Dementia epidemiology, health service utilization and treatment costs among American Indian and Alaska Native Elders

项目摘要 电子健康记录(EHR)在健康研究中越来越受欢迎，因为它们 在现实世界的医疗保健中提供对大量纵向健康数据的相对轻松访问 场景。然而，在潜在的疾病风险因素和 死亡率受到多重限制。其中包括选择偏差、错误分类偏差、 知情的在场偏见和无法测量的混乱。另一个潜在的可能性，但在很大程度上还没有研究过。 偏见产生于患者在被研究的系统之外寻求治疗，我们称之为 系统迁移。根据定义，系统迁移会导致主题中间歇性的数据丢失 水平。使这个问题进一步复杂化的是，大多数患者的迁移是未知的 研究人员认为，通常没有迹象表明患者离开了一个医疗系统，并且 到另一个地方寻求治疗。这个问题在印度卫生系统(IHS)中尤其严重，在那里 患者接受外部提供者和IHS的护理是很常见的。当建模时 事件的时间终点，如ADRD诊断的时间，由于以下原因而导致的遗漏 系统迁移的特征可以是(可能未被观察到)左、右或 间隔审查。我提议的培训和研究考虑了潜在的 未观察到的间歇性缺失，当对经过审查的事件间隔时间结果和 提出方法论解决方案，以减少此类情况下的偏见。具体来说，我们建议 可用于1)更准确地估计协变量对时间的影响的统计方法 未知系统迁移模式下的事件结果；2)更准确地估计协变量 错误指定的模型和未知系统迁移对事件发生时间结果的影响 模式；以及3)改进了对重现事件右删失预测精度的评估 生存数据。我们提出的工作将为研究人员提供更好地理解 并将与系统迁移相关的问题的影响降至最低，从而导致 我们研究结果的有效性和可复制性。

Project Summary Electronic health records (EHRs) have grown in popularity for health research because they provide relatively easy access to large amounts of longitudinal health data in real­world healthcare scenarios. However, establishing causal relationships between potential disease risk factors and mortality is subject to multiple limitations. Among these are selection bias, misclassification bias, informed presence bias, and unmeasured confounding. Another potential, yet largely unstudied, bias arises from patients seeking care outside of the system being studied, what we refer to as system migration. By definition, system migration leads to intermittent missing data at the subject level. Further complicating this issue is the fact that most migration of patients is unknown to the researcher as there is generally no indication of a patient leaving one healthcare system and seeking care at another. This problem is particularly true in the Indian Health System (IHS), where it is common for patients to receive care by outside providers as well as the IHS. When modeling a time­to­event endpoint such as time to ADRD diagnosis, the resulting missingness due to system migration can be characterized by (potentially unobserved) left­, right­, or interval­censoring. My proposed training and research consider the implications of potentially unobserved intermittent missingness when modeling censored time­to­event outcomes and proposes methodological solutions to reduce bias in such cases. Specifically, we propose statistical methods that can be used to 1) more accurately estimate covariate effects on time­to­ event outcomes under unknown system migration patterns; 2) more accurately estimate covariate effects on time­to­event outcomes under a mis­specified model and unknown system migration patterns; and 3) improve assessment of prediction accuracy for recurrent event right­censored survival data. Our proposed work will provide the researchers with methods to better understand and minimize the impact of concerns related to system migration, thereby leading to increased validity and replicability of our research findings.