Health Informatics to Model the Scott County HIV Outbreak

摘要 美国的许多地区，特别是农村和边境地区，缺乏资源， 确定孤立的传染病爆发。印第安纳州斯科特县在2014年经历了艾滋病毒爆发- 2015年，约有200人感染艾滋病毒。如果不是一个警惕的疾病干预专家 (DIS)他注意到斯科特县的艾滋病毒感染人数在短时间内不断上升， 疫情可能会更糟斯科特县等资源有限地区的孤立疫情 强调了在非城市地区建立更具创新性、自动化和实时的艾滋病毒生物监测系统的必要性。 地区迄今为止，数字艾滋病流行病学研究几乎完全依赖于Twitter;然而， 这种方法可能限制性太大，尚未产生预测艾滋病毒爆发的有希望的方法。 除了社交媒体之外，更多的异构数据来源可能会更有效地预测艾滋病毒的到来 在一个监控资源有限的社区拟议的健康信息学研究将分析 从2014年到2016年收集的历史时间序列数据，以确定模型斯科特的预测变量 县艾滋病爆发。分析的数据包括：（1）急诊室（ER）入院和出院 与阿片类药物使用和与药物滥用有关的软组织感染有关的数据;（2）艾滋病毒检测监测数据;（3）丙型肝炎病毒 发病率数据;（4）搜索引擎查询相关主题，如Google或Bing搜索“HIV检测” （5）执法逮捕记录（特别是与持有类阿片有关的记录， 分发）;以及（6）HIV相关治疗的电子医疗保健报销数据（例如，暴露后 预防）。自动数据/文本挖掘和机器学习技术也将应用于（7）社交 媒体数据（即，Twitter推文和Reddit论坛帖子），其中提到艾滋病毒，Opana，物质使用， 和其他术语来确定社交媒体数据的趋势是否可以预测艾滋病毒的到来， 整个传输，斯科特县。使用上面列出的各种数据源，我们的团队将 关键预测变量的时间序列，以确定已知数据中最具预测性的数据源（或多个数据源） 艾滋病爆发。如果我们的团队开发出一种健康信息学方法和算法， 媒体和其他电子数据表明艾滋病毒即将爆发，州和县卫生部门可以 利用这些“信号”，在受影响地区增加艾滋病毒检测和咨询点的数量， 提供者可以更积极地筛查HIV/STI感染， 更快速、更有效地针对性地开展接触者追踪活动， 在有关地理区域，为有感染艾滋病毒风险的人开处方。本研究还将研究 与收集和分析电子卫生信息和社会媒体数据有关的可行性问题， 艾滋病毒生物监测工作，如数据的可访问性，成本和普遍性。

ABSTRACT Many regions of the United States, particularly rural and frontier areas, lack the resources to proactively identify isolated infectious disease outbreaks. Scott County Indiana experienced an HIV outbreak in 2014- 2015, resulting in approximately 200 new HIV infections. If not for a vigilant Disease Intervention Specialist (DIS) who noticed an escalating number of HIV infections in Scott County over a brief period of time, the outbreak could have been much worse. An isolated outbreak in resource-limited settings such as Scott County underscores the need for more innovative, automated, and real-time HIV biosurveillance systems in non-urban areas. To date, digital HIV epidemiologic research has relied almost exclusively on Twitter; however, this approach is likely too restrictive and has not yet yielded a promising approach to predicting HIV outbreaks. More heterogeneous sources of data--in addition to social media--may more efficiently predict the arrival of HIV in a community with limited surveillance resources. The proposed health informatics research will analyze historical time series data collected from 2014 through 2016 to identify predictor variables that model the Scott County HIV outbreak. Data to be analyzed include: (1) emergency room (ER) admissions and discharges related to opioid use and soft tissue infections related to drug abuse; (2) HIV testing surveillance data; (3) HCV incidence data; (4) search engine inquires of relevant topics, such as Google or Bing searches for “HIV testing” and “Opana”; (5) law enforcement arrest records (particularly those related to opioid possession and distribution); and (6) electronic healthcare reimbursement data of HIV-related treatments (e.g., post-exposure prophylaxis). Automated data/text mining and machine learning techniques will also be applied to (7) social media data (i.e., Twitter tweets and Reddit forum posts) that make reference to HIV, Opana, substance use, and other terms to determine if trends in social media data could have predicted HIV's arrival in, and transmission throughout, Scott County. Using the diverse data sources listed above, our team will correlate the time series of key predictor variables to identify the data source (or sources) most predictive of a known HIV outbreak. If our team develops a health informatics approach and algorithm(s) identifying trends in social media and other electronic data indicating an imminent HIV outbreak, state and county health departments can use these “signals” to increase the number of HIV testing and counseling sites in the affected area, health care providers can more aggressively screen for HIV/STI infection, syringe-service programs can be mobilized rapidly and targeted more efficiently, contact tracing activities can be initiated, and PEP and PrEP can be prescribed to those at risk for HIV infection in the geographic region of concern. This study will also examine feasibility issues related to the collection and analysis of electronic health information and social media data in HIV biosurveillance efforts, such as data accessibility, costs, and generalizability.