|
Abstract:
|
Medical claims data has become an increasingly important tool in cancer diagnoses, treatments and costs given its longitudinal assessment of clinical outcomes associated with cancer care and resourceful information based on large population. The aim of this paper is to develop an estimating equation approach to integrate massive cost and survival data to access the mean medical cost trajectory in population level, allowing for nonignorable right-censoring and heterogeneity. Such adjustment is motivated by investigating the association between cancer care costs and survival for over ten thousand prostate cancer patients from Surveillance, Epidemiology, and End Results (SEER)-Medicare Linked Database, for which the cost data are not only censored but also heavily-skewed with a large proportion of zero values. Ignoring such data features may distort the accuracy of statistical inference and produce misleading results. To address these issues, we propose a two-part marginal model to depict the censoring mechanism and incorporate flexible mean and variance models to account for heteroskedasticity with multi-dimensional penalized splines and sandwich variance estimator for inference.
|
