Prediction of Alzheimer's disease progression within 6 years using speech: A novel approach leveraging language models.
Level 3 - non-randomized controlled study
Prognostic cohort study evaluating a machine learning prediction model on longitudinal follow-up data
PubMed 38924662 · doi:10.1002/alz.13886
What was done
Natural language processing and machine learning methods were applied to voice recordings from neuropsychological test interviews to predict progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD) within 6 years. Models were evaluated using speech data and demographic variables (age, sex, education level) from n = 166 Framingham Heart Study participants (90 progressive MCI and 76 stable MCI cases).
What was found
The best-performing models combining speech-derived features and basic demographics achieved an accuracy of 78.5% and a sensitivity of 81.1% in predicting MCI-to-AD progression within 6 years. Specificity and area under the ROC curve were not reported in the abstract.
Why it matters
This approach shows potential for developing fully automated, inexpensive, and remotely administrable screening tools to identify individuals at risk of developing dementia for clinical trials and early intervention.
Limits
The sample size is modest (n = 166) and derived from a single historical cohort (Framingham Heart Study), with no external validation cohort reported in the abstract. Key diagnostic metrics such as specificity and positive predictive value are not provided.