Digital Phenotyping of Nonalcoholic Fatty Liver Disease

1 2项目摘要/摘要 3. 4非酒精性脂肪性肝病(NAFLD)管理中最严重的缺陷之一是缺乏有效的 5人群早期识别方法。本研究的目标是利用数据和分析来 6在与肝脏相关的疾病发作之前，通过NAFLD的早期检测和风险分层来改善医疗结果 并发症7例。人工智能在大型电子健康记录中的应用具有识别 发病前的8个疾病性状。这项建议的中心假设是有针对性的筛查 应用于大型集成医疗数据集的9个机器学习模型可以识别患有非酒精性脂肪肝的个体 10，更具体地说，那些具有进行性表型的人。我们将在两个具体的例子中检验中心假设 11个目标。首先，我们将使用多个纵向数据点来训练NAFLD预测的机器学习模型 在以人群为基础的特征良好的队列中，被诊断为 13非酒精性脂肪肝是指一般人群中没有非酒精性脂肪肝的个人。我们假设 14无监督机器学习可以在没有人类指导的情况下识别复杂的过程和模式 15发现反映一种表型的早期共病簇(在NAFLD发展之前就存在的“潜在特征”) 16有可能在晚年患上非酒精性脂肪肝。其次，我们将对模型进行检验和优化，以进行患者预测 NAFLD队列中的17种结局(发展为肝硬变、与肝脏相关的并发症和死亡)。我们 18假设机器学习方法可以用来进一步将患者分成不同的亚组 19种不同的疾病轨迹，目的是识别那些有进展性NAFLD风险的人和 20个与肝脏相关的结果。本申请中提出的研究具有创新性，因为它扩展了 21个分析工具箱，超越传统方法，使用所有健康接触识别患有NAFLD的个体 22个以人口为基础的大型队列中的22人，并进行了长期随访。这项建议意义重大，因为 23它解决了确定和管理最普遍的慢性肝病和 24为大规模实施筛查和风险分层策略提供了实用的解决方案 25个常规收集的数据。这项建议的最终目标是改善肥胖人群的健康-- 伴发疾病26例。

1 2 PROJECT SUMMARY/ABSTRACT 3 4 One of the most critical gaps in management of nonalcoholic fatty liver disease (NAFLD) is the lack of effective 5 methods of early identification in the population. The objective of this study is to leverage data and analytics to 6 improve healthcare outcomes by early detection and risk stratification of NAFLD, before onset of liver-related 7 complications. Artificial intelligence applications in large electronic health records have the potential to identify 8 disease traits before onset of disease. The central hypothesis of this proposal is that targeted screening with 9 machine-learning models applied to large integrated healthcare datasets can identify individuals with NAFLD 10 and, more specifically, those with a progressive phenotype. We will test the central hypothesis in 2 specific 11 AIMs. First, we will train a machine learning model of NAFLD prediction using multiple longitudinal data points 12 of all health-care encounters of a well-characterized population-based cohort of individuals diagnosed with 13 NAFLD in reference to individuals without NAFLD from the general population. We hypothesize that 14 unsupervised machine learning can identify complex processes and patterns without a human's guidance and 15 discover early comorbidity clusters (“latent traits” present prior to NAFLD development) that reflect a phenotype 16 at risk to develop NAFLD later in life. Second, we will test and optimize the model for the prediction of patient 17 outcomes (development of cirrhosis, liver-related complications and death) in the NAFLD cohort. We 18 hypothesize that machine learning approaches could be used to further stratify patients into subgroups with 19 different disease trajectories, with the goal of identifying those individuals at risk of progressive NAFLD and 20 liver-related outcomes. The research proposed in this application is innovative because it expands the 21 analytical toolbox beyond conventional methods to identify individuals with NAFLD using all health-encounters 22 of a large, well-characterized population-based cohort with long follow-up. This proposal is significant because 23 it addresses a critical need of identification and management of the most prevalent chronic liver disease and 24 offers a practical solution to large scale implementation of screening and risk-stratification strategies using 25 routinely collected data. The ultimate goal of this proposal is to improve the population health in obesity- 26 associated diseases.