Abstract
BACKGROUND: In Alzheimer's disease (AD) clinical trials, efficacy inference is traditionally based on the last visit (e.g., 18 months). However, recent studies suggest that disease-modifying treatment effects may emerge as early as 3 months post-baseline. OBJECTIVE: To explore this further, our study aimed to assess the increased statistical power achieved by incorporating all or multiple post-baseline visits to estimate treatment effect, compared to relying solely on the last visit. METHODS: We developed explicit formulas for the base functions of the natural cubic spline model, ensuring compatibility with standard SAS procedures. Through simulations using disease progression trajectories from ClarityAD and TRAILBLAZER-ALZ2 trials, we comprehensively evaluated various models in terms of power and type I error. Additionally, we offer SAS codes that to facilitate seamless implementation of different modeling approaches. RESULTS: Simulations based on ClarityAD and TRAILBLAZER-ALZ2 disease trajectories demonstrated that models incorporating multiple or all post-baseline visits yield greater power than those using only the last visit, while maintaining Type I error control. Furthermore, when three post-baseline visits were included, adding more visits resulted in minimal power gains. CONCLUSIONS: Our findings support prioritizing statistical models that incorporate multiple or all post-baseline visits for treatment efficacy inference, as they offer greater efficiency than models relying solely on the last visit.