Predicting Self-Harm, Suicide Attempt, and Suicidal Death using Longitudinal EHR, Claims and Mortality Data

项目摘要 自杀是导致死亡的主要原因之一。截至2015年，美国经年龄调整后的年自杀率为13.26 每10万人中，平均每天有121人自杀。而45岁至之间的白人男性 5岁的人自杀死亡的可能性是女性的4倍，女性自杀的几率是女性的3倍 男性。最近的数据表明，自杀企图是自杀企图的20倍，这通常被认为是一种 随后自杀的高危因素和一贯的危险因素。然而，预测和监控某人何时会 企图自残和自杀几乎是不可能的。在这个项目中，我们计划利用大规模、 来自纽约市临床数据研究网络的综合电子健康记录和索赔研究 与急诊科就诊或住院有关的自杀行为。特别是，使用&gt；10上的数据 ，我们将开发新的NLP和机器学习模型来识别处于最高风险的患者 自残、自杀未遂和自杀，并进行一项试点研究，以评估这些模型的临床实用性。我们 还将使用来自Kaiser Permanente Washington的类似数据进行验证研究。

PROJECT ABSTRACT Suicide is one of the leading causes of death. As of 2015, annual age-adjusted suicide rate in the U.S. is 13.26 per 100,000 individuals, and on average, there are 121 suicides per day. While white males between 45 and 64 years of age are 4 times more likely than females to die by suicide, females attempt suicide 3 times as often as males. Recent data suggest that there are 20 times as many suicide attempts, which is generally considered a high and consistent risk factor for subsequent suicide. However, predicting and monitoring when someone will attempt self-harm and suicide has been nearly impossible. In this project, we plan to leverage large-scale, integrated electronic health record and claims from the New York City Clinical Data Research Network to study the suicidality in relation to emergency department (ED) visits or hospitalizations. In particular, using data on >10 million patients, we will develop novel NLP and machine learning models to identify patients at highest risk for self-harm, suicide attempt and suicide, and conduct a pilot study to assess the clinical utility of such models. We will also conduct a validation study using similar data from Kaiser Permanente Washington.