An artificial intelligence (AI) tool was able to accurately predict Alzheimer's disease almost eight years before a person was diagnosed, according to a study recently published in the journal EClinicalMedicine, and researchers say the tool could help providers to identify patients in early stages of the disease.

AI vendor evaluation guide: Separate the signal from the noise

Study details

For the study, which was funded by Pfizer, scientists from IBM Research used algorithms to train an AI tool to predict the eventual onset of Alzheimer's disease by analyzing over 700 written samples from 270 participants in the Framingham Heart Study. The Framingham Heart Study has been tracking thousands of participants since 1948 and collecting detailed medical histories, physical exams, and lab tests from the participants.

None of the participants who provided the writing samples used in the IBM study showed signs of memory loss at the time they submitted the samples. However, the researchers found that the AI tool was able to accurately predict up to 74% of the participants who eventually were diagnosed with Alzheimer's disease. In addition, the AI tool made those predictions, on average, 7.6 years before the participants were diagnosed with the disease.

According to the researchers, risk factors for Alzheimer's that can be found in language include repeating questions, stories, or statements, as well as agraphia, which is the loss of writing ability. The researchers noted that the AI tool more accurately predicted Alzheimer's diagnoses than other common screening methods, including evaluating whether a person is genetically susceptible to developing the disease, a person's demographics, or psychological tests.

Could the AI tool change Alzheimer's screening?

Ajay Royyuru, VP of health care and life sciences research for IBM, said the AI tool could work as a non-invasive test that "presents a better window for targeted interventions."

Royyuru suggested that health care providers could perform language pattern tracking as part of routine physicals or a behavioral health exams for patients. For example, he suggested that doctors could gather a baseline sample of a patient's language skills as a young adult, and then collect an updated sample every five years for comparison.

"That is not in normal clinical practice today," Royyuru said, but "[t]he technology allows us to think about this as something that would be possible."

In addition, Royyuru said the AI tool could be used to identify patients at the right stage of Alzheimer's disease that's needed for some clinical trials, which could help with the development of a drug to prevent or treat the disease.

But Oscar Lopez, a professor of neurology and psychiatry and director of the University of Pittsburgh's Alzheimer's Disease Research Center, said using a language test to identify patients who may develop Alzheimer's disease could be difficult, because it would first require acquiring reliable data from studies that track patients for decades to prove its effectiveness.

Instead, many Alzheimer's researchers are focused on developing blood tests to identify people who may develop the disease before symptoms occur by detecting certain biomarkers, such as the tau protein, USA Today reports. However, research on such tests is ongoing.

"Blood tests are going to be the future, but we're still not there," Lopez said (Alltucker, USA Today, 10/22; Owens, "Vitals," Axios, 10/23).