Statistical Analysis of Longitudinal Medical Cost Data

描述（由申请人提供）：本申请的目的是开发新的统计模型和方法来分析纵向医疗费用数据，例如每月或其他长期收集的医疗费用数据。与总成本数据的分析浪费大量信息相比，纵向成本数据的可用性可以使卫生经济分析更有效，解释更全面，更有见地，更有用。我们首先提出了一个纵向四部分模型，用四个联合方程分别建模：(1)就医概率，(2)住院概率，以及(3)门诊费用和(4)住院费用的实际金额。我们的模型同时考虑了每个患者的跨时间相关性和四个方程的交叉相关性，通过(1)和(2)中二元结果的联合广义线性混合模型（GLMM）和(3)和(4)中对数转换成本的线性混合模型；为了避免对数转换成本中的再转换问题，我们在第二部分中提出了一个带有Gamma GLMM的两部分随机效应模型，用于计算正成本的数量。此外，我们还考虑了一个更灵活的广义伽玛分布的数量的正成本。

DESCRIPTION (Provided by the Applicant): The objective of this application is to develop new statistical models and methods to analyze longitudinal medical cost data, e.g., monthly or other medical cost data collected over time. Compared to the analysis of total cost data which discards a great amount of information, the availability of longitudinal cost data can make health economic analyses more efficient and interpretation more comprehensive, insightful, and useful. We first propose a longitudinal four-part model, with four joint equations modeling respectively: (1) the probability of seeking medical treatment, (2) the probability of being hospitalized, and the actual amount of (3) outpatient and (4) inpatient costs. Our model simultaneously takes account of the inter-temporal correlation of each patient and the cross-equation correlation of the four equations, by joint generalized linear mixed models (GLMM) for binary outcome in (1) and (2), and linear mixed models for log transformed cost in (3) and (4); To circumvent the re-transformation issue in log transformed cost, we next propose a two part random effects model with a Gamma GLMM in part II for the amount of positive cost. Furthermore, we also consider a more flexible generalized Gamma distribution for the amount of positive cost. In Aim 3 we introduce a two-part random effects model for longitudinal medical costs, with proportional hazards model in the second part. An appealing feature of the proportional hazards model is that the baseline hazard is unspecified (nonparametric), accommodating the fact that positive medical cost data are often highly skewed and heteroscedastic, which can not be described by any simple parametric distribution. Our model can be used to ascertain the risk factors of medical costs and identify the most cost-effective treatment. To illustrate this we apply our proposed new model to monthly medical costs of 1,397 chronic heart failure patients from the clinical data repository (CDR) at the University of Virginia. We also compare the results from this model with those from the existing models in terms of consistency and accuracy for generating financial forecasts. Finally, we will develop ready-to-use software to facilitate the application of our methods in practical data analysis.