Harnessing Computerized Adaptive Testing, Transdiagnostic Theories of Suicidal Behavior, and Machine Learning to Advance the Emergent Assessment of Suicidal Youth (EASY).

摘要： 青少年急性自杀风险的紧急评估是一项艰巨的临床挑战，因为我们的 目前预测自杀未遂的能力很弱，而且因为自杀未遂的风险 青少年比例很高。尽管如此，目前还没有研究研究最好的方法来解决 到精神科急诊科就诊的自杀青年自杀行为的预测。至 为了解决这一研究差距，我们提出了一项针对1800名青年的研究，提交给地区PED，其中1350人 目前用于自杀风险评估，在PED中对青少年进行评估，并在1、3和6岁时进行跟踪 几个月的时间来确定哪些年轻人试图自杀。我们提出了三种互补的方法来 自杀风险评估。首先，在这场竞争性的更新中，我们在发展方面取得了成功 在我们之前的项目期间，对6个诊断小组进行了计算机化的适应性测试，加上自杀风险。 这些自我报告和家长报告总共可以在10-15分钟内完成。第二，因为理论驱动 自杀风险评估在成人中具有很强的预测力，但从未在 对于青少年，我们建议测试施奈德曼心理疼痛(精神疼痛)测量的预测力。 以及乔纳的人际自杀理论，该理论假定感知到的负担是交互作用的， 挫败了归属感，并获得了自杀的能力，从而导致了自杀风险。第三，我们的目标是使用机器 学习(ML)和自然语言处理(NLP)电子健康记录(EHR)以识别青年 自杀未遂的风险。我们假设这些方法中的每一种：(1)猫对自杀风险和对 抑郁症、焦虑症、躁郁症、多动症、对立违抗和品行障碍)；(2)基于理论的测量 (3)EHR的ML和NLP在以下方面将分别优于单独的临床评估 对尝试的预测，这3种方法的组合将比以下任何一种方法更强大 仅这些方法就可以。这项研究具有创新性，因为它是第一个使用CATS来预测 自杀风险，在持续的高危人群中，首次对两种主要的自杀理论进行了前瞻性测试 青少年，第一个使用机器学习和自然语言处理来确定EHR预测因素 在青少年中的自杀企图，并首次测试了识别自杀的方法的组合 在一个足够大的高危样本中，青少年中即将发生的自杀风险。这项研究的潜在价值很高。 影响，因为它可以确定简短、易于传播的评估战略，以识别高危青年 用于自杀行为，并增加临床医生将强度和类型的资源与最大限度地匹配的能力 临床需要。这项研究中要测试的方法可能会产生反映这两种情况的评估 紧急精神卫生保健的要求：简洁和准确。能够更好地识别谁处于危险之中 对于自杀行为，我们将处于更有利的地位，能够识别谁需要干预，并扭转 令人不安的是，十年来青少年自杀和自杀行为呈上升趋势。

Abstract: The emergency assessment of acute suicidal risk in adolescents is a daunting clinical challenge because our current ability to predict suicide attempts is weak, and because the risk for suicide attempts in suicidal adolescents is high. Nevertheless, there have been no studies that have examined the best approaches to the prediction of suicidal behavior in suicidal youth presenting to a psychiatric emergency department (PED). To address this research gap, we propose a study of 1800 youth presented to a regional PED, 1350 of whom present for evaluation of suicidal risk, in which youth are assessed in the PED, and followed up at 1, 3, and 6 months to determine which youth have made a suicide attempt. We propose 3 complementary approaches to assessment of suicidal risk. First, in this competitive renewal, we build on our success in developing computerized adaptive tests for 6 diagnostic groups, plus suicidal risk, during our previous project period. These self- and parent-reports can be completed in a total of 10-15 minutes. Second, because theory-driven assessments of suicide risk have strong predictive power in adults, but have never been tested prospectively in adolescents, we propose to test the predictive power of measures of Shneidman’s psychache (mental pain) and Joiner’s Interpersonal Theory of Suicide, which posits interactive roles of perceived burdensomeness, thwarted belonging, and acquired capacity for suicide in driving suicidal risk. Third, we aim to use machine learning (ML) and natural language processing (NLP) of electronic health records (EHRs) to identify youth at risk for suicide attempts. We hypothesize that each of these approaches: (1) CATs for suicide risk and for depression, anxiety, bipolar, ADHD, oppositional defiant, and conduct disorders); (2) theory-derived measures of suicidal risk; and (3) ML and NLP of EHRs, will each be superior to clinical assessment alone in the prediction of attempts, and that the combination of the 3 approaches will be more powerful than any one of these approaches alone. This study is innovative because it is one of the first to use CATs for the prediction of suicidal risk, in a consistently high risk population, the first prospective test of two leading theories of suicide in adolescents, the first to use machine learning and natural language processing to identify EHR predictors of suicide attempts in adolescents, and the first to test a combination of approaches to the identification of imminent suicidal risk in adolescents in a sufficiently large, high risk sample. The study is of potentially high impact because it could identify brief, easily disseminated assessment strategies to identify youth at high risk for suicidal behavior and add to clinicians’ ability to match intensity and type of resources to those at greatest clinical need. The approaches to be tested in this study could yield assessments that reflect the two imperatives of emergency mental health care: brevity and accuracy. With better ability to identify who is at risk for suicidal behavior, we will be in a much stronger position to identify who needs intervention and reverse the disturbing, decade-long trend of increases in adolescent suicide and suicidal behavior.