Clustered semi-competing risks analysis in quality of end-of-life care studies

描述：美国医学研究所最近的一份报告强调了在不牺牲医疗质量的情况下控制美国医疗成本的迫切需要。作为医疗汽车费用的最大支付者，医疗保险和医疗补助服务中心(CMS)在全国范围内开展全面努力，监测医疗质量。然而，这些努力侧重于治愈率高、死亡率低的急性疾病。癌症和阿尔茨海默氏症等一系列日益普遍的“高级健康状况”，治愈率低，短期死亡率高，疾病管理的重点是临终关怀。然而，这样的护理是昂贵的。2010年，全国癌症护理费用估计为1250亿美元。尽管有这些巨大的成本，但没有全面的国家努力来监测EOL护理的质量。这些努力的一个关键障碍是缺乏适当的统计方法。为了估计特定医院的再住院率，CMS目前使用Logistic-正态广义线性混合模型(GLMM)。然而，这个模型忽略了死亡是一个截断的事件。因此，天真地应用当前的CMS方法来评估高级健康状况的EOL质量是不合适的，可能会导致偏见，并可能对医院因护理质量优秀/差而受到奖励/惩罚的方式产生重大影响。在统计学文献中，他研究了一种非终末期事件(例如，再入院)，它受到终末期事件(例如，死亡)的影响，我称之为“半竞争风险”问题。目前的国家护理质量评估工作忽视了半竞争性风险问题。一个主要的因素是，统计文献中没有考虑到聚集的半竞争性风险数据。因此，必须开发和评估半竞争性风险数据的新统计方法。我们将为半竞争性风险数据开发一个全面、统一的贝叶斯分析框架。拟议的框架将允许研究人员利用贝叶斯范式提供的众多好处。一个关键的贡献将是为重复测量半竞争数据开发新的贝叶斯分层模型，其中个人聚集在医院内。随着时间的推移，将开发新的多变量医院一级措施，共同适应非临终事件和临终事件，以及评估、推断、排名和确定优秀/较差医院的方法。最后，使用2000-2010年所有联邦医疗保险参保人的数据和SEER-Medicare的肿瘤数据，我们将应用我们的方法来提高胰腺癌、肺癌、结肠癌和脑癌的EOL护理质量。拟议的工作将立即大幅改进和扩大用于EOL护理质量评估的统计工具集，并提供美国癌症护理的关键流行病学结果。这些方法将广泛适用于除癌症以外的所有高级健康状况，其中许多直接影响到日益老龄化的美国人口的很大一部分人。

DESCRIPTION: A recent Institute of Medicine report highlighted the pressing need to control health care costs in the US without sacrificing quality of care. As the largest payer of health car costs, the Centers for Medicare and Medicaid Services (CMS) conducts comprehensive national efforts to monitor quality of care. However, these efforts focus on acute conditions for which cure rates are high and mortality low. For a broad range of increasingly prevalent 'advanced health conditions', such cancer and Alzheimer's disease, cure rates are low, short-term mortality is high and the focus of disease management is end-of-life (EOL) palliative care. Such care is expensive, however. In 2010 national cost of cancer care was estimated to be $125 billion. Despite these huge costs, there are no comprehensive national efforts to monitor quality of EOL care. A key barrier to these efforts is the lack of appropriate statistical methodology. To estimat hospital-specific readmission rates, CMS currently uses a logistic-Normal generalized linear mixed model (GLMM). However, this model ignores death as a truncating event. As such, naïve application of the current CMS approach for quality of EOL assessments for advanced health conditions is inappropriate, would likely lead to bias and could have a major impact on how hospitals are rewarded/penalized for excellent/poor quality of care. In the statistics literature, he study of a non-terminal event (e.g. readmission) that is subject to a terminal event (e.g. death) i known as the 'semi-competing risks' problem. Current national quality of care assessment efforts ignore the semi-competing risks problem. A major contributing factor is that clustered semi-competing risks data has not been considered in the statistical literature. Novel statistical methods for semi-competing risks data must, therefore, be developed and evaluated. We will develop a comprehensive, unified Bayesian analysis framework for semi-competing risks data. The proposed framework will permit researchers to take advantage of the numerous benefits afforded within the Bayesian paradigm. A crucial contribution will be the development of novel Bayesian hierarchical models for repeated measures semi- competing data, where individuals are clustered within hospitals. Novel multivariate hospital-level measures that jointly accommodate non-terminal and terminal events over time will be developed, as will methods for estimation, inference, ranking and the identification of excellent/poor hospitals. Finally, using data on all Medicare enrollees from 2000-2010 and tumor data from SEER-Medicare, we will apply our methods to quality of EOL care for cancers of the pancreas, lung, colon and brain. The proposed work will immediately and substantially improve and expand the set of statistical tools use for EOL care quality assessments, as well as provide key epidemiological results on cancer care in the US. The methods will be broadly applicable to all advanced health conditions, beyond cancer, many of which directly affect large segments of an increasingly aging US population.