|
Abstract:
|
Recurrent event data occur in clinical trials when multiple events of the same type are observed for each patient, for example, heart-failure hospitalizations in cardio-vascular studies. Interest lies in the number of events that occur over a fixed period of time and the negative binomial model is a preferred approached for analysis. A challenge in analyzing such data arises when a large proportion of patients discontinues before the end of the study. Using the negative binomial distribution implicitly assumes a missing-at-random mechanism. Thus, there is need for sensitivity analysis to assess robustness across different missingness mechanisms. Our approach is to use a Gamma-Poisson process to complete the data for those who withdraw, conditional upon the history of events that they had prior to withdrawal. This was studied by Keene et al. (2014) using discrete time intervals while Akacha et al. (2016)’s approach models in continuous time. The results from the multiply-imputed completed datasets are combined using Rubin’s rule. This is then extended to handle death or any competing terminal event. The operating characteristics will be illustrated through a simulation study.
|
