Sean Oosterholt (1), Alienor Berges (2), Oscar Della Pasqua (2), Monica Simeoni (2)
(1) Leiden University, Leiden, the Netherlands; (2) GlaxoSmithKline, London, United Kingdom
Objectives: Although in the past two decades hundreds of molecules were tested against Alzheimer disease (AD) only five for symptomatic treatment were approved, the last one in 2003. Regulatory agencies have recently recognized the importance of simulation modelling tools to better design clinical studies and consequently decrease the high drug development attrition rate for this indication [1]. Population characteristics are an important determinant in AD studies and therefore, an exhaustive assessment of the relationships between parameters and population covariates would be highly valuable. In this work we used the ADAS-cog indirect response model published by Gomeni et al [2] as base model, then the covariate models were selected using a multi-study dataset.
Methods: The analysis was done on placebo/background therapy data from 10 aggregated AD studies for the CAMD initiative [3]. Together the studies offered a wide range of patient characteristics, in terms of demographics (age: 73 ± 8.1, 55% female) and disease features (bMMSE: 20 ±3.8, ADAS-cog: 23.5 ± 9.9). Parameter-covariate relationships were assessed using stepwise covariate model building (SCM) alone or in combination with different validation methods; e.g. a bootstrap SCM. Tested covariates included age, baseline MMSE and gender, while the considered parameter-covariate relationship states were: linear, exponential and power. With the exception of the standard SCM, all methods used a linearization procedure and allowed the parallel evaluation of the parameter-covariate states. To verify the relevance of the selected covariates a bootstrap was performed on the final model. Modelling analyses were executed using NONMEM 7.2 and PsN 3.4.2. Data manipulation, as well as graphical and statistical summaries were done in R 3.0.2.
Results: Standard SCM method selects 7 or 8 out of the 12 possible relationships with and without linearization respectively. The bootstrap SCM narrows down the results to 3 relations selected more than 80% of the time. Gender did not result to be a relevant covariate. Age and baseline Mini Mental State Examination (bMMSE) were selected at least 85% with all parameters showing a dependency on bMMSE.
Conclusions: Using an automated tool for the covariate model building we were able to confirm the inclusion of a subset of covariates on the base structural model published in [2], even when applied to a wider dataset.
References:
[1] http://c-path.org/wp-content/uploads/2013/09/CAMD-AD-Model-press-release.pdf
[2] Gomeni R et al. Alzheimer’s & Dementia 8 (2012) 39–50
[3] Coalition Against Major Diseases; http://www.c-path.org/CAMD.cfm
Reference: PAGE 23 () Abstr 3221 [www.page-meeting.org/?abstract=3221]
Poster: Methodology - Covariate/Variability Models