P3-045: Reducing noise in clinical trials: Improving the ADAS-cog scoring

Abstract

The ADAS-Cog test is a primary outcome measure in clinical trials of Alzheimer's disease (AD). Changes in a subject's ADAS-Cog scores result from factors unrelated to treatment effect, including AD progression, practice effects, anxiety, etc., which can mask positive treatment effects. Reducing within-subject variability in ADAS-Cog scores due to such factors can help identify treatment effects. The Memory Performance Index (MPI) is an improved method of scoring the CERAD wordlist memory test that greatly reduces random variability in subject performance over time. The MPI was adapted to scoring the memory sub-test of the ADAS-Cog to see if it could reduce random variability in subject performance. One ADAS-Cog wordlist parallels the CERAD and MCI Screen wordlists except for differences in the order of the 10 words across the three learning trials. We measured the effect of word order in these wordlists by applying Bayesian and correspondence analysis (CA) methods to data from a large sample of normal-to-moderately demented subjects. A second ADAS-Cog wordlist parallels the wordlist used in some AD Centers except for differences in word order across trials. We also measured the effect of word order using CA and Bayesian methods. The effects of the different ADAS-Cog wordlists used were also measured using CA and Bayesian methods. We then applied these results to adapt the MPI score to the ADAS-Cog by adjusting it for word order and wordlist differences between MCI Screen and the ADAS-Cog. The variances in the ADAS-Cog adjusted MPI and delayed free recall scores were then compared using a longitudinal patient sample with at least three repeated measures. On exactly the same set of patient data, the variance of the ADAS-Cog adjusted MPI score was substantially lower than that of the ADAS-Cog delayed free recall score. The MPI score, adjusted for differences between the MCI Screen and ADAS-Cog memory sub-tests, substantially reduces random within-subject variability, making it easier to identify treatment effect for AD and related disorders.