A machine learning program that analyzes recordings of primary care visits might help identify people with cognitive impairment, according to a
study published today in
JAMA Neurology.
Why It’s Relevant
While the U.S. Preventive Services Task Force has found
insufficient evidence to recommend universal screening for dementia in older adults, there are potential advantages to early detection of cognitive impairment—including the opportunity to address comorbid conditions or medications that may be contributing to the problem or enable patients to begin therapies that may
slow cognitive decline.
Primary care is an ideal setting to conduct cognitive tests, but time and resources are limited. Automated screening methods that use existing data such as health records or patient dialogue might be a solution.
By the Numbers
- Researchers tested multiple speech-recognition programs to identify possible cognitive impairment from audio clips of primary care dialogue from 787 older adults (average age 67) using acoustic clues such as vocal pitch and pause duration.
- The best-performing model (OpenAI’s Whisper) was then validated using audio from 179 different patients. It performed with an overall accuracy of around 73% (32% false negative rate and 36% false positive rate).
- In terms of predictive value, about 30% of patients identified as having cognitive impairment did indeed have it as assessed with the Montreal Cognitive Assessment (MoCA).
The Other Side
The recruited patients were oversampled for cognitive impairment to improve training and validation; more than 20% of participants had elevated MoCA scores. The researchers noted that if the speech program were used in a clinic where 10% of patients had cognitive impairment as opposed to 20%, the predictive value would drop to 16%.
Takeaway Message
The researchers speculated that speech-recognition programs might be useful as an initial screening assessment that would flag patients who might benefit from a brief clinician- or staff member–administered assessment such as the MoCA.
Related Information
Source
Joseph T. Colonel, et al. Acoustic analysis of primary care patient–clinician conversations to screen for cognitive impairment. JAMA Neurology. Published June 15, 2026. doi:10.1001/jamaneurol.2026.1868
(Image: Getty Images/iStock/FatCamera)