When Doctor Doesn’t Check for Cognitive Decline, AI Might

Obstacles to Dementia Screening

Although a cognitive assessment is covered under Medicare’s annual physical, fewer than one-third of Medicare beneficiaries receive one. Another study from 2020 found that primary care physicians do only a brief screening for possible dementia in about half of patients aged sixty-five and older. 

A doctor may not have the time or inclination to do a formal cognitive assessment during a patient's visit. Appointment time is limited. Patients can find it stressful. A thorough dementia screening is more expensive than checking for other chronic conditions. How to deliver a diagnosis the patient or family is an emotional and practical burden for the doctor. 

In fact, some experts even argue that early detection is not necessary at this point when there are so few treatments. Advocates for dementia screening say that early diagnosis allows patients to understand their future and make plans for long-term care. 

Artificial Intelligence Screening 

Recently, researchers developed an artificial intelligence system that can identify signs of cognitive impairment simply by analyzing the sound patterns in conversations patients have with their doctors during an appointment.

Rather than utilizing specialized memory tests or lengthy evaluations, this AI detects subtle changes in speech that may signal early cognitive problems. The findings, published in JAMA Neurology, show that ordinary conversations contain clues about brain health that are difficult for humans to recognize but can be detected by machine learning algorithms.

The study included 966 adults ages 55 and older who had no previous diagnosis of mild cognitive impairment or dementia. Participants were recruited from primary care practices in New York City and Chicago.

Researchers recorded routine doctor-patient conversations and extracted multiple 30-second speech samples. They then used machine learning models to analyze acoustic features of speech, including characteristics related to pitch, timing, rhythm, and variability.

By comparing speech analysis results with participants' scores on a standardized cognitive assessment, researchers sought to determine whether the technology could identify people with previously unrecognized cognitive impairment.

The AI Successfully Identified Cognitive Impairment

About 21% of participants met the study's definition of cognitive impairment despite not having a prior diagnosis.

When used as a screening tool, the algorithm correctly identified about 68% of people with cognitive impairment and correctly ruled out impairment in about 64% of people without it. The best-performing AI model was able to identify these individuals with a moderate to high level of accuracy. The findings remained consistent when the system was tested in a separate group of patients from another city, suggesting the results may generalize beyond a single healthcare setting.

At this point, this technology is not accurate enough to replace formal cognitive testing. It would help identify patients for further evaluation.

Speech and Brain Health

Speaking is a surprisingly complex task.

The brain must retrieve words, organize thoughts, control breathing, coordinate facial muscles, and monitor the flow of conversation. Even subtle changes in these processes can alter the rhythm and acoustic qualities of speech.

Previous research has shown that neurological conditions such as Alzheimer's disease and other forms of dementia can affect speech long before severe symptoms appear. Characteristics related to pitch, timing, and variability were among the most important predictors used by the AI system.

A Potential New Screening Tool

Very appealing about this technology is that it could be integrated into healthcare visits. While a doctor is focused on what a patient is saying, AI can focus on how a patients is speaking.  

If implemented, AI results could provide a time-efficient way to identify those patients who should have a more comprehensive evaluation. Rather than asking patients of a certain age to complete separate screening tests, clinicians could potentially use speech analysis as a passive tool to identify people who may need further assessment.

Such an approach could be particularly valuable because many individuals with early cognitive impairment remain undiagnosed.

More work is needed before this type of technology can be adopted in routine clinical practice.

Future studies will need to determine how well the system performs in more diverse populations. Also, it is possible that with future research and development, AI technology could detect cognitive decline at even earlier stages.