Machine learning spots the time to treat Huntington disease

Abstract

\n We propose a new approach, based on machine learning, to evaluating new treatments for neurodegenerative diseases. Using data from two longitudinal studies of 299 participants with early Huntington Disease, we learned the range of likely trajectories of 15 imaging and clinical biomarkers from the premanifest to manifest disease stages. We positioned independent 11,510 patients on these maps using their baseline data and hence forecast the values of their biomarkers at any timepoint. Applied to trial design, we showed that sample size can be decreased by up to 50% by selecting individuals for whom we predict a significant change for their outcome measures during trial. Reduction occurs whatever the selected outcome measures and the targeted disease stage. This approach does not only select the right patient at the right time for the right trial, but also guides decisions about when to start preventive treatments.