Developing a Robust Measure of Hospital End-of-Life Intensity

描述（由申请人提供）：一些研究人员认为，老年医疗保险受益人在生命结束时接受的高强度医疗护理是护理质量差和效率低下的潜在指标。然而，通过死者随访（病例系列）方法测量治疗强度的传统方法可能存在缺陷。具体而言，许多死者并不知道是“死亡”，对他们进行强化治疗可能是适当的；相反，目标是识别预后不良的患者，对他们进行重症监护可能具有低边际价值。我们开发了一种新的生命结束强度测量方法，该方法关注于高死亡概率（HPD）患者的治疗模式，HPD定义为在预测死亡概率的第95百分位入院。我们的HPD测量在理论上是对“垂死”患者决策的一种较少偏差的估计；它比较了不同医院的相似人群，这可能是效率低下的标志。我们计划使用我们的HPD强度测量，通过研究个体医院行为对这些模式的贡献，进一步阐明在生命末期决策中种族和条件特异性变化的原因和后果。黑人使用维持生命治疗（LST）的比例更高，并且光顾ICU使用率更高的医院，总体而言，癌症患者比患有严重生命限制疾病的非癌症患者接受临终ICU护理的可能性要小得多。其次，我们试图测试我们的HPD测量在不同种族和条件下的表现特征，以确定是否应该为患者亚组单独计算和报告该测量。第三，我们试图探索HPD方法在宾州缺乏临床和风险预测数据的绝大多数行政数据中的普遍性。我们的目标是：1)计算医院临终治疗强度的种族特异性度量，并探索医院种族特异性强度与入院后生存之间的关系；2)计算医院临终治疗强度的具体条件测度，探讨医院临终治疗强度与入院后生存的关系；3)开发一种行政数据衍生的HPD测量方法，并将其与我们的“黄金标准”临床数据增强HPD测量方法进行比较。所有的分析都将使用宾夕法尼亚州卫生保健成本控制委员会的数据与州生命统计数据相关联。我们提出的用于开发医院特定强度测量的统计程序将依赖于最先进的贝叶斯技术，生存分析将扩展到卫生服务研究，即最初在流行病学中开发的边缘结构模型，以解决时变混杂因素。我们试图改进的新措施有可能填补当前政策努力的空白，以公开介绍医院的表现，并帮助我们更好地了解使用重症监护和LST的决定如何因种族和条件而异。

DESCRIPTION (provided by applicant): Several researchers have argued that the high intensity of medical care that elderly Medicare beneficiaries receive at the end of life is a potential indicator of poor quality of care and of inefficiency. However, the traditional method for measuring intensity of treatment through the decedent follow-back (case-series) approach may be flawed. Specifically, many decedents were not known to be "dying" and that intensive treatment for them may be appropriate; instead, the goal is to identify patients with poor prognosis for whom intensive care may have low marginal value. We have developed a new measure of end-of-life intensity that focuses on treatment patterns among patients with a high probability of dying (HPD), defined as admissions in the 95th percentile of predicted probability of death. Our HPD measure is a theoretically less biased estimate of decision making for "dying" patients; it compares similar populations across hospitals, and it may be a marker of inefficiency. We plan to use our HPD intensity measure to further elucidate the causes and consequences of race- and condition-specific variations in decision making near the end of life by studying the contribution of individual hospital behavior to these patterns. Blacks have higher rates of life-sustaining treatment (LST) use and patronize hospitals with greater ICU use, and in the aggregate, cancer patients are much less likely than non-cancer patients with serious life limiting illness to receive end-of-life ICU care. Second, we seek to test the performance characteristics of our HPD measure across racial groups and conditions to ascertain whether the measure should be calculated and reported separately for patient subgroups. Third, we seek to explore the generalizability of the HPD approach to the vast majority of administrative data lacking the clinical and risk prediction data available in Pennsylvania. Our aims are:1) To calculate race-specific measures of hospitals' end-of-life treatment intensity and explore the relationship between a hospital's race-specific intensity and post-admission survival; 2) To calculate condition-specific measures of hospitals' end-of-life treatment intensity and explore the relationship between a hospital's condition-specific intensity and post-admission survival; and 3) To develop an administrative data-derived HPD measure and compare it to our "gold standard" clinical data-augmented HPD measure. All analyses will use Pennsylvania Health Care Cost Containment Council data linked to state vital statistics data. Our proposed statistical procedures for developing hospital-specific intensity measures will rely on state-of-the-art Bayesian techniques, and survival analyses will extend to health services research the marginal structural models originally developed in epidemiology to address time-varying confounders. The new measure we seek to refine has the potential to fill a niche in current policy efforts to publicly profile hospitals' performance and to help us better understand how decisions to use intensive care and LST vary by race and condition.