Frank Harrell ([email protected]) wrote: > I wonder whether a simpler approach > would also work: include the censoring/event time, event indicator, and > product of these two in an imputation model. In imputation methods that > allow for nonlinear effects, nonlinear terms could also be added. -Frank
Might work, somebody should try some simulations and see! The method I proposed is closer to the "correct" model (in which censoring means loss of information for later time points), but since the intent is just an ad hoc approximation, your proposal might be adequate and is certainly simpler.
