V-71
Disorders of the Nervous System
Digital Speech Biomarkers for Cognitive and Brain Health Across the Alzheimer’s Disease Continuum
Ivan Caro1,2, Gonzalo Pérez1,2,3, Joaquín Valdés Bize1, Franco J. Ferrante1,2,3, Lara Gauder3, Alejandro Sosa Welford1,2, Joaquín Ponferrada1, Luciana Ferrer3, Agustin Ibañez1,5, Andrea Slachevsky4, Adolfo M. García1,5
1. Cognitive Neuroscience Center (CNC), Departamento de Ciencias de la Vida y del Comportamiento, Universidad de San Andrés, Buenos Aires, Argentina.
2. Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Buenos Aires, Argentina.
3. Universidad de Buenos Aires, Buenos Aires, Argentina.
4. Geroscience Center for Brain Health and Metabolism (GERO), Santiago, Chile.
5. Global Brain Health Institute (GBHI), University of California San Francisco (UCSF); & Trinity College Dublin, Dublin, Ireland.
Presenting Author:
ivan.caro.strokes@gmail.com
Assessing the Alzheimer’s disease continuum (ADC) relies on resource-intensive neuropsychological testing and neuroimaging. Digital speech biomarkers (DSBs) provide a scalable approach, but their ability to predict cognitive and brain health across the ADC is unexplored in Latino populations. We collected speech, cognitive, and neuroimaging data from 150 participants (17 healthy controls, 55 with subjective cognitive decline, 57 with mild cognitive impairment, 21 with Alzheimer’s disease dementia [ADD]). DSB features came from brief TELL-app fluency tasks, cognitive features from standardized-test scores (ACE-III, IFS), and brain features from dementia-related MRI regional volumes. Separate machine-learning models used DSB or brain features to predict ACE-III/IFS scores, and DSB or cognitive features to predict hippocampal and ADD-atrophy mask volumes. Performance was assessed with age-adjusted partial Pearson correlations. DSB-gray matter volume associations were tested with multiple regression. DSBs outperformed brain features for ACE-III (r=.67 vs .23) and IFS (r=.55 vs .31), and cognitive features for ADD-atrophy mask volume (r=.38 vs .27), but not hippocampal volume (r=.42 vs .37). Higher word frequency was associated with lower prefrontal cortex volume. DSBs captured cognitive status and structural brain health across the ADC via a fully automated multivariate pipeline, supporting DSBs as a low-cost, scalable alternative for dementia screening and monitoring.
