Purpose: To evaluate the psychometric properties and patterns of decline on the total score and item scores of the cognitive subscale of the Alzheimer's Disease Assessment Scale (ADAS-Cog) in patients with Alzheimer Disease (AD).
Methods: We analyzed data from 536 AD outpatients randomized to the placebo group in two identical 26-week multicenter drug trials.
Results: Mean deterioration at week 26 on the ADAS-Cog total score for subjects with moderate dementia was 84% greater than that for those with milder severity ( p < 0.001). After adjusting for this effect, age ( p = 0.015) and educational level ( p = 0.01) also predicted cognitive decline. In a model, absolute change for most individual ADAS items was less than 10% of the possible change, and it was generally smaller than one-third of the standard deviation of the measure. Measurement error variability was greatest for word recognition and the "placebo" effect was greatest for word recall. Variability increased with trial duration in a model.
Conclusions: There is a relationship between baseline severity and magnitude of cognitive decline. In 6-month trials, measurement error makes a substantial contribution to the variance in ADAS-Cog change scores. Sensitivity to intervention effects will therefore depend both on the variability and magnitude of change. Such data must be considered when designing future trials to minimize measurement error variability and increase sensitivity for specific populations.