Structural MRI markers may serve as surrogate endpoints in clinical trials on disease modification in Alzheimer's disease (AD). Here, we used a longitudinal MRI data set of total brain and cortical grey matter volumes from 66 patients with AD recruited across seven centres of the German Dementia Competence Network. We compared effect size estimates for the detection of a 25% reduction of atrophy progression between a priori segmentation of brain tissue, implementing an anatomical model of brain tissue distribution, and a posteriori segmentation that was not informed by an anatomical model. Additionally, we compared effect size estimates between fixed effects analysis and a mixed effects model, implementing a random effects term to account for variable spacing of observation times. A priori segmentation reduced the required sample size by 50%. Introducing a random effects term for time led to an additional 50% reduction of required samples sizes compared to fixed effects analysis. In summary, using a priori segmentation with mixed effects analysis reduced the sample size to detect clinically relevant treatment effects more than fourfold. The implementation of mixed effects models will enhance the power to detect treatment effects also with other classes of biological endpoints including molecular biomarkers of disease.
Structural MRI markers may serve as surrogate endpoints in clinical trials on disease modification in Alzheimer's disease (AD). Here, we used a longitudinal MRI data set of total brain and cortical grey matter volumes from 66 patients with AD recruited across seven centres of the German Dementia Competence Network. We compared effect size estimates for the detection of a 25% reduction of atrophy progression between a priori segmentation of brain tissue, implementing an anatomical model of brain tissue distribution, and a posteriori segmentation that was not informed by an anatomical model. Additionally, we compared effect size estimates between fixed effects analysis and a mixed effects model, implementing a random effects term to account for variable spacing of observation times. A priori segmentation reduced the required sample size by 50%. Introducing a random effects term for time led to an additional 50% reduction of required samples sizes compared to fixed effects analysis. In summary, using a priori segmentation with mixed effects analysis reduced the sample size to detect clinically relevant treatment effects more than fourfold. The implementation of mixed effects models will enhance the power to detect treatment effects also with other classes of biological endpoints including molecular biomarkers of disease.