Sensitivity analysis for informative censoring in parametric survival models: an evaluation of the method

In a paper by Siannis, Copas and Lu in Biostatistics, the authors proposed and studied a sensitivity analysis for informative censoring in parametric survival analysis. More specifically, they introduced a parametric model that allows for dependence between the failure and censoring processes in terms of a parameter delta which can be thought of as measuring the size of the dependence between the two processes, and a bias function that measures the pattern of this dependence. Based on this model, for small values of delta, they also derived simplified closed form expressions (approximations) for the sensitivity analysis of the associated parameters of the model. Since then, some extensions of this approach have also appeared in the literature. In this paper, some theoretical issues concerning the above approach are discussed. Then the results of an extensive simulation study are reported, which indicate some shortcomings of the proposed sensitivity analysis, particularly in the presence of nuisance parameters.