to Identify Veterans with Nonalcoholic Fatty Liver Disease and Predict Progression

对退伍军人医疗保健的预期影响：该提案将使用自然语言处理（NLP） 提供和比较非酒精性脂肪酸的预测模型的方法和机器学习途径 退伍军人中的肝脏疾病（NAFLD）。拟议的分析还将检查以下方面的种族/民族差异： NAFLD的诊断、治疗和结局，目的是确定最高风险的患者群体。 进展为肝硬化和肝硬化相关并发症。这项研究的长期目标是， 这项试点研究将促进，是综合多学科的发展和有效的目标， 治疗算法，以及简单，文化上适当的，具有成本效益的干预措施，以遏制 NAFLD及其并发症在退伍军人中的流行。 背景：非酒精性脂肪肝是一个重要的和日益增长的健康问题密切相关的肥胖，2型 糖尿病（T2 DM）、高血压和血脂异常。在弗吉尼亚州，NAFLD的患病率一直是 估计高达46%。NAFLD的患病率因人群而异 研究和使用的测试。在达拉斯心脏研究中，估计超过30%的患者 通过MR波谱分析的NAFLD。重要的是，研究人员发现，NAFLD的最高患病率 发生在西班牙裔（58%）和T2 DM（超过70%）。西班牙裔人口 NAFLD的发生率和进展为晚期纤维化的可能性更高，与非NAFLD相比， 西班牙裔白色（NHW）患者。目前的治疗旨在优化心血管和肝脏相关风险 因素（即T2 DM、高血压、高脂血症、肥胖、吸烟等）。饮食驱动的生活方式变化 干预和运动是诱导和维持体重减轻，减少脂肪量， 高胰岛素血症和胰岛素抵抗，从而减少脂毒性肝损伤和多系统代谢 后果VA NAFLD诊所提供强化减肥，包括营养，运动， 行为，VA批准的药物（例如，安非他酮/纳曲酮、氯卡西林）和减肥手术。 因此，重要的是要识别出处于进展至不良结局的高风险患者， 与先进的NAFLD，并提供治疗可在VA NAFLD诊所。 目的：在这项为期1年的试点研究中，我们建议使用VA NAFLD团队策划的队列（n= 61，900）， 来自国家退伍军人事务信息学和计算基础设施（芬奇）系统的退伍军人， 接受了肝脏活检。将对数据集进行扩充，以包括8年前和1年前的医疗记录。 活检后一年。我们将使用聚类和机器学习预测分析方法来识别 发生肝硬化、肝硬化相关并发症和心血管事件风险较高的患者 重点分析种族和民族差异。 方法：卷积神经网络和随机森林的机器学习方法将在 用于使用NLP变量、实验室检查值和合并症识别NAFLD患者， 芬奇系统中的患者记录。为了识别快速进展的NAFLD患者，我们将聚类 纤维化风险评分趋势数据。我们将量身定制NAFLD和进展的识别方法， 用机器学习分析来增强它我们的试验结果将是NAFLD的预测模型 患者沿着其严重性估计，其可用于确定哪些患者组处于 进展为肝硬化、肝硬化并发症和心血管事件的风险更高，因此， 从临床干预中受益，以积极降低其风险。下一步是后续研究， 使用在试点中得出的预测高危患者的模型，作为干预措施的一部分， 具有肝脏并发症和心血管事件进展高风险的退伍军人获得 在VA NAFLD诊所进行适当的护理。

Anticipated Impacts on Veterans Health Care: This proposal will use natural language processing (NLP) methods and machine learning approaches to provide and compare predictive models of non-alcoholic fatty liver disease (NAFLD) among Veterans. Proposed analyses will also examine racial/ethnic differences in NAFLD diagnosis, treatment, and outcomes with the goal of identify patient groups at highest risk of progression to liver cirrhosis and cirrhosis-related complications. The long-term goal of this research, which this pilot study will facilitate, is the development and effective targeting of integrated multidisciplinary treatment algorithms alongside simple, culturally appropriate, and cost-effective interventions to curb the epidemic of NAFLD and its complications among Veterans. Background: NAFLD is a significant and growing health problem closely associated with obesity, type 2 diabetes mellitus (T2DM), hypertension, and dyslipidemia. In the VA, NAFLD prevalence has been estimated as high as 46%. The prevalence of NAFLD varies significantly depending on the population studied and on the tests used. In the Dallas Heart Study, it was estimated that over 30% of patients had NAFLD by MR spectroscopy. Importantly, investigators found that the highest prevalence of NAFLD occurred among Hispanics (58%), and those with T2DM (over 70%). Hispanic populations have higher incidence of NAFLD and potentially higher rates of progression to advanced fibrosis, compared to non- Hispanic White (NHW) patients. Current therapy aims to optimize both cardiovascular and liver-related risk factors (i.e. T2DM, hypertension, hyperlipidemia, obesity, smoking etc.). Lifestyle changes driven by dietary intervention and exercise are the first line of therapy to induce and maintain weight loss, reducing fat mass, hyperinsulinemia and insulin resistance, thus decreasing lipotoxic liver damage and multisystem metabolic consequences. The VA NAFLD Clinic provides Intensive Weight Loss that includes nutrition, exercise, behavioral, VA approved pharmaceuticals (e.g., Bupropion/Naltrex, Lorcascerin) and bariatric surgery. Hence it is important to identify patients that are at high risk of progression to the poor outcomes associated with advanced NAFLD and provide treatments available at VA NAFLD Clinics. Objectives: In this 1-year pilot, we propose using the VA NAFLD Team curated cohort (n=61,900) of Veterans from the national Veteran Affairs Informatics and Computing Infrastructure (VINCI) system who have received liver biopsies. The dataset will be augmented to include medical records 8-years prior and 1- year post biopsy. We will use clustering and machine learning predictive analytic approaches to identify patients with higher risk of developing cirrhosis, cirrhosis-related complications, and cardiovascular events with a focused analysis on racial and ethnicity disparities. Methods: The machine learning methodology of convolutional neural networks and random forests will be used to identify NAFLD patients using NLP variables, laboratory values and comorbidities available in the patient records in the VINCI system. In order to identify rapidly progressing NAFLD patients we will cluster fibrosis risk score trend data. We will tailor the approach to identification of NAFLD and progression and augment it with machine learning analysis. The outcome of our pilot will be predictive models of NAFLD patients along with their severity estimate that can be used to determine which groups of patients are at higher risk of progression to cirrhosis, cirrhosis complications and cardiovascular events and thus, would benefit from a clinical intervention to proactively reduce their risk. The next steps is a follow on study that uses the models predicting high risk patients, derived in the pilot, as part of an intervention to improve access of Veterans with a high risk of progression to liver complications and cardiovascular events to appropriate care in VA NAFLD Clinics.