Multisite Electronic Health Record-Based Surveillance of the Burden of Diabetes by Type in Young Adults

目前，美国约有300万年龄在18-44岁的年轻人患有糖尿病。这 预计到2060年，这一数字将增加到约580万。区分糖尿病类型至关重要，因为 糖尿病的病因、治疗和结果因类型而有很大不同。1型糖尿病(T1D)占 ~17%和2型糖尿病(T2D)~75%的美国青年糖尿病患者。糖尿病类型的这种分布 不断进化。我们没有一个大规模的监测系统来监测流行率和 美国青壮年T1D和T2D的发病率。电子产品的广泛使用和日益增长的功能 健康记录(EHR)系统极大地提高了数据的数量、广度和及时性 与基于人口的登记和调查相比，监测和降低成本。EHR算法具有 在识别糖尿病病例方面显示出巨大的潜力。这项研究将分析结构化的EHR数据(例如， 诊断代码、药物和实验室结果)和非结构化临床记录。我们将申请专家 知识、机器学习和自然语言处理，以开发识别的最佳算法 流行和偶发的T1D和T2D病例。本研究的主要目标是建立以电子病历为基础的 监测美国青壮年T1D和T2D负荷的监测系统。我们将与3家公司合作 EHR研究网络来自国家以患者为中心的临床研究网络(PCORnet)，涵盖 约600万来自4个州(伊利诺伊州、洛杉矶、纽约州和德克萨斯州)的不同种族、民族和社会经济背景的年轻人 3个人口普查区域。这项研究中的患者群体大致代表了 集水区。这项研究的具体目的是1)估计T1D和T2D在美国的患病率 2019年按年龄、性别、种族/民族和地理区域划分的青壮年；2)估计T1D的发病率 和2019年按年龄、性别、种族/民族和地理区域划分的美国年轻人的T2D；3)估计10年 美国年轻人中T1D和T2D的患病率和发病率按年龄、性别、种族/民族和 2014-2023年地理区域；以及4)按类型比较糖尿病患病率和发病率 美国年轻人的时间趋势与其他国家和地区的年轻人的趋势。这项研究是 创新，因为它将检测到低至0.2%的假阴性率，利用EHR进行监控(更多 比登记和调查更有效率和成本效益)，使用先进的统计方法(例如，机器 学习和自然语言处理)，使用患者邮政编码估计分母，为 根据当地可获得的临床病历采取监测方法，并采用两阶段抽样方法 提高审图效率。这项研究将促进我们对年龄、性别、种族/民族和 美国年轻人T1D和T2D负担的地理差异。所获得的监视数据将 为医疗保健需求规划提供信息，确定医疗保健资源分配的优先顺序，并减少健康 通过确定预防糖尿病和相关合并症的亚群并确定其优先顺序，来区分不同人群之间的差异。

Approximately 3 million young adults aged 18-44 years currently have diabetes in the United States. This number is projected to increase to ~5.8 million by 2060. Differentiating diabetes types is crucial, because the etiology, treatments, and outcomes of diabetes differ substantially by type. Type 1 diabetes (T1D) accounts for ~17% and type 2 diabetes (T2D) ~75% of total diabetes in US young adults. This distribution of diabetes types continuously evolves. We do not have a large-scale surveillance system to monitor the prevalence and incidence of T1D and T2D in US young adults. The widespread use and increasing functionality of electronic health record (EHR) systems substantially increase the quantity, breadth, and timeliness of data available for surveillance and reduce costs compared with population-based registries and surveys. EHR algorithms have shown great potential in identifying diabetes cases. This study will analyze both structured EHR data (e.g., diagnosis codes, medications, and laboratory results) and unstructured clinical notes. We will apply expert knowledge, machine learning, and natural language processing to develop the best algorithms for identifying prevalent and incident T1D and T2D cases. The primary objective of this study is to establish an EHR-based surveillance system for monitoring the burden of T1D and T2D in US young adults. We will collaborate with 3 EHR research networks from the National Patient-Centered Clinical Research Network (PCORnet), covering ~6 million racially, ethnically, and socioeconomically diverse young adults from 4 states (IL, LA, NY, and TX) in 3 Census regions. The patient populations in this study are roughly representative of the source populations in the catchment areas. The specific aims of this study are 1) to estimate the prevalence of T1D and T2D in US young adults by age, sex, race/ethnicity, and geographic region in 2019; 2) to estimate the incidence of T1D and T2D in US young adults by age, sex, race/ethnicity, and geographic region in 2019; 3) to estimate 10-year trends in the prevalence and incidence of T1D and T2D in US young adults by age, sex, race/ethnicity, and geographic region, 2014-2023; and 4) to compare the prevalence and incidence of diabetes by type, as well as temporal trends, in US young adults with those in young adults from other countries and regions. This study is innovative, because it will detect a false negative rate as low as 0.2%, leverage EHRs for surveillance (more efficient and cost-effective than registries and surveys), use advanced statistical approaches (e.g., machine learning and natural language processing), estimate a denominator using patient zip codes, build flexibility into the surveillance methods according to local availability of clinical notes, and use a 2-staged sampling approach to improve chart review efficiency. This study will advance our understanding of the age, sex, racial/ethnic, and geographic differences in the burden of T1D and T2D in US young adults. The obtained surveillance data will inform planning for healthcare needs, prioritize the allocation of healthcare resources, and reduce health disparities via identifying and prioritizing subpopulations for prevention of diabetes and related comorbidities.