Response type modelling and clinical trial simulation.
J. Grevel (1), R. Austin (1), M. Lavielle (2)
(1) BAST Inc Ltd. Nottingham, UK; (2) Inria, Saclay, France
Objectives: Drug development in depression is a particular challenge since high apparent variability in response is obscuring a clear dose-response against placebo. New drugs are often interacting at multiple targets, and it seems appropriate to identify patients that share a common response type. We present a statistical model for response type analysis.
Methods: A between subject model mixture (BSMM) was implemented in Monolix 4.1 [1] that estimated the probabilities of four different response types. The types were: no response, short, long, and continued response. Considering dose as a categorical covariate, the probability distribution of the response type was estimated as a composite of the probability distribution in each treatment arm. The population parameters of the model were estimated using the SAEM algorithm for BSMM. The prediction distribution of the response in each treatment arm was estimated by simulation using the population parameters.
Results: A clinical anti-depression trial was simulated using the first prototype of the clinical trial simulator developed by Inria for DDMoRe [2] with 200 patients being randomly allocated to four equal treatment arms: placebo, 50, 100, and 150 mg. The treatments were administered for 3 weeks and response was simulated up to 8 weeks. Without a response type analysis all three active treatments seemed to be equally effective in comparison to placebo. The distinct probability distributions for the response type under each treatment demonstrated a clear dose-response. The prediction distribution of the response in each treatment group demonstrated the superiority of the 100 mg dose over the two others. Minimal group sizes could be determined according to outcome expectations.
Conclusions: A between subject model mixture estimated the probabilities of different response types in the treatment arms. The response type analysis paired with trial simulation elucidated a dose-response where a standard analysis failed.
References:
[1] Monolix 4.1 User Guide, http://www.lixoft.com/wp-content/uploads/2011/10/UserGuide.pdf.
[2] DDMoRe, http://www.ddmore.eu/.
