Electronic Health Record-based Surveillance of Diabetes by Type in Young Adults in Pennsylvania

1目前，美国有460万年轻人(18至44岁)患有糖尿病，其发病率 2和患病率在这一年龄段呈上升趋势。然而，由于传统监控的局限性， 3种策略，目前尚不清楚这些增加是在1型糖尿病还是在2型糖尿病中。电子健康 4基于记录(EHR)的监控是一种相对简单、可持续和及时的替代传统监控方式 5种方法。认识到以电子健康记录为基础的监测的属性，疾病控制中心和 6预防(CDC)资助了开发、评估和部署基于EHR的糖尿病监测的努力， 7包括青年糖尿病研究(SEARCH)和青年糖尿病研究(DIYA)。 8然而，这些研究的地理覆盖面有限。这些研究中的地理差距 9个是有问题的，因为已知糖尿病患病率和发病率存在地理差异。此外， 10关于这些研究中应用的方法将如何在该国其他地区发挥作用，目前知之甚少 11个农村社区和其他卫生系统。这项拟议的研究将使用二十多年的电子病历 12数据和行政索赔数据，以开发和实施基于电子健康记录的第1类和第2类监测 13在宾夕法尼亚州一大片地区的年轻人中患糖尿病，该州的糖尿病患病率位居第五 这一年龄段的糖尿病患者14例。这些信息对于制定公共卫生战略、评估疾病 15负担，并对特定类型的卫生服务进行优先排序。我们将使用来自医疗系统Geisinger的EHR数据 16服务于宾夕法尼亚州的一个大而多样化的地区，以扩大现有监视的地理范围 大西洋中部地区年轻人中的17种糖尿病亚型，这是一个以前没有EHR糖尿病的地区 18项监视估计。这一地区包括农村和城市社区，使我们能够 19评估基于EHR的算法在按社区类型确定病例时的性能差异。 20在研究的第一阶段，我们将评估基于电子健康记录的有效性、简单性和一致性 21识别糖尿病亚型的算法。这项工作将建立在以前开发的 22搜索和Diya研究。我们将使用临床医生记录的人工审查作为黄金标准来确定 23这些算法的阳性预测值、敏感性和特异性。我们建议使用一种创新的、 24高效、严格的验证方法，结合了对临床医生笔记的自然语言处理。我们 25将使用辅助数据源--行政索赔数据--来评估数据的完整性和我们的能力 26区分偶发病例和流行病例。在第二阶段，我们将用表现最好的 27种算法报告糖尿病的年发病率和患病率，按类型、年轻人、 2014年和2024年在宾夕法尼亚州的38个县。所有分析将按年龄、性别、种族/民族、保险 29地位和社区类型(农村/城市)。最后，我们将与疾控中心等地协调开展联合 对汇总数据进行了30次分析，大大扩大了研究对象。

1 Currently there are 4.6 million young adults (18 to 44 years of age) with diabetes in the US and the incidence 2 and prevalence are increasing in this age group. However, due to limitations of traditional surveillance 3 strategies, it remains unknown whether these increases are in type 1 or type 2 diabetes. Electronic health 4 record (EHR)-based surveillance is a relatively simple, sustainable, and timely alternative to more traditional 5 methods. Recognizing the attributes of EHR-based surveillance, the Centers for Disease Control and 6 Prevention (CDC) has funded efforts to develop, evaluate, and deploy EHR-based surveillance of diabetes, 7 including the SEARCH for Diabetes in Youth (SEARCH) study and the Diabetes in Young Adults (DiYA) Study. 8 However, the geographic coverage of these studies has been limited. The geographical gaps in these studies 9 are problematic, as there are known geographic disparities in diabetes prevalence and incidence. Moreover, 10 little is known about how the methods applied in these studies will perform in other regions of the country, in 11 rural communities, and in other health systems. The proposed study will use more than two decades of EHR 12 data and administrative claims data to develop and implement EHR-based surveillance of type 1 and type 2 13 diabetes among young adults in a large region of Pennsylvania, the state with the 5th highest prevalence of 14 diabetes in this age group. This information is essential to informing public health strategies, assessing disease 15 burden, and prioritizing type-specific health services. We will use EHR data from Geisinger, a health system 16 serving a large and diverse region of Pennsylvania, to expand the geography of existing surveillance of 17 diabetes subtypes in young adults to the Middle Atlantic, an area without prior EHR-based diabetes 18 surveillance estimates. This region includes a combination of rural and urban communities, enabling us to 19 evaluate differences in the performance of EHR-based algorithms for case ascertainment by community type. 20 In the first phase of the study, we will evaluate the validity; simplicity; and consistency of EHR-based 21 algorithms for identifying diabetes subtypes. This work will build on previously developed algorithms from the 22 SEARCH and DiYA studies. We will use manual review of clinician notes as the gold standard to determine the 23 positive predictive value, sensitivity, and specificity of these algorithms. We propose to use an innovative, 24 efficient, and rigorous validation approach that incorporates natural language processing of clinician notes. We 25 will use a secondary data source, administrative claims data, to assess data completeness and our ability to 26 distinguish between incident and prevalent cases. In the second phase, we will use the best performing 27 algorithms to report on the annual incidence and prevalence of diabetes, by type, in young adults, between 28 2014 and 2024 in 38 Pennsylvania counties. All analyses will be stratified by age, sex, race/ethnicity, insurance 29 status, and community type (rural/urban). Finally, we will coordinate with CDC and other sites to conduct joint 30 analyses of aggregated data, greatly expanding the population under study.