Development of digital voice biomarkers and associations with cognition, cerebrospinal biomarkers, and neural representation in early Alzheimer's disease.
Level 3 - non-randomized controlled study
Observational cohort study assessing diagnostic accuracy and prospective disease progression
PubMed 36777093 · doi:10.1002/dad2.12393
What was done
Connected speech recordings, neuropsychological assessments, neuroimaging, and cerebrospinal fluid (CSF) Alzheimer's disease (AD) biomarker data were collected from 206 participants: 92 cognitively unimpaired (40 amyloid-β positive [Aβ+]) and 114 cognitively impaired (63 Aβ+). Machine learning models were used to extract lexical-semantic and acoustic features from audio recordings to evaluate diagnostic performance for mild cognitive impairment (MCI), biological AD status, and 2-year disease progression.
What was found
Lexical-semantic scores achieved an area under the curve (AUC) of 0.80 and acoustic scores achieved an AUC of 0.77 for detecting MCI, outperforming the Boston Naming Test (AUC = 0.66). Only lexical-semantic scores differentiated amyloid-β status (p = 0.0003) and correlated with CSF amyloid-β (p = 0.007). Acoustic scores correlated with hippocampal volume (p = 0.017). Both voice biomarker measures were significantly associated with 2-year disease progression.
Why it matters
Automated acoustic and lexical-semantic voice features provide a non-invasive, scalable digital biomarker reflecting molecular and structural pathology in early Alzheimer's disease.
Limits
The sample size is moderate (n = 206) and lacks reported external validation across diverse accents, dialects, and recording conditions. Exact effect sizes, hazard ratios, and numerical risk estimates for 2-year progression were not provided in the abstract.
Cited by
- supports Linguistic and speech cues of neurodegenerative conditions like Alzheimer's disease can be detected up to 10 years before typical clinical symptoms appear.