Separate vs. simultaneous analysis of co-primary endpoints in Alzheimer’s disease clinical trials
Jacob Brogren
AstraZeneca R&D
Objectives: In clinical trials of drugs intended for treatment of Alzheimer's Disease (AD), cognitive and functional measures may be used as co-primary endpoints. These variables are commonly evaluated in separate statistical tests. In theory, two endpoints measured simultaneously will have a joint distribution with respect to random effects. The power of detecting a treatment effect one or both of the endpoints may differ depending on if simultaneous or separate analyses are performed. The objective of this investigation was to calculate the power of finding a treatment effect on disease progression in a clinical AD trial comparing separate and simultaneous analysis of endpoints.
Methods: The ADNI database [1] was used for modeling. A dataset including ADAS-Cog 70 point total score and Functional Assessment Questionnaire (FAQ) was prepared. A mixed effects model was fitted to the data. A treatment effect influencing the disease progression was simulated using the final model. Further power calculations were performed using the MCMP method [2]. A dataset with a large number of individuals was simulated from the model to be used in the MCMP calculation. Target values of 80% and 5% respectively were assumed for power and type-I error rate.
Results: The linear mixed effects model for ADAS-Cog and FAQ total scores fitted the data reasonably well. The required sample size for 80% power of detecting a treatment effect differed depending on whether simultaneous or separate analyses were performed.
Conclusions: Recognizing the random effects covariance of co-primary endpoints in AD studies may influence power of detecting a treatment effect in clinical studies.
References:
[1] Alzheimer's Disease Neuroimaging Initiative database http://adni.loni.ucla.edu/
[2] Camille Vong, Martin Bergstrand, Mats O. Karlsson Rapid sample size calculations for a defined likelihood ratio test-based power in mixed effects models PAGE 19 (2010) Abstr 1863 [www.page-meeting.org/?abstract=1863]
