A data-driven framework for biomarker discovery applied to optimizing modern clinical and preclinical trials on Alzheimer's disease.

PET is used to measure tau protein accumulation in Alzheimer's disease. Multiple biomarkers have been proposed to track disease progression, most notably the standardized uptake value ratio of PET tracer uptake in a target region of interest relative to a reference region, but literature suggests these region choices are nontrivial. This study presents and evaluates a novel framework, BioDisCVR, designed to facilitate the discovery of useful biomarkers, demonstrated on [F]AV-1451 tau PET data in multiple cohorts. BioDisCVR enhances signal-to-noise by conducting a data-driven search through the space of possible combinations of regional tau PET signals into a ratio of two composite regions, driven by a user-defined fitness function. This study compares ratio-based biomarkers discovered by the framework with state-of-the-art standardized uptake value ratio biomarkers. Data used is tau PET regional measurements from 198 individuals from the Alzheimer's Disease Neuroimaging Initiative database, used for discovery, and 42 from the Mayo Clinic Alzheimer's Disease Research Center and Mayo Clinic Study of Aging (MCSA), used for external validation. Biomarkers are evaluated by calculating clinical trial sample size estimates for 80% power and 20% effect size. Secondary metrics are a measure of longitudinal consistency (standard deviation of linear mixed-effects model residuals), and separation between cognitive groups (-statistic of the change over time due to being cognitively impaired). When applied to preclinical (secondary prevention with CU individuals) and clinical (treatment aimed at cognitively impaired individuals) trials on Alzheimer's disease, our data-driven framework BioDisCVR discovered ratio-based tau PET biomarkers vastly superior to previous work, both reducing measurement error and sample size estimates for hypothetical clinical trials. Our analysis suggests remarkable potential for patient benefit (reduced exposure to health risks associated with experimental drugs) and substantial cost savings, through accelerated trials and reduced sample sizes. Our study supports the leveraging of data-driven methods like BioDisCVR for clinical benefit, with the potential to positively impact drug development in Alzheimer's disease and beyond.