New Machine Learning Model May Help Predict Dementia, Harvard Medical School Researchers Report

Harvard Medical School researchers created a machine learning model which may help diagnose and predict dementia in older adults, according to a study published last Monday in the open-access medical journal JAMA Network Open.

The researchers' brain age index is derived from a machine learning algorithm that analyzes brain waves during different stages of sleep and models normal brain activity across the human lifespan, according to M. Brandon Westover, an assistant professor of neurology at the Medical School who supervised the research.

First author Elissa M. Ye, a data scientist at Massachusetts General Hospital who works in Westover’s lab, said patients with dementia tend to have brain activity that appears “older” than the patient’s chronological age.

“The brain age index is basically the brain age subtracted by the chronological age,” Ye said.

A high brain age index is "a great indicator of accelerated aging,” she added.

The study relies on 9,834 polysomnograms — records of brain waves, blood oxygen levels, heart rates, respiration, and muscle movements during sleep — collected by the Sleep Laboratory of Massachusetts General Hospital between 2009 and 2017.

“We are equipped with the world's largest sleep dataset,” Haoqi Sun, an instructor at Harvard Medical School who contributed to the study, said. “It's always growing.”

Westover said the team hopes the brain age index — in combination with other screening tools — can act as an “early warning system” for dementia and other neurological conditions.

In 2014, an estimated five million adults over the age of 65 suffered from dementia. That number is projected to reach nearly 14 million by 2060. A recent study suggests 61.7 percent of dementia cases go undiagnosed.

“There's a great diagnostic gap,” Ye said. “There's no really, really good, state-of-the-art, official way to diagnose dementia in the early stages. Usually, it's really hard to even conclude that someone has dementia, even when they're in the later stages.”

Westover also pointed to applications in testing the efficacy of preventative therapies and medicines.

“I'm thinking things like exercise or meditation,” he said. “There's actually a bunch of crazy things that people are trying that may actually work, like playing certain kinds of sounds that are time-locked to the brain activity that you have, or flashing lights in your eyes at a certain frequency.”

Unlike past algorithms of its kind — which require “expensive” and “cumbersome” procedures like lumbar puncture or magnetic resonance imaging — the brain age index can be measured at home using “easy-to-wear hats” that measure brain activity during sleep, according to Westover and Sun.

Sleep-tracking apps are popular on the Apple Store and Google Play Store, though Westover said more research is needed before his team’s algorithm will be ready for consumer or clinical applications. The next step on that road, he said, is removing the problem of night-to-night variability in brain activity, which he believes can be solved by averaging an individual’s brain age index over a period of multiple nights.

The team plans to use the algorithm to examine the effects of exercise on brain health and cognitive performance, he added.

“We're hoping to show that you can actually, by exercising, make your brain younger,” he said.