Advancing Real-Time Suicide Risk Detection Through the Digital Phenotyping Smartphone Application Screenomics

项目总结/摘要 自杀是一个主要的公共卫生问题，仅在2017年就有超过45，000人死亡。与自杀 率继续上升2，自杀想法和行为（STBs）的预测保持停滞3， 有必要将重点从确定谁有自杀风险转移到个人何时有自杀风险。研究 利用生态瞬时评估每天几次收集数据表明， 自杀意念和STB风险因素在第4天的过程中迅速变化;然而，需要改善 评估的粒度，以提高对实时风险升级的识别。为了能够可靠地检测 在相对短的时间窗口内的STB（例如，将需要技术创新的方法 可以持续捕捉自杀风险的动态本质。 我们建议使用一种新形式的数字表型，称为Screenomics 5-6， 每五秒就有一条来自参与者手机的短信然后，这些数据可以用于在真实的环境中间接识别STB。 时间（通过生成和查看的文本），以及前瞻性地预测STB通过个人参与 产生和消费社交互动（通过应用程序使用，短信和社交媒体文本）， 有已知的链接到机顶盒7.在80个人与过去一个月的STB，两个主要目标将进行调查。 目的1是证明通过智能手机使用收集的文本（即，web浏览器、文本消息）可以提供 作为直接评估STB的准确替代。目标2将确定预期的短期STB风险 与生产和消费的社会互动相关，而不是通过直接评估来证明。 研究小组（联合PI：Ammerman，Jacobucci;联合I：Jiang;顾问：Kleiman，Ram，罗宾逊， Reeves，Bourgeois，Liu）拥有世界一流的专业知识，在EMA数据收集方面拥有丰富的经验 在高风险样本中，机器学习用于预测自杀，收集和建模连续数据流， 包括屏幕截图数据，以及技术创新所特有的道德和隐私惯例。 为了有意义地降低自杀率，需要对性传播疾病和相关风险因素有更细致的了解， 需要实时。Screenomics提供近乎连续的监测，允许更接近 风险因素和性传播疾病之间的真实联系。的确，有必要确定近期风险因素 在STB发生之前成功进行干预并防止STB。这些发现将奠定 在STB检测和干预中利用被动数据的必要基础。鉴于严重的个人和 自杀的社会成本，这项工作具有重要的公共卫生意义。

PROJECT SUMMARY/ABSTRACT Suicide is a leading public health problem, accounting for over 45,000 deaths in 2017 alone 1. With suicide rates continuing to rise 2, and the prediction of suicidal thoughts and behaviors (STBs) remaining stagnant 3, there is a need to shift the focus from identifying who is at risk to when individuals are at risk for suicide. Studies utilizing ecological momentary assessment to collect data at several intervals per day have demonstrated that suicidal ideation and STB risk factors change rapidly across the course of the day 4; yet, there is a need to improve the granularity of assessment to improve identification of real-time risk elevation. To enable reliable detection of STBs within a relatively short window of time (e.g., minutes) will require technologically innovative methodologies that can continuously capture the dynamic nature of suicide risk. We propose the use of a novel form of digital phenotyping, termed Screenomics 5-6, that captures screenshots from participant’s phones every five seconds. These data can then be utilized to indirectly identify STBs in real- time (via generated and viewed text), as well as prospectively predict STBs via individual engagement in produced and consumed social interactions (via application usage, text messages, and social media text), which have knowns links to STBs 7. Among 80 individuals with past-month STBs, two primary aims will be investigated. Aim 1 is to demonstrate that text collected through smartphone use (i.e., web browser, text messages) can serve as an accurate proxy for the direct assessment of STBs. Aim 2 will identify prospective, short-term STB risk associated with produced and consumed social interactions not demonstrated via direct assessment. The research team (Co-PIs: Ammerman, Jacobucci; Co-I: Jiang; Consultants: Kleiman, Ram, Robinson, Reeves, Bourgeois, Liu) has access to world-class expertise, with extensive experience in EMA data collection in high-risk samples, machine learning for predicting suicide, collecting and modeling continuous data streams, including screenshot data, and ethical and privacy practices unique to technological innovations. To meaningfully reduce suicide rates, a more nuanced understanding of STBs and associated risk factors in real-time is required. Screenomics provides near continuous monitoring, allowing for a closer approximation of the true associations between risk factors and STBs. Indeed, there is a need to identify near-term risk factors prior to STB occurrences to successfully deliver an intervention and prevent STBs. These findings will lay the groundwork necessary for utilizing passive data in STB detection and intervention. Given the grave personal and societal cost of suicide, this work has important public health implications.