Instrumental Variable Methods for Censored Cost Data and an Application in Prosta

描述（由申请人提供）：在许多研究中，分析任务是估计特定条件或慢性疾病治疗组在规定时间内的总成本。其中包括疾病成本研究，如疾病的增量成本或总生命周期成本，成本效益的组成部分，以及卫生保健中的其他形式的经济评估。然而，回答这些问题需要解决一系列统计和计量经济学问题，其中包括对个别患者观察的可变时间段、疾病或治疗过程中的波动成本、患者死亡时成本的上升以及在感兴趣的终点（例如治愈、死亡）之前对数据的行政审查。这些分析障碍由于成本数据的偏差和治疗决策的内生性而变得更加复杂，这些决策是由基于预期的特殊效益、成本和治疗风险的自我选择造成的。我们的建议将发展创新的方法，以综合的方式解决每一个问题。拟议的工作将推动保健研究中的成本分析和经济评价四个广泛领域。首先，它将提供并验证一种一致的估算成本的方法，这种方法的特点是随着时间的推移而变化的支出率，正确的审查，以及通常在死亡前观察到的支出的增加。其次，它将连接两个关于方法的大型文献，一个是关于扭曲成本数据的建模（右审查），另一个是关于在存在治疗异质性的情况下使用IV。第三，它可以突出未被观察到的患者偏好和疾病严重程度可能在前列腺癌治疗效果异质性中所起的作用。最后，这项工作很有可能产生关于前列腺癌比较成本估算的新信息，这可以作为成本效益和其他政策分析的重要投入。

DESCRIPTION (provided by applicant): There are many studies where the analytical task is to estimate the total costs over a defined period of time for a treatment group with a particular condition or chronic disease. These include cost of illness studies such as incremental or total lifetime costs for a disease, the components of cost effectiveness, and other forms of economic evaluation in health care. However, answering such questions requires addressing a range of statistical and econometric issues that include variable time periods of observation for individual patients, fluctuating costs over the course of the disease or treatment, the escalation of costs when the patient is dying and administrative censoring of data prior to the endpoint of interest (e.g. cure, death). These analytical obstacles are compounded by skewed cost data and the endogeneity of treatment decisions caused by self-selection into treatment based on expected idiosyncratic benefits, costs and risks of treatments. Our proposal will develop innovative methods to address each of these problems in a consolidated manner. The proposed work will advance four broad areas in the fields of cost analysis and economic evaluation in health studies. First, it will provide and validate a consistent method of estimating costs in a setting characterized by varying rates of spending over time, right censoring, and the escalation of expenditures commonly observed just before death. Second, it will bridge two large literatures on methods, one on modeling skewed cost data (with right censoring) and the other on the use of IV in the presence of treatment heterogeneity. Third, it can highlight the role that unobserved patient preferences and disease severity may have in generating treatment effect heterogeneity in the context of prostate cancer. Lastly, this work has a high likelihood of generating new information about the comparative cost estimates in prostate cancer which can serve as important inputs into cost-effectiveness and other policy analyses.