Simulation‐based effect size analysis in the absence of drug effects to inform the design of clinical trials in Alzheimer’s disease

Daniel Andrews,Douglas L Arnold,Danilo Bzdok,Simon Ducharme,Howard Chertkow,D Louis Collins,

DOI: https://doi.org/10.1002/alz.074336

2023-12-01

Abstract:Abstract Background Randomized clinical trials of Alzheimer’s‐modifying drugs typically report a clinical effect as the final observation difference in means for a cognitive endpoint between treated and placebo groups. However, randomization of subjects naturally following different cognitive decline trajectories could produce a between‐group endpoint difference even when a drug has no disease‐modifying effect. Recent work used data from ADNI subjects matching trial inclusion criteria to simulate patient trajectories [Jutten et al. (2021). https://doi.org/10.1212/WNL.0000000000012022 ]. We build on such simulations by incorporating a trial’s specific design parameters, including multiple visits, visit time windowing, dropout rate, and measurement noise using jittering. We estimate treatment‐independent group differences in a cognitive endpoint’s rate of change in simulations of the Phase 3 aducanumab and lecanemab trials. Method 563 ADNI subjects matched inclusion criteria for the identically designed Phase 3 aducanumab trials EMERGE and ENGAGE [clinicaltrials.gov/ct2/show/NCT02484547]. We fitted a continuous time linear mixed effect model tracking CDSRB change from baseline (CDRSBΔbl). ADNI subjects’ baseline data were resampled and jittered, generating data for 1,070 synthetic subjects total while preserving the real trials’ baseline Alzheimer’s stage distribution. Subjects were randomized 1:1 to “treated” and “placebo” groups, but without any treatment effect. We simulated 4 visits, ∼26 weeks apart, with 30% dropout. Our sample size and visit scheme matches the real trials’ Statistical Analysis Plans. We generated 10,000 simulated trials, each a unique randomization of subject trajectories. The “treated” vs. “placebo” CDRSBΔbl linear slope difference was extracted from each simulation. We repeated this procedure for the demographics and design of lecanemab’s Phase 3 trial Clarity AD [clinicaltrials.gov/ct2/show/NCT03887455]. Result Histograms of simulated slope differences differ between the trial designs (see Figures). Averages are zero because there is no simulated treatment effect. Each successful trial’s real‐world difference (EMERGE high dose: –22%, Clarity AD: –27%) corresponds to a statistically significant reduction in cognitive decline rate for treated subjects vs. placebo. Conclusion Our results suggest low false positive probabilities for the successful trials. Cohort composition, sampling, and other trial characteristics might influence treatment‐independent group difference in cognitive decline rate. With our method, future trials could target treatment effects outside the target population’s false positive range.

clinical neurology