Speech patterns in responses to questions asked by an intelligent virtual agent can help to distinguish between people with early stage neurodegenerative disorders and healthy controls.

The sample size is very small, particularly for Alzheimer's disease (n=7), which limits the reliability of the findings. Participants were recruited via convenience sampling and were all white, of British descent, and native English speakers, limiting generalisability. Diagnoses were clinical and based on cognitive testing and imaging; no protein-based biomarkers were used to confirm Alzheimer's pathology. The study used an Intelligent Virtual Agent for data collection, which may elicit different speech patterns than a human clinician. Some participants had companions present during the recording, which could have influenced responses, although interactions where the companion spoke were excluded.

The research was partly funded by the NIHR Sheffield Biomedical Research Centre and the NIHR Sheffield Clinical Research Facility. The authors declared no other conflicts of interest.

Do not interpret the 76.6% prediction accuracy or the distinct pausing patterns as evidence that this method is ready for clinical diagnosis. The study had a very small sample size (only 7 people with Alzheimer's), and the participants were a homogeneous group (all white, British, native English speakers), so these results may not apply to your broader patient population.