Hierarchical Bayesian cognitive processing models to analyze clinical trial data

Abstract Identifying disease-modifying treatment effects in earlier stages of Alzheimer's disease (AD)—when changes are subtle—will require improved trial design and more sensitive analytical methods. We applied hierarchical Bayesian analysis with cognitive processing (HBCP) models to the Alzheimer's Disease Assessment Scale—Cognitive subscale (ADAS-Cog) and MCI (mild cognitive impairment) Screen word list memory task data from 14 Alzheimer's disease AD patients of the Myriad Pharmaceuticals' phase III clinical trial of Flurizan (a γ-secretase modulator) versus placebo. The original analysis of 1649 patients found no treatment group differences. HBCP analysis and the original ADAS-Cog analysis were performed on the small sample. HBCP analysis detected impaired memory storage during delayed recall, whereas the original ADAS-Cog analytical method did not. The HBCP model identified a harmful treatment effect in a small sample, which has been independently confirmed from the results of other γ-secretase inhibitor. The original analytical method applied to the ADAS-Cog data did not detect this harmful treatment effect on either the full or the small sample. These findings suggest that HBCP models can detect treatment effects more sensitively than currently used analytical methods required by the Food and Drug Administration, and they do so using small patient samples.