Statistical modeling of longitudinal medical cost trajectory: renal cell cancer care cost analyses

Estimating the current cost of cancer care is important to health policy makers. An indispensable step in cost projection is to estimate cost trajectories from an incident cohort of cancer patients using longitudinal medical cost data, accounting for terminal events such as death, and right censoring due to loss of follow-up. Since the cost of cancer care and survival are correlated, a scientifically meaningful quantity for inference in this context is the mean cost trajectory conditional on survival. We propose a two-stage semiparametric approach to estimate the longitudinal cost trajectories from a joint model of longitudinal medical costs and survival. The longitudinal cost trajectories corresponding to various survival times form a bivariate surface in a triangular area. The cost trajectories are estimated using the tensor products of discretized measurement time and survival, as well as effective ridge penalties for data in 2D arrays. The proposed approach balances the practical considerations of model flexibility, statistical efficiency, and computational tractability. We used the proposed method to estimate the cost trajectories of renal cell cancer patients using the Surveillance, Epidemiology, and End Results-Medicare linked database.