Prediction of suicide death using EHR and polygenic risk scores

摘要 自杀是死亡的主要原因，而且还在继续增加，每年有超过47000人死于可预防的自杀 尽管我们在使用电子健康记录(EHR)和其他方面取得了很大进展 预测自杀意念和行为的因素，我们可靠地预测自杀死亡的能力接近于零。 从医疗保健的角度来看，预测自杀死亡是一件棘手的事情。我们知道自杀的发生率 与自杀死亡(每年约0.01%-0.02%)相比，自杀行为更常见(每年约4%-5%)。 从本质上讲，只有一小部分从事自杀行为的人会最终自杀身亡。 了解这些最高风险的人是谁，对于指导预防工作和 制定未来有针对性的干预措施。此外，超过一半的自杀死亡是在没有前科的情况下发生的 尝试，甚至考虑到在诊断代码中缺少尝试的记录。这些“突如其来的” 病例表明，一个或多个高危群体更难以准确预测和预防。 包括自杀死亡的基因数据可能会提供实质性的预测改进；遗传因素解释 近50%的自杀死亡风险。利用广泛的遗传数据，全州范围内的纵向EHR 犹他州自杀遗传风险研究(USGRS)可用的资源、人口统计和家庭数据，我们是 独一无二地准备解决这一关键的知识差距。我们的主要重点将是使用机器学习 方法开发预测自杀死亡的模型。此外，我们的大型自杀死亡研究资源将 这也让我们可以模拟有没有自杀企图的自杀死亡人数的差异。犹他州约9000人自杀 死亡人数与人口统计和环境数据、家庭数据和20年的纵向电子健康记录数据一起， USGRS目前还拥有来自6,000英镑的DNA，在获奖期间将增加到约10,000英镑。基因组- 犹他州5000多人自杀的广泛分子数据已经掌握，允许测试自杀的关联性 使用EHR数据识别具有多基因风险分数所代表的“遗传表型”的亚型。这个 USGRS也有来自5个年龄/性别匹配的犹他州的人口统计数据、家庭数据和纵向EHR数据 每一次自杀死亡的人口控制，允许将非致命企图与自杀死亡进行比较。在……里面 此外，我们将与西奈山医学院的同事合作，他们目前 开发EHR和多基因风险模型来研究物质使用障碍、焦虑和重度抑郁 在西奈山生物群生物库的37,510名参与者中出现了疾病。他们将把这项工作扩大到包括 自杀性为我们的模型开发和测试提供了一个额外的自杀企图资源。我们会 此外，使用我们的资源，研究与自杀死亡和未遂相关的多基因风险分数 以及与范德比尔特大学合作访问他们的生物库和自杀未遂 在英国的生物库..通过对收集到的犹他州新自杀病例进行基因分型，将有可能进行独立验证 在整个项目中，通过其他比较来尝试在大型数据集中尝试案例，这些数据集可以通过 心理医生联盟。

ABSTRACT Suicide is a leading cause of death that continues to increase, with over 47,000 preventable suicide deaths per year in the U.S. Although we have made great strides in using electronic health records (EHR) and other factors to predict suicidal ideation and behavior, our ability to reliably predict suicide death is close to zero. From a healthcare standpoint, predicting suicide deaths is tricky. We know that the incidence of suicide behaviors is far more common (~4%-5% per year) compared to suicide death (~0.01%-0.02% per year). Essentially, only a small fraction of those who engage in suicidal behaviors will go on to die by suicide. Knowledge of who these highest risk individuals are is critically important in directing prevention efforts and development of future targeted interventions. In addition, well over half of suicide deaths occur with no prior attempts, even accounting for lack of documentation of attempts in diagnostic codes. These “out of the blue” cases suggest one or more high-risk groups even more elusive to accurate prediction and prevention. Including genetic data of suicide deaths may offer substantial predictive improvement; genetic factors account for close to 50% of the risk of suicide death. Using the extensive genetic data, statewide longitudinal EHR resources, demographic, and familial data available to the Utah Suicide Genetic Risk Study (USGRS), we are uniquely poised to address this critical knowledge gap. Our primary focus will be to use machine learning methods develop models that predict suicide deaths. In addition, our large suicide death research resource will also allow us to model differences of suicide deaths with vs. without prior attempts. Of the ~9,000 Utah suicide deaths with demographics and environmental data, familial data, and 2 decades of longitudinal EHR data, the USGRS also currently has DNA from >6,000, which will increase to ~10,000 during the award period. Genome- wide molecular data is in hand for over 5,000 of these Utah suicides, allowing for tests of association of suicide subtypes identified using EHR data with “genetic phenotypes” represented by polygenic risk scores. The USGRS also has demographics, familial data, and longitudinal EHR data from 5 age/sex- matched Utah population controls for each suicide death, allowing for comparisons of non-lethal attempts to suicide deaths. In addition, we will collaborate with colleagues at the Mount Sinai School of Medicine, who are currently developing EHR and polygenic risk models to study substance use disorder, anxiety, and major depressive disorder in 37,510 participants in the Mount Sinai BioMe Biorepository. They will expand this work to include suicidality to provide an additional resource of suicide attempt for our model development and testing. We will additionally study polygenic risk scores associated with suicide death vs. attempt using our resources, Mount Sinai BioMe, and a collaboration with Vanderbilt University for access to their Biobank and to suicide attempts in the UK Biobank.. Independent validation will be possible through genotyping of new Utah suicides collected throughout the project, with additional comparisons to attempt cases in large datasets available through the PsychEMERGE consortium.