P4-024: the relative efficiency of time-to-progression and continuous measures of cognition in preclinical alzheimer's

Clinical trials in Preclinical Alzheimer's are challenging due to the slow rate of disease progression as measured by periodic cognitive performance tests or by transition to a diagnosis of Mild Cognitive Impairment (MCI). We use a simulation study to demonstrate that models of repeated cognitive assessments detect treatment effects more efficiently compared to models of time-to-progression to MCI or dementia. We also explore the bias induced by hypothetical non-ignorable missingness. Multivariate continuous data are simulated from a Bayesian joint mixed effects model trained on data from the Alzheimer's Disease Neuroimaging Initiative. Simulated progression events are algorithmically derived from the continuous assessments using a random forest model trained on the same data. Simulated clinical trials are then analyzed using either (1) a continuous cognitive composite outcome or (2) time-to-progression to MCI or dementia outcome (Figure 1). Non-ignorable missing data due to tolerability or perceived lack of efficacy is simulated according to the pattern described in Table 1. We find that power is approximately doubled with models of repeated continuous outcomes compared to the time-to-progression analysis (Figure 2). The simulations also demonstrate that a plausible non-ignorable missing data pattern can induce a bias which inflates treatment effects, yet 5% Type I error is maintained.