Using Predictive Models to Reduce Heterogeneity in Alzheimer’s Disease Clinical Trials
Ali Ezzati1, Kellen Petersen1, Bhargav Nallapu2, Richard Lipton1
1Albert Einstein College of Medicine, 2NEurology, Albert Einstein College of Medicine
Objective:
1. To investigate the proportion of individuals showing meaningful cognitive decline (MCD) in the placebo arm of Alzheimer’s disease (AD) trials during trials. 2. To evaluate data-driven predictive models for identifying participants who will show MCD if given placebo.
Background:
Enrolling individuals unlikely to show MCD with placebo treatment may make it more difficult to demonstrate the benefits of active treatment on cognition.
Design/Methods:
We used data from the placebo arm of five Phase III trials: two trials of semagacestat (LFAN and LZAN) and three trials of solanezumab (EXPEDITON1, EXPEDTION2, EXPEDITION3). We identified a group with MCD based on a score change on the cognitive subscale of the Alzheimer’s Disease Assessment Scale (ADAS-cog11; change of ≥4 from baseline to last trial visit), and a group with stable cognition (SC), defined as no MCD (change of <4 from baseline to last trial visit). Machine learning (ML) models were trained to classify participants into individuals with MCD vs. SC using baseline predictors (demographics, neuropsychological tests, genotype, and imaging when available). Predictive values of these models were evaluated against longitudinal rates of disease progression.
Results:
Total pooled placebo sample included 1982 individuals. MCD was not observed in 42-58% of placebo-treated participants at the end of trials. Positive predictive values of the predictive ML models were approximately 12-25% higher than the sample rate of MCD, while negative predictive values of models were approximately 15-24% higher than the base rate of participants who had SC at the end of trial.
Conclusions:
Despite enrollment in a clinical trial, 42-58% of participants in the placebo arm of five AD trials did not show MCD. Using predictive models can potentially increase sensitivity to treatment effects and reduce sample size requirements by increasing the proportion of participants with MCD enrolled in both arms of clinical trials.