Patterns and predictors of viral suppression: A Big Data approach

摘要 病毒抑制是艾滋病毒治疗级联的最后阶段，这是联合国艾滋病规划署的框架， 90-90-90进球。持续抑制病毒是“结束艾滋病毒流行”的四个战略领域之一 （EtHE）：美国计划”联邦运动，于2019年2月启动，旨在减少新的 到2025年和2030年，美国的艾滋病毒感染率分别下降75%和90%。EtHE运动 重点关注美国48个县和7个州，包括南卡罗来纳州（SC）。鉴于病毒的重要性 抑制结束美国艾滋病毒流行，病毒状态的最佳预测模型可以帮助临床医生 确定那些病毒控制不佳的风险，并为艾滋病毒治疗和护理的临床改进提供信息。各种 在文献中已经提出了表征纵向病毒学结果的指标， 持续的病毒抑制、病毒反弹、低水平病毒血症（LLV）、持续性LLV和病毒学信号。 然而，在我们开发病毒抑制的最佳预测模型的努力中仍然存在一些关键的差距。 这些差距包括使用有限的病毒学结果指标，随访时间有限， 数据来源，缺乏对结构和社会环境数据的考虑，数据量小或不具代表性 艾滋病毒感染者的样本，以及将研究结果转化为服务准备的努力有限 临床使用的工具。自2017年以来，在NIH的支持下（R 01 AI 127203），我们利用大数据方法， 检查治疗差距（例如，错过诊断和与护理联系的机会） 这项正在进行的研究提取了纵向电子健康记录数据，从六个 州机构然后将患者级数据与县级数据（例如，社会经济指标， 医疗保健专业人员、医院和医疗保健机构的数量） 源由此产生的综合数据库使我们能够成功地“跟踪”11，470名患者， 2005年至2016年在SC诊断为艾滋病毒，并确定艾滋病毒治疗联系和保留的差距。 基于R 01 AI 127203的经验和成就，我们提交此申请，以审查 病毒抑制的纵向动态模式，开发各种病毒的最佳预测模型 抑制指标，并将模型转化为临床使用的服务就绪工具。拟议 研究，我们将：1）继续“跟踪”我们的队列五年（并扩大队列， 增加2016-2020年诊断的艾滋病毒携带者）; 2）扩大我们的数据库，以包括关键艾滋病毒携带者的额外数据。 病毒抑制的预测因子（例如，治疗和实验室数据，酒精和物质使用数据） 新参与的全州数据源; 3）采用基于人工智能（AI）的建模来了解 动态病毒载量模式及其预测因素;以及4）制定并试点测试多因素决策 临床使用的系统。这些结果将能够识别病毒控制不良的PLWH，并建议 “何时”和“如何”帮助那些PLWH实现和维持病毒抑制。拟议的研究将 提高我们对病毒抑制的纵向动态的理解，并为定制的艾滋病毒护理提供信息 在SC和其他地区的PLWH之间进行管理。

Abstract Viral suppression is the final stage of the HIV treatment cascade, which serves as the framework for UNAIDS’ 90-90-90 goals. Sustained viral suppression is one of four strategic areas of the “Ending the HIV Epidemic (EtHE): A Plan for America” federal campaign, launched in February 2019, which aims for the reduction of new HIV infections in the United States (US) by 75% and 90% by 2025 and 2030, respectively. The EtHE campaign focuses on 48 US counties and 7 states, including South Carolina (SC). Given the importance of viral suppression in ending the US HIV epidemic, an optimal predictive model of viral status can help clinicians identify those at risk of poor viral control and inform clinical improvements in HIV treatment and care. Various indicators to characterize the longitudinal virologic outcomes have been proposed in the literature such as sustained viral suppression, viral rebound, low-level viraemia (LLV), persistent LLV, and virologic blips. However, some critical gaps still exist in our efforts to develop an optimal predictive model of viral suppression. These gaps include the use of limited indicators of virologic outcomes, limited duration of follow-up, limited data sources, lack of consideration of structural and socioenvironmental data, small or unrepresentative samples of people living with HIV (PLWH), and limited efforts to translate research findings into service-ready tools for clinical use. With NIH support (R01AI127203) since 2017, we have utilized a Big Data approach to examine treatment gaps (e.g., missed opportunities for diagnosis and linkage to care) among a statewide cohort of PLWH in SC. This ongoing research extracted longitudinal electronic health records data from six state agencies and then linked the patient-level data with county-level data (e.g., socioeconomic indicators, number of health care professionals, hospitals, and health care facilities) from multiple publicly available data sources. The resultant integrated database has enabled us to successfully “track” 11,470 patients who were diagnosed with HIV from 2005 to 2016 in SC and identify the gaps in HIV treatment linkage and retention. Based on the experience and accomplishment of the R01AI127203, we submit this application to examine the longitudinal dynamic pattern of viral suppression, develop optimal predictive models of various viral suppression indicators, and translate the models to service-ready tools for clinical use. In the proposed research, we will: 1) continue to “follow” our cohort for another five years (and also expand the cohort by adding PLWH diagnosed between 2016-2020); 2) expand our database to include additional data on critical predictors of viral suppression (e.g., treatment and laboratory data, alcohol and substance use data) from two newly participating statewide data sources; 3) employ artificial intelligence (AI)-based modeling to understand the dynamic viral load patterns and their predictors; and 4) develop and pilot-test a multifactorial decision system for clinical use. The results will enable the identification of PLWH with poor viral control and suggest “when” and “how” to help those PLWH achieve and maintain viral suppression. The proposed research will improve our understanding of the longitudinal dynamics of viral suppression and inform tailored HIV care management among PLWH in SC and beyond.