Exposure-Response Analysis Using Time to Event Data: An Example Using Sleep Onset.

PDF of poster

Background/Aims: Exposure-response analysis of different types of clinical data (e.g., continuous, categorical, count) requires different modeling approaches. The time required to fall asleep is used as the primary endpoint in assessing sleep onset in insomnia patients. The aim of the analysis was to develop a dose-response model to describe latency to sleep (LS).

Methods: Data from 6 studies were combined (773 patients) and analyzed using a survival model in NONMEM.  The Weibull distribution is used to describe the time to an event, such as LS and is described by 2 parameters: f, the median time to event, and g.  When g=1, <1 or >1, the probability of the event remain constant, decreases or increases, respectively, over time.  In addition, an Emax model was used to describe drug effect.  As a posterior predictive check, LS data were simulated (N=100 trials) based on final parameter estimates and compared to observed data.

Results: The shape parameter g was 1.33 indicating an increased probability with time: the longer the patient is awake, the more likely it is that he/she will eventually fall asleep.  Median time to fall asleep (f) at baseline was dependent on study population (patients with difficulty falling asleep [f=34.0 min] vs. difficulty maintaining sleep [f=21.5 min]).  Placebo response averaged 12%, while Emax was 72% and was greater in elderly patients.  Simulated and observed LS by dose were in good agreement. 

Conclusions: The Weibull distribution was successfully implemented to describe time to event data (such as LS) in NONMEM.

Support: Pfizer Global Research and Development