|
Abstract:
|
In pharmacoepidemiology, estimation of incidence and prevalence of drug use is of key interest. The basic epidemiologic methods for incidence and prevalence do however impose a chronic treatment assumption which is unrealistic for most medications. Hence we suggest to extend the waiting time distribution model (time from a fixed date to first subsequent treatment event for each patient) to three compartments, corresponding to no previous, paused, or current treatment, respectively. As the two first compartments can not be distinguished in the waiting time distribution alone, we incorporate information from the interarrival distribution for time between treatment events. This transformation is non-trivial as it introduces informative censoring into the waiting time distribution, and we therefore consider how censoring can be modeled to allow valid estimation of the interest parameters.
|
