Automated Risk Assessment for School Violence Prevention

预防校园暴力的自动风险评估

• 批准号: 10096109
• 负责人: Drew Barzman
• 金额: $ 45.8万
• 依托单位: CINCINNATI CHILDRENS HOSP MED CTR
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10096109
• 关键词: Achievement; Adolescent; Aggressive behavior; Area; Attitude; Behavior; Characteristics; Childhood; Clinical; Crime; Data; Data Set; Descriptor; Development; Dropout; Early Intervention; Etiology; Evaluation; Family; Feeling; Foundations; Geographic Locations; Goals; Healthcare; Impulsivity; Individual; Institution; Interview; Judgment; Knowledge; Language; Legal Guardians; Linguistics; Machine Learning; Manuals; Medical center; Methods; Modeling; National Institute of Child Health and Human Development; Natural Language Processing; Observational Study; Outcome; Participant; Pediatric Hospitals; Performance; Population; Prevention; Prevention program; Process Assessment; Property; Psychiatrist; Public Health; Questionnaires; ROC Curve; Randomized; Recommendation; Reporting; Research; Research Priority; Risk; Risk Assessment; Risk Factors; Safety; Sampling; Schools; Scientist; Site; Structure; Students; System; Technology; Testing; Thinking; Time; Training; Universities; Validation; Violence; Work; Youth; base; bullying; clinical practice; computerized; cost; deep learning; design; emerging adult; experience; follow-up; high risk; impression; improved; individualized prevention; innovation; predictive modeling; prevent; prospective; recruit; school violence; school violence prevention; screening; study population; youth violence

PROJECT SUMMARY Acts of school violence have increased over the past decade and over 20% of students report being bullied at school. School violence has a far-reaching impact on the entire school population, including staff, students and families. It was noted that the largest crime-prevention results occurred when youth at elevated risk were given effective prevention programs. As such, there is a critical need for developing a rapid and accurate approach to interview students, assess their risk characteristics, and provide supportive evidence for prevention. Our study focuses on detecting and preventing youth aggression, the predominant form of school violence. Several risk assessment scales, ranging from simple clinical impressions to structured professional judgments, have been proposed to identify youth violence. However, these assessments heavily rely on clinicians' subjec- tive impressions and their predictive validities remain a major issue. In addition, none of the risk assessments include direct analysis of the words (language) used by students and hence, provide little information to sup- port subsequent prevention. Our long-term goal is to develop an Automated RIsk Assessment (ARIA) system to analyze participant interviews, detect elevated-risk students, and provide risk characteristics (e.g., impul- sivity, negative thoughts) to assist prevention. In our earlier study we developed a risk assessment approach to interview students and evaluate their risk of aggression. The overall objective of this study is to validate our risk assessment approach with real-world evidence, and to develop an AIRA system to automate the assessment process. We hypothesize that our risk assessment approach will have sufficient predictive validity in predicting aggression at school, and a computerized system leveraging machine learning and natural language pro- cessing (NLP) will be able to detect high-risk students, identify violence-related predictors from linguistic con- tent, and improve subsequent prevention by assisting recommendations. The hypothesis will be tested by pur- suing three specific aims: 1) Evaluate the predictive validity and generalizability of our risk assessment approach with prospectively collected school-based outcomes; 2) Develop a high-performing ARIA system to identify risk characteristics and predict risk of school violence; and 3) Compare actionable recommenda- tions and school outcomes with and without using the ARIA system in a prospective observational study. The study is highly innovative in that it will be among the first efforts that leverage NLP and machine learning to analyze interviews, identify risk characteristics from student language, and predict violence outcomes. The study will have a significant impact on several fronts. Successful validation of our risk assessment approach on multiple sites (Aim 1) will provide a valid mechanism to detect youth aggression at school. The AIRA system developed in Aim 2 will enable accurate and scalable risk screening for individual students. Aim 3 is a bench-to-practice translational aim to rapidly transfer our findings to clinical practice. The study will help establish a nationwide solution for school violence risk assessment, which will benefit healthcare institutions, schools, and students.

项目摘要 在过去的十年中，学校暴力行为有所增加，超过20％的学生报告被欺负 学校。学校暴力行为对整个学校人口有深远的影响，包括员工，学生和 家庭。值得注意的是，最大的预防犯罪结果是在给予较高风险的青年时发生的 有效的预防计划。因此，迫切需要开发快速准确的方法 采访学生，评估其风险特征，并为预防提供支持证据。 我们的研究重点是检测和预防青年侵略，这是学校暴力的主要形式。 从简单的临床印象到结构化的专业判断，几个风险评估量表， 已提议确定青年暴力。但是，这些评估在很大程度上依赖于临床医生的子名称 印象及其预测有效性仍然是一个主要问题。此外，没有风险评估 包括对学生使用的单词（语言）的直接分析，因此提供很少的信息来支持 随后的港口预防。我们的长期目标是开发自动风险评估（ARIA）系统 分析参与者的访谈，检测较高的风险学生并提供风险特征（例如，IMPUL- 态度，消极的想法），以帮助预防。在较早的研究中，我们开发了一种风险评估方法 采访学生并评估他们的侵略风险。这项研究的总体目的是验证我们的风险 带有现实世界证据的评估方法，并开发AIRA系统以自动化评估 过程。我们假设我们的风险评估方法将具有足够的预测有效性 学校的侵略性，以及一个计算机化系统利用机器学习和自然语言培训 CESSING（NLP）将能够检测到高危学生，从语言上确定与暴力相关的预测指标 帐篷，并通过协助建议来改善随后的预防。该假设将通过pur- 起诉三个具体目标：1）评估风险评估的预测有效性和普遍性 前瞻性收集的基于学校的成果的方法； 2）开发高性能的咏叹调系统 确定风险特征并预测学校暴力的风险； 3）比较可行的推荐 - 在前瞻性观察研究中，有或不使用ARIA系统的情况和学校成果。 这项研究具有很高的创新性，因为它将是利用NLP和机器学习的首批努力之一 分析访谈，从学生语言中确定风险特征，并预测暴力结果。研究 将对几个方面产生重大影响。成功验证我们的风险评估方法 站点（AIM 1）将提供一种有效的机制来检测学校的青年侵略性。 AIRA系统开发了 在AIM 2中，将为个别学生提供准确且可扩展的风险筛查。 AIM 3是练习台 转化目标迅速将我们的发现转移到临床实践中。这项研究将有助于建立全国 解决学校暴力风险评估的解决方案，这将使医疗机构，学校和学生受益。