AI gives primary care a larger role in battling dementia

©Ivelin Radkov-stock.adobe.com

“The computer will see you now” headline, or something similar, flashes across my news feeds regularly. They represent the very real concern over the dehumanization of medical care as technologies get more advanced. However, as a primary care doctor, I believe the right technology can aid in providing compassionate and comprehensive care, augmenting what providers do rather than replacing them. One of those technologies is the use of artificial intelligence (AI) to identify, treat, and ultimately help prevent dementia.

As generative AI and ChatGPT news has sparked public interest, a recently-published study added to the excitement by demonstrating GPT-3’s potential role in detecting Alzheimer’s dementia from speech. The study showed high accuracy in detecting dementia based on voice and text analysis from tasks like describing a complex picture. The results are promising, but the greatest benefit of the article may be that it has highlighted what researchers have demonstrated for years: AI can detect dementia.

Over the last decade, researchers have demonstrated the ability to detect dementia with speech, drawing tasks, targeted MRI algorithms, and adaptive testing. Studies have also demonstrated AI’s ability to detect more subtle declines in thinking, known as mild cognitive impairment (MCI).

Recent studies have even shown that some forms of imaging AI and drawing process analysis correlate with brain pathway impairment, changes in the structure of the brain, and even brain pathology, such as amyloid beta and tau deposits, which are the earliest signs of Alzheimer’s. However, rather than being tools to replace clinicians, these technologies can help maintain the humanity of medical care if they can be used effectively by frontline clinicians, especially in primary care.

The ability for primary care providers to identify cognitive impairment before it progresses to dementia can give hope to millions. Treatment options for early Alzheimer’s have advanced. For example, Leqembi (Lecanemab), an FDA-approved medication that has shown a 27% decrease in the progression of the disease, is now available and covered by the Veterans Health Administration; though this treatment is not a fit for all patients or their specific cases of MCI.

Lifestyle interventions, on the other hand, are appropriate for all patients with MCI. Studies indicate that 40% of all dementia is preventable and intervention trials like the FINGER study have demonstrated that lifestyle interventions targeted at the highest-risk elderly can reduce progression to dementia by over 25%. Recent international studies have also demonstrated that lifestyle interventions can reduce the risk of progressing from cognitive impairment to dementia by more than 30%.

Because all known treatment options are shown to prevent or delay - but not reverse - dementia, early identification is key. Just like we need mammograms to detect breast cancer early, we need actionable technology to identify cognitive impairment early. The most effective AI solutions won’t just detect cognitive concerns, but also give PCPs a more efficient way to determine if patients have MCI and inform further testing (blood or imaging-based), streamlining diagnosis for those who would benefit from drug treatments and reducing unnecessary testing for those who do not. This knowledge can help PCPs guide their patients on appropriate, multifaceted interventions.

The ability for AI to augment early detection is both exciting and very real. Devices and technologies are already on the market to detect MCI and help guide diagnostic pathways in primary care clinics. Imaging algorithms can already be used by radiologists to detect subtle changes in brain pathology and definitive pathology testing can be done with Positron Emission Tomography amyloid imaging. The challenge is getting these incorporated into the standard of care and ensuring their cost-effectiveness.

It is important to remember that despite all these advances, AI that doesn’t integrate into clinicians’ workflows, or requires additional provider time to use won’t become the standard of care. It’s also necessary to be mindful that not all “AI solutions” are created equal and their utility relies on their ability to accurately detect cognitive impairment/pathology without bias. AI trained on limited data sets, or designed poorly, may appear accurate on the surface, but have hidden errors that increase inequalities.

The economic value of using effective and actionable AI solutions to identify and triage individuals with MCI is clear; the ability to give patients hope and provide them with definitive health action plans is priceless. Combined, this is one of the best uses of AI in modern medicine.

John Showalter, MD, MSIS, is a dual board-certified primary care physician and chief product officer at Linus Health, a digital health company focused on early detection in brain health.