ADAS-Cog Placebo Modelling in Alzheimer’s Disease
Pinky Dua, Alienor Berges, Chao Chen, Roberto Gomeni
Clinical Pharmacology Modelling and Simulation (Neurosciences), GlaxoSmithKline
Background: The cognitive portion of the Alzheimer Disease Assessment Scale (ADAS-Cog) is the standard cognitive endpoint for clinical trials of Alzheimer's disease (AD). It is scored by number of errors, ranging from 0 to 70 [1]. An increase in ADAS-Cog score implies worsening cognition. Several recent 6 month clinical trials of investigative medications for AD have failed to detect cognitive decline in the placebo groups by ADAS-Cog, potentially obscuring true treatment effects. The lack of cognitive decline in the placebo groups has renewed interest in a better understanding of the time course of placebo response.
Objective: The aim of this work is to investigate the ADAS-cog placebo model in AD to aid the design of future clinical trials by taking into account the information about which patients are more likely to be placebo responders.
Methods: Data from the placebo arms of 3 recent internal clinical trials including 307 subjects were pooled to investigate the placebo response as a function of time and disease severity given by Mini-Mental Status Exam (MMSE). Nonlinear Mixed Effects Modelling Approach was applied to this data using NONMEM V6. The following models were explored:
Model A: ADASij = ADAS0j + Kj tij - Aj [exp(-koffj tij) - exp(-kj tij)] + eij [2]
Model B: ADASij = ADAS0j × exp(-kj tij) + Kj tij + eij [3]
where ADAS0 = baseline, Kj is the disease progression slope, Aj is the magnitude of placebo contribution, koff is the offset rate of placebo response and kj is the rate of placebo response. For each parameter, between-subject variability was tested and covariate analysis was investigated. Results were assessed in terms of goodness of fit plots, Akaike Information Criterion and posterior predictive check. External validation was carried out using data from internal GSK studies.
Results: Numerical convergence problems were encountered with Model A, which failed to adequately describe the data with FOCE interaction. A theoretical analysis was carried out to explore the reasons of this failure. Structural Identifiability Analysis based upon Taylor's Series approach [4] showed that in presence of no or positive placebo response Model A parameters were not uniquely identifiable. To alleviate this problem, Model B was used. The covariate analysis was carried out using Model B. The addition of MMSE at baseline and on each parameter further improved the data fitting. The placebo response model was finally described as a function of time and of the disease severity at inclusion.
Conclusions: Model B adequately describes the data when a flat placebo response is observed. The advantage of model B is that ADAS-cog response based on the MMSE at baseline can be predicted and thus maximize the possibility of detecting a clinical response by selecting the subjects who can deliver greater signs of drug activity.
References:
[1] R. F. Zec et. al., Alzheimer Dis.Assoc.Disord. 6 (1992) 89-102.
[2] N. H. G. Holford and K. E. Peace, Proc. Natl. Acad. Sci. 89 (1992) 11466-11470.
[3] R. Gomeni and E. Merlo-Pich, Br. J. Clin. Pharmacol. 63 (2007) 595-613.
[4] M. J. Chappell et. al., Mathematical Biosciences. 102 (1990) 41-73.
