Abstract:
|
Recurrent event data is common in clinical trials like the tumor recurrence, recurrent episodes of acute lower respiratory tract infections, and recurrent malaria episodes etc. Often when analyzing recurrent event data, missing at random (MAR) is assumed. Under the situation like informative censoring occurs (e.g. subjects drop out due to side effects in treatment arm), this assumption may not lead to conservative analysis. For confirmatory analysis of clinical trials, sensitivity analyses are required to investigate alternative estimate that depart from MAR assumption. It has been challenge to perform missing not at random sensitivity analysis when Andersen-Gill Model is used as a regression analysis of the intensity of the recurrent event. A method, using multiple imputation framework and Bayesian model to generate the event time for the group who are censored prior to the event of interest and who are missing not at random (MNAR), is illustrated and evaluated.
|