Brain experts have a pretty good handle on some of the major risk factors that contribute to Alzheimer’s—from a person’s genes to their physical activity levels, how much formal education they’ve received, and how socially engaged they are.
But one promise of AI in medicine is that it can spot less obvious links that humans can’t always see. Could AI help uncover conditions linked to Alzheimer’s that have so far been overlooked?
[time-brightcove not-tgx=”true”]
To find out, Marina Sirota and her team at University of California San Francisco (UCSF) ran a machine-learning program on a database of anonymous electronic health records from patients. The AI algorithm was trained to pull out any common features shared by people who were ultimately diagnosed with Alzheimer’s over a period of seven years. The database includes clinical data, such as lab and imaging test results and diagnoses of medical conditions.
“There were some things we saw that were expected, given the knowledge that we have about Alzheimer’s, but some of things we found were novel and interesting,” says Sirota. The results were published in Nature Aging.
Heart disease, high cholesterol, and inflammatory conditions all emerged as Alzheimer’s risk factors—not surprising, since they’re known to contribute to the buildup of protein plaques in the brain. But the less expected conditions included osteoporosis in women and depression in both men and women. The researchers also saw unexpected patterns emerge closer to when people are diagnosed, such as having lower levels of vitamin D.
Sirota and Alice Tang, a medical student in bioengineering who is the lead author of the paper, stress that these factors do not always mean that a person will develop Alzheimer’s. But they could be red flags that a patient can address to potentially lower their risk. “Picking up these factors gives us clues that a diagnosis of Alzheimer’s might be coming, and things like [high cholesterol] and osteoporosis are modifiable [with treatments],” says Tang.
Whether or not treating these issues can actually lower a person’s risk of developing Alzheimer’s isn’t clear yet; the study wasn’t designed to answer that question. Sirota and her team plan to continue mining the database of health records to determine if people receiving treatments for conditions like osteoporosis or high cholesterol, for example, eventually had a lower risk of Alzheimer’s than patients who had those conditions but didn’t treat them. “We can retrospectively look at treatment data in the electronic medical records, so that’s definitely a direction forward to determine if we can leverage any existing therapies to lower risk,” says Sirota.
Tang also hunted for genetic factors associated with things like high cholesterol or osteoporosis and Alzheimer’s that could further explain the connection between these risk factors. The link between cholesterol and Alzheimer’s turns out to be related to the ApoE gene; scientists have known that a specific form of the gene, ApoE4, is associated with a higher risk of developing Alzheimer’s. Tang also identified a gene associated with both osteoporosis and Alzheimer’s that could become a new research target for a possible treatment.
The study shows the power of machine learning in helping scientists to better understand the factors driving diseases as complex as Alzheimer’s, as well as its ability to suggest potential new ways of treating them.
