Computational Models for the Prediction and Prevention of Child Traumatic Stress - Resubmission - 1

项目总结/摘要 至少40%的儿童会经历创伤性事件。在经历过创伤的人中，15-40%的人会 发展创伤后应激障碍（PTSD）和其他不良的精神、健康和功能结果 （以下称为儿童创伤性应激（CTS））。尽管对CTS的风险因素进行了数十年的研究，但该领域 没有达到特定的风险因素模型，可以准确预测CTS结果的可能性或识别 这些因素--如果改变的话--会改变他们的可能性。了解导致 根据定义，结果的变化是因果关系。文献中关于CTS风险的绝大多数发现 我不能提供这样的因果知识，因为这样的发现是基于应用相关性 观察数据的方法。实验研究不能-出于所有实际目的-进行， 人类研究CTS风险。因此，该领域只剩下相关的观测研究， 关于CTS风险的经验知识的唯一生成器，这种知识不适合指导 必须采取的行动（即干预措施），以改变儿童获得CTS结果的可能性。我们 我建议通过应用能够使自信的因果关系成为可能的方法来解决这一巨大的进步障碍。 使用包含广泛多样的CTS风险变量的大型观察数据集进行推断。机 学习（ML）预测和因果建模方法将被应用于发现因果关系， 从观测数据中测量变量：并从这种确定的因果关系中估计 当操纵因果变量时（即干预模拟）对CTS结果的影响。我们将建立 与文献中的儿童创伤相关的结果模型， 儿童的福祉，功能和发展：创伤后应激障碍，抑郁症，药物滥用，健康， 教育绩效。

Project Summary/Abstract At least 40% of children will experience a traumatic event. Of those who experience a trauma, 15-40% will develop Posttraumatic Stress Disorder (PTSD), and other adverse psychiatric, health, and functional outcomes (herein called Child Traumatic Stress - CTS). Despite decades of research on risk factors for CTS, the field has not arrived at specific risk factor models that can accurately predict the likelihood of CTS outcomes or identify factors that – if changed – would change their likelihood. Knowledge about changes in factors that result in changes in outcomes is, by definition, causal. The vast majority of findings in the literature on risk for CTS cannot provide such causal knowledge because such findings were based on the application of correlational methods to observational data. Experimental research cannot – for all practical purposes - be conducted for human research on risk for CTS. Thus, the field is left with correlational observational research as the near exclusive generator of empirical knowledge on risk for CTS, and such knowledge is unsuitable to guide the actions (i.e. interventions) that must be taken to change children's likelihood of acquiring CTS outcomes. We propose to address this considerable barrier to progress by applying methods that can enable confident causal inference with large observational data sets containing a broad diversity of risk variables for CTS. Machine Learning (ML) predictive and causal modeling methods will be applied to discover causal relationships among measured variables from observational data: and from such determined causal relationships, to estimate the effect on a CTS outcome when a causal variable is manipulated (i.e. intervention simulation). We will build models for outcomes associated with childhood trauma in the literature and that entail significant burden to children's well-being, functioning, and development: PTSD, Depression, Substance Abuse, Health, and Educational Performance.