The assistant professor in the department of neurology at the Albert Einstein College of Medicine and Montefiore Medical Center talked about predictive models in Alzheimer disease for clinical trials.
The enrollment of patients who are unlikely to show meaningful cognitive decline with placebo may make it more difficult to show the benefits of active treatment for cognition. Recent research used data from the placebo arm of 5 phase 3 trials, showing that predictive machine learning models can potentially increase sensitivity to effects from treatment and reduce the requirements for sample size in clinical trials.1
In total, 1982 patients were included in the pooled placebo analysis, with meaningful cognitive decline not observed in 42% to 58% of individuals at the end of trials. Using the predictive machine learning models, positive predictive values were approximately 12% to 25% higher than the sample rate of meaningful cognitive decline. Notably, negative predictive values of models were approximately 15% to 24% higher than the base rate of patients who had stable cognition at the end of trial.
Ali Ezzati, MD, assistant professor, department of neurology, at the Albert Einstein College of Medicine and Montefiore Medical Center, presented this study during the experimental therapeutics in dementia session at the 2023 American Academy of Neurology (AAN) Annual Meeting, April 22-27, in Boston, Massachusetts. During the meeting, Ezzati sat down with NeurologyLive® in an interview to talk about the reason behind the difficulties and failures in clinical trials for Alzheimer disease (AD). He also spoke about the findings from his study that were presented, and the proposal to improve the design of trials using machine learning predictive models.
Original Article in NeurologyLive
