Alternative Endpoints and Models for AD Prevention Trials

Abstract

AbstractBackgroundThere are several challenges for AD prevention trials. First, the subtle cognitive change in asymptomatic disease stages requires large sample sizes, e.g. 400‐500/arm for two phase 3 confirmatory trials, or 800‐900/arm for a single phase 3 confirmatory trial, using the traditional MMRM. Second, large, international trials may necessitate the use of different tracers for imaging biomarker assessment due to global availability issues, however, statistical models that harmonize tracer differences within clinical trials are still lacking. Third, primary prevention trials using time to event endpoint demands even larger sample sizes, e.g., 1650/arm for TRAILBLAZER‐ALZ 3 with time‐to‐first‐conversion from CDR global 0 to >0 as the outcome. The percentage slowing relative to placebo decline (e.g., lecanemab led to 27% slowing at 18 months) is a popular concept to describe treatment effect. Instead of estimating the absolute difference using MMRM, we proposed to model this percentage treatment effect directly using the percentage or proportional MMRM (pMMRM). pMMRM can harmonize assessments from different tracers by converting the absolute difference into a shared proportional treatment effect across tracers. When using time‐to‐conversion as the primary outcome, instead of using a single event, Cox proportional hazards models for recurrent events can capture the treatment effect after the first event, where a recurrent event is defined as any increment over the previous visit in CDR global.MethodWe first demonstrated that disease modifying treatment effects are essentially proportional retreatment effects. Then, we evaluated the power advantage of pMMRM over MMRM and Cox recurrent events model over single event model through simulations. Finally, we demonstrated how to harmonize data from different tracers using pMMRM.ResultFigure 1 demonstrates that disease modifying treatment effects are like proportional treatment effects, leading to diverged disease progression trajectories and increased absolute differences. Figure 2 shows the power comparison between MMRM and pMMRM. pMMRM led to a power increase up to 30% over MMRM.ConclusionAlternative models such as pMMRM or Cox recurrent event models can yield greater power and pMMRM can harmonize assessment from different tracers, which may help enable global prevention trials in AD with disease modifying drug effects.