AMHERST, Mass. — University of Massachusetts Amherst researchers and scientists at Embr Labs, a Boston-based start-up, have developed an AI-driven algorithm that can accurately predict nearly 70% of hot flashes before they're perceived. The work, featured in the journal Psychophysiology , will be incorporated into the Embr Wave, a wearable wrist device clinically proven to manage hot flashes.
In the U.S. alone, an estimated 1.3 million women transition into menopause annually, and 80% of women experience hot flashes — sudden feelings of intense heat, often radiating in the upper body. Most hot flashes occur during this transition, but half of women experience them for 7 years or more. The physiological response had, for decades, been considered more of a nuisance than a treatable condition by medical professionals. However, recent research has linked hot flash severity and related sleep disturbances to risk of cardiovascular disease.
"Hot flashes have been dramatically overlooked," says Matt Smith, co-founder and CTO of Embr Labs . "Even 50 years ago, hot flashes were still considered to be psychosomatic. To our knowledge, our work is the first attempt to undertake a rigorous effort to achieve the prediction of hot flashes. This breakthrough discovery and the publication of our findings are the result of incredibly deep data science that answers the question: "Is it possible to accurately predict a hot flash before a woman feels it?"
The answer is "yes," and presents the opportunity to incorporate this science into the next version of the Embr Wave that already provides relief for women in menopause. Adding the predictive feature into its next-generation device means that hot flashes can be mitigated in real time.
"This study opens this door for real-time, closed-loop digital therapeutics to exist," says Mike Busa , clinical professor and director of the Center for Human Health & Performance at the UMass Amherst Institute for Applied Life Sciences, and corresponding author on the paper. "Not every target has to be druggable. If we can detect hot flashes early, Embr has a wearable solution that might be able to mitigate symptoms and provide meaningful relief."
This industry-academic partnership leverages the full breadth of both the university's and Embr Labs' expertise, talent, and innovation to spur job creation, entrepreneurship and workforce development locally, regionally and across the state.
To develop the model, which accurately identified 82% of hot flashes and predicted nearly 70% of them, on average, 17 seconds before they were perceived, Busa's team collected various physiological data points from peri- or postmenopausal women. They found that skin conductance, the amount of electricity that can be conducted by the surface of the skin, provided the best signal. While hot flashes are notorious for causing profuse sweating, the researchers found that even the initial, imperceptible increase in water and salt on the skin that accompanies the start of a hot flash was sufficient to predict an impending hot flash event.
The researchers then evaluated the peaks in skin conductance around a hot flash compared to self-report perception of a hot flash, as well as expert characterization of events that occur without being perceived by the test subject, such as when they are distracted or asleep.
"We went through a very rigorous tuning process and a lot of different permutations of different ways that you can examine the data," says Busa. "The mathematical consequences of any one function really can impact the fidelity of what you're able to see in that data," particularly when considering the size of the time window around the peak event.
Next, they used a separate data pool that was not used to train the model to see which model had the best predictive capabilities. Their best model was able to identify 82% of hot flashes that occurred from 60 seconds before to 30 seconds after a hot flash was perceived by the study participants. When looking just at its predictive capacity, the model identified 69% of hot flashes 17 seconds, on average, before the onset of perceived symptoms.
Busa highlights that this academic-industry partnership keeps a constant eye on the practicality for the end-user. "We're [creating this model] in a way that it goes beyond having the potential of only a retrospective research tool," he says. "[Embr's] wearable sensors have the potential to make real-time meaningful impacts on people's lives, opposed to just creating a graph or chart that can be analyzed later."
"The end goal of this collaboration, which started in 2019, was always focused on enabling a closed-loop solution that could predict hot flashes and then deliver cooling, using the thermal technology that Embr Labs has developed and the deep expertise of Mike and his team at the Institute for Applied Life Sciences," Smith adds.
This research was supported by grants from the Massachusetts Life Sciences Center Women's Health Initiative, awarded to UMass Amherst, and the U.S. National Science Foundation to Embr Labs.
