Modeling right-censored medical cost data in regression and the effects of covariates

This paper focuses on the problem of modeling medical costs with covariates when the cost data are subject to right-censoring. The prevailing methods are divided into three categories, (a) the inverse probability weighted (IPW) regressions; (b) the generalized survival-adjusted estimators; and (c) the joint-modeling methods. Comparisons are made both in and between categories to demonstrate their different mechanisms to handle the informative censoring, to take into account the covariates and the way they interpret the covariates effects. Based on the above discussion, we believe that the linear or generalized linear regressions using the IPW scheme are very popular due to its convenience to fit and interpret, which could be a good choice in practice with additional conditional means to address the role of survival to some extent. The recently proposed generalized survival-adjusted estimator is very intuitive as the derivative of the estimation function naturally decomposes the effects of covariates into the intensity part and the survival part, therefore especially useful when the covariates have substantial effect on survival. The joint-modelling methods have the advantage in providing the access to the correlation between medical cost and survival, although they suffer from theoretical and computational complexity. The effect of covariates on cost through survival in this kind of joint-modelling methods could be a desirable topic for further research.