Speech-based digital cognitive assessment for clinical trials: Detecting cognitive impairment stages and AD biomarker relations across European cohorts.
Level 3 - non-randomized controlled study
Multi-cohort observational validation study of a diagnostic marker
PubMed 42151661 · doi:10.1002/alz.71462
What was done
Researchers evaluated an automated speech biomarker for cognition (SB-C) in 736 participants across five European cohorts (BBRC Alzheimer's at-risk cohort, EPAD Scotland, DESCRIBE, LOCIDE, and BioFINDER-Primary Care). Participants completed verbal learning and semantic fluency tasks administered via automated phone or app-based platforms. SB-C scores were compared against standard cognitive measures (Mini-Mental State Examination, Clinical Dementia Rating, Preclinical Alzheimer Cognitive Composite 5) and cerebrospinal fluid (CSF) amyloid-beta and phosphorylated tau181 status.
What was found
SB-C significantly differentiated cognitively unimpaired and impaired participants (P < 0.001) and correlated with standard cognitive tests. For classifying CSF amyloid-beta and phosphorylated tau181 biomarker positivity, SB-C achieved an area under the curve (AUC) ranging from 0.56 to 0.82 across cohorts, with the strongest performance observed in the BioFINDER-Primary Care cohort. Exact correlation values and cohort-specific baseline statistics were not provided in the abstract.
Why it matters
Remote, automated speech analysis provides a scalable and low-burden method to screen for cognitive impairment and flag individuals with underlying Alzheimer's pathology for clinical trials without immediate invasive testing.
Limits
Classification performance varied widely across cohorts (AUC 0.56 to 0.82), indicating inconsistent accuracy across different clinical settings. Specific numerical correlation coefficients, sensitivity, specificity, and demographic breakdowns were not reported in the abstract. Cross-sectional speech assessment does not establish longitudinal disease tracking ability.