Zampeta-Sofia Alexopoulou, Stefanie Köhler, Johannes Tröger, Elisa Mallick, Nicklas Linz, Josef Priller, Anja Schneider, Klaus Fliessbach, Bjoern Falkenburger, Jens Wiltfang, Inga Zerr, Frank Jessen, Thomas Klockgether, Emrah Düzel, Annika Spottke, Michael Wagner, Christoph Laske, Stefan Teipel, and Alexandra König
* Talk presented at AAIC24, the Alzheimer’s Association International Conference, USA
Abstract
Background
Speech and language impairments are associated with cognitive decline in neurodegenerative dementias, particularly Alzheimer’s Disease (AD), where subtle speech changes may precede clinical dementia onset. As clinical trials prioritize early identification for disease-modifying treatments, digital biomarkers for timely screening become imperative. Digital speech-based biomarkers can be employed for screening populations at the earliest AD stages. An automated phone-based screening battery has been created, encompassing speech-related neurocognitive tests (Semantic Verbal Fluency and Verbal Learning Test). This allows the extraction of speech-based biomarkers alongside classical cognitive scores. This study aims to validate digital speech biomarkers in early-stage AD by comparing them to traditional evaluation methods.
Method
Within the PROSPECT-AD project, speech and gold-standard clinical data were obtained from the German DELCODE and DESCRIBE cohorts. We used data from N = 14 healthy controls (HC), N = 75 participants presenting with Subjective Cognitive Decline (SCD) and N = 18 participants presenting with amnestic Mild Cognitive Impairment (aMCI). Spearman rank correlations were computed between speech biomarkers and gold-standard clinical measures. Kruskal-Wallis test assessed group differences and regression analysis adjusted for age, sex and education, associated domain-specific speech biomarkers and cognitive assessment scores.
Results
There was a significant difference in the speech biomarker for cognition composite score (ki:e SB-C) between diagnostic groups (x2(2) = 18.06, p <0.001). Significant correlations were found between ki:e SB-C and all global anchor scores including MMSE (r = 0.48, d = 0.97, p <0.001), CDR-SoB (r = -0.49, d = -0.98, p <0.001) and PACC5 (r = 0.56, d = 1.12, p <0.001) (Figure 1). All domain-specific biomarker composite scores (memory, executive function, processing speed) significantly correlated with CDR, with strongest correlations found with the memory biomarker. All correlations remained significant when controlling for age, sex and education. Finally, based on the regression analysis results, domain-specific biomarkers were significantly associated with respective domain-specific anchors (Clock Drawing, TMT-A/B, Digits Span, Figure Drawing, ADAS-Cog, Verbal Fluency) (Table 1).
Conclusion
Findings support prior research, emphasizing speech biomarkers as a promising tool for remote early-stage AD screening, with potential implications for scalable screening in research trials and healthcare.