Predictive models for incident cirrhosis in non-alcoholic fatty liver disease using genetic and electronic medical record-based risk factors

项目摘要/摘要 非酒精性脂肪性肝病(NAFLD)影响着美国8000万人，并与 与肝脏相关的死亡人数上升36%。而非酒精性脂肪肝是增长最快的肝硬变和与肝脏相关的疾病 并发症，并不是所有NAFLD患者最终都会发展为肝硬变。我们识别哪些患者的能力 处于最高风险的人是有限的，这使得分配密集的生活方式干预和 对高危人群进行药物治疗。纤维化分期是预测肝硬变发生的最强指标， 但现有的纤维化只能确定已经进展为肝硬变的患者，需要 先进的表型，如活组织检查或瞬时弹性成像，并不是普遍可用的。会是 对于开发改进的疾病进展模型至关重要。该项目侧重于两个因素，它们可能 改善进展为肝硬变的风险分层：使用电子医学的遗传学和机器学习 记录(电子病历)数据。肝纤维化和肝硬变的遗传力高达50%，而一些遗传性 变异已经被认为与肝硬变的风险有关。电子病历是临床使用的丰富而复杂的数据来源 练习一下。在构建具有如此高维数据的模型时，非线性效应和相互作用 在预测值之间是常见的；机器学习算法的性能可能会优于更常用的 Logistic回归模型在这方面的应用。该项目的总体目标是为以下项目生成预测模型 哪些NAFLD患者通过结合遗传学和EMR最有可能进展为肝硬变 带有机器学习的预测器。具体目标是(1)表征遗传风险因素对 从NAFLD到肝硬变的进展率，(2)训练和验证事件的机器学习模型 基于EMR数据的肝硬变，以及(3)生成融合了EMR和遗传数据的集成模型。 为了实现这些目标，陈博士将接受电子病历数据处理的进一步培训，这是 统计遗传学、机器学习和预测建模。陈博士的长期目标是成为一名 领先的独立研究人员生成预测NAFLD结果的模型，最终甚至 相应地对患者进行优先处理。NIDDK K08奖将为陈博士提供必要的 时间和培训，以实现他的职业目标和改善对非酒精性脂肪肝患者的护理。总体而言，这个项目 将提高预测哪些NAFLD患者最有可能发展为肝硬变的能力，因此 通过帮助医疗机构将高危人群优先安排为更高风险人群，提高精准健康水平 干预。

Project Summary/Abstract Non-alcoholic fatty liver disease (NAFLD) affects >80 million people in the United States and is implicated in up 36% of liver-related deaths. While NAFLD is the fastest-growing cause of cirrhosis and liver-related complications, not all patients with NAFLD ultimately develop cirrhosis. Our ability to identify which patients are at highest risk is limited, which makes it challenging to allocate intensive lifestyle intervention and pharmacologic therapy to those at highest risk. The strongest predictor of incident cirrhosis is fibrosis stage, but existing fibrosis only identifies patients who have already progressed toward cirrhosis and requires advanced phenotyping such as biopsy or transient elastography which are not universally available. It will be critical to develop improved models for disease progression. This project focuses on two factors which may improve risk stratification of progression to cirrhosis: genetics and machine learning using electronic medical record (EMR) data. Heritability of liver fibrosis and cirrhosis is as high as 50%, and a number of genetic variants have been linked to risk of cirrhosis. The EMR is a rich but complex source of data used in clinical practice. When constructing models with such high-dimensional data, non-linear effects and interactions between predictors are common; machine learning algorithms may outperform the more commonly-used logistic regression models in this respect. The overall goal of this project is to generate predictive models for which patients with NAFLD are most likely to progress to cirrhosis by integrating genetics and EMR-based predictors with machine learning. The specific aims are (1) characterizing the effect of genetic risk factors on rate of progression from NAFLD to cirrhosis, (2) training and validating machine learning models for incident cirrhosis based on EMR data, and (3) generating integrated models incorporating both EMR and genetic data. To accomplish these aims, Dr. Chen will obtain further training in processing of EMR data, the fundamentals of statistical genetics, and machine learning and predictive modeling. Dr. Chen’s long-term goal is to become a leading, independent investigator generating models to predict outcomes in NAFLD and eventually even prioritize patients for treatment accordingly. An NIDDK K08 award will provide Dr. Chen with the necessary time and training to achieve his career goals and improve care for patients with NAFLD. Overall, this project will improve ability to predict which patients with NAFLD are most likely to develop cirrhosis and therefore enhance precision health by helping medical providers prioritize persons at highest risk to more intensive intervention.