Suicide as a contagion: modeling and forecasting emergent outbreaks

摘要 2016年，美国每个年龄段和几乎每个州的自杀率都在继续上升， 美国有44965人死于自杀。制定科学严谨的监控、报告和 自杀预测系统对于制定适当的公共卫生应对措施至关重要。在这里，我们将带来 来自Google Extended Trends、国家自杀预防生命线、Health Cost和 利用项目和生命统计与国家暴力死亡登记相结合，以建立和验证 关于询问、呼叫、未遂和自杀未遂的统计和数学模型。我们的模型旨在 通过将其作为一个可传染的过程进行研究来应对这一流行病。我们的首要目标是提供一个 预警系统，告知学校和社区的预防和治疗能力。 我们的第一个目标是估计自杀的询问、呼叫、企图和完成在空间和时间上是如何聚集的。 自杀行为的时间和空间自相关性与传染病有许多相似之处 这可以用流行病学三合会代理人(媒体报道、人与人之间的报道)来概念化 传播、致死媒介)、宿主(企图和精神障碍史)和环境(天气、 海拔和温度)-已使用严格的统计模型来研究地理和 即使在监测数据不完整的情况下，与传染病相关的临时风险因素也是如此。我们 将估计自杀查询、呼叫、未遂和完成的唯一和共享的自关联，以及 检验不同发育阶段(即青少年和青壮年)的自相关程度 与老年人相比)。我们的第二个目标是了解不同发育阶段自杀风险的动态变化 通过使用代表自杀的数学模型模拟异常自杀爆发的阶段 作为一种传染病。这些模型将与贝叶斯推理算法相结合以实现模拟， 优化和估计表征系统动态的关键流行病学参数。这一努力 将汇集4个数据源(查询、调用、尝试和完成)，以将系统视为 而不是作为单独的信息流。我们将回答有关以下方面的关键问题 自杀结果的局部和暂时异常增加在不同事件中是不同的，以及 传染性传播、传染性自杀危机的持续时间以及致命性手段的贡献。为我们的 第三个目标，我们将使用模型推理框架来产生细粒度的、局部的、6个月的自杀预测 爆发事件。生成的预测将帮助告知我们可以在何时、何地以及在哪些人中进行预测 除非迅速传播预防努力，否则自杀暴发将持续多久。这个项目 汇集了数学建模、传染病和自杀流行病学方面的专家， 预防和干预，世卫组织将把最先进的建模方法应用于自杀监测和干预 预测。

Abstract Suicide rates continue to increase in every age group in the United States and in almost every state—in 2016, 44,965 individuals in the US died by suicide. Developing scientifically rigorous surveillance, reporting, and forecasting systems for suicide is essential to craft appropriate public health responses. Here we will bring together geo-located data from Google Extended Trends, National Suicide Prevention Lifeline, Health Cost and Utilization Project, and vital statistics coupled with the National Violent Death Registry to build and validate statistical and mathematical models of queries, calls, attempts, and completions of suicide. Our models aim to address the epidemic by studying it as a communicable process. Our overarching goal is to provide an anticipatory warning system to inform school-based and community-based prevention and treatment capacity. Our first Aim is to estimate how suicidal queries, calls, attempts, and completions cluster in space and time. Temporal and spatial autocorrelation of suicidal behavior shares many features with communicable diseases that can be conceptualized in terms of the epidemiological triad—agent (media reporting, person-to-person transmission, lethal vectors), host (history of attempts and psychiatric disorder), and environment (weather, elevation, and temperature)—for which rigorous statistical models have been used to study geographic and temporal risk factors associated with infectious disease, even in the face of incomplete surveillance data. We will estimate the unique and shared autocorrelation of suicide queries, calls, attempts, and completions and test the extent to which autocorrelation varies by developmental stage (i.e., adolescents and young adults versus older adults). Our second Aim is to understand the dynamics of suicide risk across developmental stages through simulation of anomalous suicidal outbreaks using mathematical models that represent suicide as a contagion. The models will be coupled with Bayesian inference algorithms to enable simulation, optimization, and estimation of key epidemiological parameters that characterize system dynamics. This effort will bring together 4 data sources (queries, calls, attempts, and completions) to consider the system as a whole, rather than as separate streams of information. We will answer critical questions about the extent to which local and temporal anomalous increases in suicidal outcomes vary across events, as well as the force of contagious transmission, length of time of contagious suicidal crises, and contribution of lethal means. For our third Aim, we will use the model-inference framework to produce granular, local, 6-month predictions of suicide outbreak events. The generated forecasts will help inform when, where, and among whom we can expect suicide outbreaks to develop, and for how long unless prevention efforts are rapidly disseminated. This project brings together experts in mathematical modeling, communicable disease and suicide epidemiology, prevention, and intervention, who will apply state-of-the-art modeling approaches to suicide surveillance and forecasting.