Digital Phenotyping & Deep Learning: Substance Use Impact on PrEP Adherence among Black Sexual and Gender Minorities

美国一直在与多种流行病作斗争，特别是持续的艾滋病毒流行，主要影响黑人、性少数族裔男性和性别少数群体。这些群体的艾滋病毒发病率和流行率都有所上升。与此同时，随着几个州倾向于大麻合法化，我们看到大麻使用量激增。吸食大麻的黑人和少数性别群体表现出高水平的高危行为，包括酗酒、受影响的无保护性行为以及与可能艾滋病毒阳性的伴侣发生性行为。有趣的是，现代的地理社交网络和约会应用程序正在加入一些功能，允许用户专门与大麻使用者进行匹配。这种性别化的大麻使用可能会放大艾滋病毒的传播，因为没有公寓的性行为参与度更高，性伴侣数量增加。 由于艾滋病毒暴露前预防(PrEP)在依从者中显示出近95%的有效性，它成为抗击这一流行病的潜在斗士。尽管有两种PrEP制剂可供使用，但口服和长效注射在目标人群中的知晓率、使用率和依从性仍然低得惊人。初步数据表明，黑人、性少数群体男性和性别少数群体的坚持取决于他们的药物使用情况。 为了解决这些复杂的交叉问题，我们对艾滋病毒传播风险、大麻消费和PrEP结果之间的关系进行了全面检查。这将利用创新的数字表型、社交媒体语言和在线行为、事件级别的衡量标准以及关键的生物标志物数据。鉴于智能手机无处不在(超过95%的目标群体使用)，这些设备被选为持续监测和数据收集的设备。我们部署了一个由我们实验室开发并之前经过验证的应用程序，以积累被动移动数据并管理及时的生态瞬时评估。利用尖端的深度学习技术，我们将把这些数据合成成动态的数字表型，能够向参与者分配每日的风险评分。首要目标是确定大麻消费与两个关键结果之间的相关性：危险的性行为和减少遵守PrEP。 为了实现上面列出的目标，我们将跟踪纵向使用大麻的黑人少数性群体和性别少数群体(与男性发生性关系)，以生成一个数据集，该数据集将用于训练数字表型背后的算法。这包括审查关于性别化大麻使用和/或PrEP意图和态度的在线通信。该项目将涉及以下概念： 识别大麻使用和嗜好的高级计算算法 我们将使用先进的计算算法来预测大麻的使用，包括性别化的大麻使用和对大麻的渴望。我们的第一个目标是对机器学习算法进行比较分析，以确定我们可以用来提供准确分类的最佳分类器。我们的第二个目标是实现一个特征选择算法，该算法将从临床、EMA和数字表型数据中提取最相关的特征，提供对大麻渴望和使用的最佳分类，特别是在性行为方面。 人工智能识别PrEP不遵守和危险性行为 高危性行为和不遵守PrEP的原因有很多，包括吸食大麻。解决这些问题的最佳方法将涉及能够感知背景并为其量身定做的大数据。数字表型数据包含丰富的信息集，包括人口统计数据、情绪、性行为、社会支持、约会行为、物质使用和地点。我们还将获得临床健康记录。结合起来，我们将能够开发和验证PrEP遵守和HIV风险人工智能(AI)工具，以识别谁有PrEP不遵守和高风险性行为的风险。这个人工智能工具将被编程为在检测到风险上升时提供定制的消息传递。 总之，该项目深入探讨了大麻使用的错综复杂的动态、其社会技术影响及其与艾滋病毒风险和遵守PrEP的交集。这些方法和成果对于努力为这一多方面挑战寻求全面解决办法的主要利益攸关方来说将是无价的。

The United States has grappled with multiple epidemics, notably the persistent HIV epidemic, predominantly affecting Black sexual minority men and gender minorities. These groups register elevated HIV incidence and prevalence. Parallelly, as several states are gravitating towards cannabis legalization, we witness a surge in its usage. Black sexual and gender minorities consuming cannabis have shown heightened levels of high-risk behaviors, including binge drinking, unprotected sex under the influence, and intercourse with partners potentially HIV-positive. Intriguingly, modern geosocial networking and dating applications are incorporating features allowing users to match specifically with cannabis users. This blend of sexualized cannabis use may be amplifying HIV transmission due to heightened engagement in condomless sex and an increased number of sexual partners. With HIV pre-exposure prophylaxis (PrEP) showcasing near 95% efficacy in adherent individuals, it emerges as a potential combatant against this epidemic. Despite having two PrEP formulations availabledaily oral and long-acting injectablethe awareness, usage, and adherence amongst the target demographic remains alarmingly low. Preliminary data suggests that adherence among Black sexual minority men and gender minorities is contingent on their substance use. To address these complex intersections, we use a holistic examination of the relationships among HIV transmission risk, cannabis consumption, and PrEP outcomes. This would harness innovative digital phenotyping, social media language and online behaviors, event-level measures, and critical biomarker data. Given the ubiquity of smartphones (used by over 95% of the target group), these devices were chosen for continuous monitoring and data collection. An application, developed by our lab and previously validated, was deployed to accumulate passive mobile data and administer timely ecological momentary assessments. Employing cutting-edge deep learning techniques, we will synthesize this data into dynamic digital phenotypes, capable of allotting daily risk scores to participants. The overarching objective is to ascertain correlations between cannabis consumption and two pivotal outcomes: risky sexual practices and diminished PrEP adherence. To accomplish the goals listed above, we will follow Black sexual minorities and gender minorities (who have sex with men) who use cannabis longitudinally to generate a dataset that will be used to train the algorithm behind the digital phenotypes. This includes examining online communication regarding sexualized cannabis use and/or PrEP intentions and attitudes. This project will address the following concepts: Advance Computational Algorithms to Identify Cannabis Use and Cravings We will use advanced computational algorithms to predict cannabis use, including sexualized cannabis use, and cannabis cravings. Our first goal will be to carry out a comparative analysis of machine learning algorithms to determine the optimal classifier we can use to provide accurate classifications. Our second goal is to implement a feature selection algorithm that will extract the most relevant features (from clinical, EMA, and digital phenotype data) that provide the best classification of cannabis cravings and use, especially in the context of sexual behavior. Artificial Intelligence to Identify PrEP Non-Adherence and Risky Sexual Behaviors High-risk sexual behaviors and non-adherence to PrEP occur for many reasonsincluding cannabis use. The best approach to tackling these problems will involve big data that is context-aware and individually tailored. Digital phenotype data contains a rich set of information that includes demographics, mood, sexual behavior, social support, dating behavior, substance use, and location. We will also have access to clinical health records. In combination, we will be able to develop and validate a PrEP adherence and HIV risk artificial intelligence (AI) tool to identify participants who are at risk for PrEP non-adherence and risky sexual encounters. This AI tool will be programmed to deliver tailored messaging when elevations in risk are detected. In summary, this project delves into the intricate dynamics of cannabis use, its socio-technological implications, and its intersection with HIV risk and PrEP adherence. The methodologies and results would be invaluable for key stakeholders striving for comprehensive solutions to this multifaceted challenge.