Modeling the Co-evolution of Substance Use behavior and peer Networks of risk/support (CoSUN)

多年来，美国吸毒过量死亡人数持续增加，死亡人数超过7万人 2019年。与药物使用和滥用有关的犯罪、医疗保健和工作损失的经济成本 美国的生产率每年超过6000亿美元。为了帮助个人并减少 相关联的成本效益，我们必须将适当的资源分配给最大限度的个人 新出现的需求(例如，未来大量使用物质的需求)。然而，目前还没有数据驱动的 允许利益相关者(随机)预测个人的药物使用情况的工具。当前 方法论只依赖趋势分析，建立风险因素(例如朋友)之间的相关性 使用毒品)和物质使用。我们在模拟未来物质使用时面临的一个关键挑战 在于行为(即吸毒)和同伴风险/支持网络的共同进化，这可以 随着时间的推移而变化，并可能相互依赖。为了解决这一科学障碍，我们认为 创新方法，使物质使用和同伴风险/支持的共同演变过程脱钩 通过(目标1A)首先模拟个人属性(例如，吸毒和不良童年 历史)以及他们的同伴网络(例如，同伴和知己吸毒的程度)影响 个人物质使用行为和(目标1B)然后建模个人的同伴风险/支持 网络链路响应于端点属性中的相似或不同而形成或断开(例如， 他们的吸毒行为)。因此，这个项目第一次寻求发展随机预测 长期的未来物质使用(FSU)和未来同行风险/支持网络(FPN)模型 在几个月内(目标1-2)，对于FSU，在几天内以短时间尺度(目标3)使用协变量数据 个人属性和对等网络特征。我们将使用成功的机器学习方法来 建立这些模型并严格评估模型泛化/预测性能(目标1-3) 利用从模型构建过程中保留的数据。AIMS将提供一个基础 对于未来的创新NIH R01，它开发了现实中与SUD相关的行为的随机模拟 在看似合理的动态网络中的传染性，为资源分配和应急计划提供信息。这 该项目还为及时干预奠定了基础，以检测FSU和 使用这些风险预测来触发社会和行为干预的交付。

Drug overdose deaths in the U.S. have continued to increase over the years, with over 70,000 deaths in 2019. The economic cost of substance use and abuse related crimes, healthcare, and loss in work productivity in the U.S. exceeds $600 billion each year. In order to aid individuals and reduce the associated cost effectively, we must allocate appropriate resources to individuals in the greatest emerging need (e.g., those of high future substance use). However, there are currently no data-driven tools that allow stakeholders to (stochastically) forecast an individual's substance use. Current methodologies only rely on trend analyses, establishing correlations between risk factors (e.g., friends that use drugs) and substance use. A key challenge we are facing when modelling future substance use lies in the co-evolution of behaviors (i.e., drug use) and peer networks of risk/support, which can change over time and may depend on each other. To address this scientific obstacle, we consider an innovative approach that decouples the co-evolution process of substance use and peer risk/support networks by (Aim 1A) first modelling how individual attributes (e.g., drug use and adverse childhood history) along with their peer networks (e.g., the extent of peer and confidant drug use) impact the individual’s substance use behavior and (Aim 1B) then modeling how individuals’ peer risk/support network links form or break in response to similarities or differences in the endpoints’ attributes (e.g., their drug use behaviors). Thus, for the first time, this project seeks to develop stochastic forecasting models for future substance use (FSU) and future peer risk/support networks (FPN) at long timescales within months (Aims 1-2) and for FSU at short timescales within days (Aim 3) using data on covariates of individual attributes and peer network features. We will use successful machine-learning methods to build these models and rigorously assess model generalizability/prediction performance (Aims 1-3) by making use of data that is held-out from the model building process. The Aims will provide a foundation for a future innovative NIH R01 that develops stochastic simulations of realistic SUD-related behavioral contagion in plausible dynamic networks, to inform resource allocation and contingency planning. This project also lays the groundwork for just-in-time interventions to detect imminent increases in FSU and use these risk forecasts to trigger the delivery of social and behavioral interventions.