Alexandra König, Felix Dörr, Elisa Mallick, Johannes Tröger, Anika Wuestefeld, Erik Stomrud, Pontus Tideman, Oskar Hansson, & Sebastian Palmqvist
*Poster was presented at ADPD 2025
Abstract
Aims: This study explores the potential of a digital speech biomarker for cognition (SB-C) to discriminate between the following FDA-recognised AD stages: 2 (asymptomatic, CDR=0, Aβ42/p-tau181- positive), 3 (subtle cognitive symptoms,CDR=0.5, Aβ42/p-tau181-positive) and 4 (mild cognitive impairment, CDR ≥1, Aβ42/p-tau181-positive), in order to provide an accessible screening solution in real-life clinical settings.
Methods: We utilized data from the Swedish BioFINDER-Primary Care study (N=144).Participants were classified into AD stages following the FDA Guidance for Industry definition based on amyloid and tau pathology confirmed through CSF, along with clinical assessments. The SB-C and subscores (executive function, memory, processing speed) were automatically extracted from recordings of Semantic Verbal Fluency (SVF) and RBANS List Learning tasks applying speech recognition and feature extraction.The SB-C is a z-score corrected for age and education. We found pairwise optimal cut-offs for the SB-C between the AD stages and evaluated discriminative performance by assessing sensitivity, specificity and balanced accuracy.
Results: Participantsʼ characteristics are presented in Table1. The SB-C demonstrated robust discriminatory power across all three FDA stages. The cutoff analysis showed optimal cut-offs of the SB-C normed z-scores of -0.8 to separate stage 2 and 3 and -1.3 to separate stage 3 and 4 respectively. With those cut-offs classification performance was at 77% accuracy (sens =.88, spec=.67) for stage 2 vs. 3 and for stage 3 vs. 4 (sens= .66, spec = .77) respectively. Combining FDA stage 2 and 3 and separating it from stage 4 yielded a balanced accuracy of 75% (sens = .75,spec =.77).
Conclusions: The SB-C demonstrates potential as a non-invasive tool for differentiating between AD stages, offering a practical and accessible solution for broader application in clinical settings and specifically primary care, where automated speech extraction could provide cost and time benefits.
