Robust Suicide/Reinjury Risk Models to Assess Healthcare Systems

描述（由申请人提供）：最近的研究表明，自杀状态的最有效治疗发生在医疗保健提供系统内的多层次预防措施的背景下。这些发现强调了创建可以解决自杀和再伤害风险的服务的重要性。虽然提供关键服务与降低自杀率有关，但重要的是要了解服务如何有助于防止自我伤害者随后再次受伤。例如，2012年国家预防自杀行动联盟研究工作队利益相关者调查将自杀和自杀再伤害的研究列为其三个最高优先研究项目之一。目前，高达25%的自杀完成者和高达70%的非致命性再伤害患者在随后的再伤害12个月内在医疗保健环境中接受已知的非致命性自残行为治疗。尽管多状态、基于人群的、患者级和系统级数据可用，但这些数据库中的信息没有被有效地提取。我们希望证明，先进的统计建模将提供机制，提取这些信息，使决策者能够区分医疗保健提供系统，更有效 防止那些经历更高再伤害率的人故意再伤害。 这项第一阶段的研究探讨了应用新的统计方法来分析模型的照顾自我伤害的个人使用从AHRQ的广泛的医疗保健成本和利用项目[HCUP]获得的患者层面的医疗记录的可行性。描述连续向医疗保健系统提交的数据，以治疗来自五个州的州卫生服务部门或州医院协会的自伤，这些数据将首先用于开发医疗保健系统风险模型，然后用于测试损伤后医疗与致命或非致命故意再损伤的可能性之间的关联。将使用最佳近似模型（BAM）方法开发模型，该方法选择并验证稳健的模型搜索，同时处理常见的建模问题，例如存在可能的模型错误指定、缺失值、过度拟合、多重共线性、罕见事件结局偏倚和多重比较的夸大误差。将进行模拟研究，以表征BAM策略的优势，以开发一个强大的自杀/再伤害风险模型，而不是广泛使用的多元统计方法，如逐步回归。可行性研究结果将为更先进的II期医疗风险模型开发、评估和传播提供初步研究，进而为III期产品商业化奠定必要基础。

DESCRIPTION (provided by applicant): Recent research suggests that the most effective treatment of suicidal states occurs in the context of multi- level prevention initiatives within healthcare delivery systems. Such findings underscore the importance to create services that can address suicide and reinjury risks. While offering key services have been associated with reduced suicide rates, it is important to understand how services contribute to prevent self-harming persons from subsequent reinjury. For instance, the 2012 National Action Alliance for Suicide Prevention Research Task Force Stakeholder Survey ranked studies of suicide and suicide reinjury among its three highest priorities for research. Presently, up to 25% of all suicide completers and up to 70% of all nonfatal reinjuring patients are treated in a healthcare environment for a known nonfatal act of self-harm within 12 months of subsequent reinjury. Despite the availability of multi-state, population-based, patient-level, and system-level data, th information in such databases is not being effectively extracted. We wish to demonstrate that advanced statistical modeling will provide mechanisms for the extraction of such information that will enable policy makers to differentiate healthcare delivery systems that are more effective in preventing intentional reinjury from those that experience higher reinjury rates. This Phase I study investigates the feasibility of applying new statistical methods to analyze models of care for self-harming individuals using patient-level medical records obtained from AHRQ's extensive Healthcare Cost and Utilization Project [HCUP]. Data characterizing consecutive presentations to healthcare systems for treatment of self-harm from the state health services divisions or state hospital associations in five states will be used first to develop healthcare system risk models and then to test associations between post-injury medical treatment and the likelihood of fatal or nonfatal intentional reinjury within 12 months of an index, nonfatal intentional self-injury. Model will be developed using the Best Approximating Model (BAM) approach that selects and validates robust model searches while simultaneously handling common modeling problems such as the presence of possible model misspecification, missing values, over fitting, multicollinearity, rare event outcomes bias, and inflated error from multiple comparisons. Simulation studies will be performed to characterize the advantages of the BAM strategy for developing a robust suicide/reinjury risk model over widely-used multivariate statistical methods such as stepwise regression. Feasibility study results will provide the preliminary research for more advanced Phase II healthcare risk model development, evaluation, and dissemination that, in turn, establishes the essential foundation for Phase III product commercialization.