Joint Modeling of Medical Expenditure and Survival in Complex Sample Surveys
Laura Sands, Purdue University School of Nursing 
*Huiping Xu, Indiana University School of Medicine 
Danni Yu, Purdue University Department of Statistic 

Keywords: Complex survey, medical cost data, random effects, sample weights, survival analysis

Medical expenditure data are typically highly skewed to the right in that a small percentage of patients have very serious medical conditions and incur extremely high costs compared to others. Another characteristic of these data is that a large portion of zero values usually exist because many people do not incur any costs during a certain period of time. These characteristics must be accounted for in order to accurately estimate costs. In addition, it is common that expenditure data are accompanied by the incomplete follow-up due to informative censoring such as death. It is important to analyze the expenditure and survival simultaneously to account for their potential dependence, which might occur since people with a higher death rate tend to be frailer and hence incur a larger medical cost. Liu (2009) developed a joint model of longitudinal monthly medical expenditure and survival using shared random effects. We propose to extend this model to complex survey data by introducing sample weights into the model parameter estimation. The complex survey design is accounted for through the use of balanced repeated replication approach for the standard error estimation. Simulation studies show that ignoring the survey design results in biased parameter estimates. Our approach is applied to the longitudinal monthly Medicare expenditure of 2,313 respondents from the community interview of the 2004 National Long-Term Care Survey (NLTCS) who had difficulty in performing activities of daily life such as eating, dressing, getting around, and bathing, and toileting.