Machine learning–enhanced screening funnel for clinical trials in Alzheimer's disease

Abstract

INTRODUCTION: Alzheimer's disease (AD) clinical trials with therapeutic interventions require hundreds of subjects to be studied over many months/years due to variable and slow disease progression. This article presents a novel screening paradigm integrating disease progression models to improve trial efficiency by identifying appropriate candidates for early phase clinical studies. METHODS: A traditional screening funnel is enhanced using machine learning models, including 3D convolutional neural networks and ensemble models, which integrate neuroimaging, demographic, genetic, and clinical data. RESULTS: This approach predicts clinical progression (2-year Clinical Dementia Rating Sum of Boxes change > 1) with an area under the curve of 0.836. Incorporating it into trials (with maximized sensitivity/specificity optimization) could reduce the number of subjects required by 55%, shorten recruitment by 13 months, and reduce screening amyloid positron emission tomography scans by 72%. DISCUSSION: By reducing patient burden and shortening timelines in clinical trials, this enhanced screening funnel could accelerate the development of AD therapies. Highlights: An innovative screening funnel was developed to improve Alzheimer's disease clinical trial efficiency. The funnel incorporates machine learning (ML)–based disease progression models. The ML model identifies patients with progression rate optimal for clinical trials. Unsuitable patients would fail early in the funnel before burdensome imaging procedures. This screening funnel is customizable to specific study needs.