| dc.description.abstract | Cost-effectiveness analysis is widely conducted in the economic evaluation of new treatments, due to skyrocketing health care costs and limited resource available. Censored costs data poses a unique problem for cost estimation due to “induced informative censoring” problem. Thus, many standard approaches for survival analysis are not valid for the analysis of cost data. We first derive the confidence interval for the incremental cost-effectiveness ratio for a special case, when terminating events are different for survival time and costs. Then we study how to intuitively explain some existing estimators for costs, based on the generalized redistribute-to-the-right algorithm. Motivated by that idea, we also propose two improved survival estimators of costs, based on generalized redistribute-to-the-right algorithm and kernel method.
We first consider one special situation in conducting cost-effectiveness analysis, when the terminating events for survival time and costs are different. Traditional methods for statistical inference cannot deal with such data. We propose a new method for deriving the confidence interval for the incremental cost-effectiveness ratio under this situation, based on the counting process theory and the general theory for missing data process. The simulation studies and real data example show that our method performs very well for some practical settings.
In addition, we provide intuitive explanation to a mean cost estimator and a survival estimator for costs, based on generalized redistribute-to-the-right algorithm. Since those estimators are derived based on the inverse probability weighting principle and semiparametric efficiency theory, it is not always easy to understand how these methods work. Therefore, our work engenders a better understanding of those theoretically derived cost estimators.
Motivated by the idea of generalized redistribute-to-the-right algorithm, we propose an estimator for the survival function of costs. The proposed estimator is naturally monotone, more efficient than some existing survival estimators, and has a quite small bias in many realistic settings. We further propose a kernel-based survival estimator for costs. The latter estimator, which is asymptotically unbiased, overcomes the deficiency of the former estimator, while preserving the nice properties. Our proposed estimators outperform existing estimators under various scenarios in simulation and real data example. | en |
