By Sree Roy

When I think of artificial intelligence (AI) in sleep medicine, the first applications I think of are on the diagnostics side, such as autoscoring software solutions that shorten sleep study turnaround times. But increased efficiency is just the beginning of the uses for AI in sleep medicine. 

Within the next three to five years, I predict we’ll see AI-driven technologies launch on the therapy side—driving a personalized sleep medicine revolution.

Indeed, the article “CPAP Mask Selector Software Gets Smarter” begins the conversation of how sleep therapy—in this case, mask selection for CPAP devices—is changing due to machine learning and AI. What was beyond the scope of that article but bears mentioning here is a next likely step will be custom CPAP masks designed specifically for each patient. 

Patrick Karem, vice president at sovaSage, makers of TherapistAssist CPAP mask fitting software, says, “This is down the road for sure, but with 3-D mapping and printers, each patient will be able to buy a design guide from the manufacturer and then have a plethora of mask sizes to choose from.” While custom masks have been tried before, the increased efficiencies of AI-driven mask fitter platforms coupled with technology costs trending downward in the long-run makes me see a greater likelihood of success for such solutions in the future.

With its ability to swiftly analyze both big data from hundreds of thousands of individuals and enormous data sets from a single person, artificial intelligence in sleep may also help identify specific daytime behaviors that impact nighttime sleep quality.

One company working on developing such sleep AI is Cerebra Health. “We know certain people metabolize caffeine fast and some people do so more slowly. So it would be valuable to identify personal behavioral triggers, for example, maybe a person would learn not to drink coffee after 1 pm because that makes them susceptible to sleep problems,” says Patrick Crampton, chief commercial officer of Cerebra. Whereas, for someone else, coffee late in the day may have no sleep impacts but a large meal in the afternoon may tank their sleep quality.

Because Cerebra’s home sleep test system incorporates a validated measure of sleep depth (ORP or “Odds Ratio Product”), the company has the means to get appropriate inputs for a sleep therapy-focused AI analysis. The key, Crampton says, is “How do we turn that objective, really detailed nightly feedback loop, combine it with an engaging platform, and combine that with user behavior during the day? And that’s where AI comes in, with lifestyle factors that can improve sleep.”

Crampton envisions a future in which an individual could learn how many hours of sleep they need when no behavioral triggers exist, as well as what triggers get them away from optimal sleep. He says, “That’s where this can go—into that true personalization of understanding your sleep need.”

AI can also incorporate horizontal data sets, that is, data from other people with similar tendencies—accelerating sleep therapy data knowledge further.

While these ideas are ambitious, there are iterations of therapy-side sleep AI in use today. For example, in Australia, Ambulance Victoria and the Victorian Level Crossing Removal Program have implemented a shift work scheduling system dubbed “AlertSafe” into their workflows. Developed by Monash University researchers and optimization tech company Opturion, AlertSafe’s algorithms use a mathematical model based on the underlying biology of sleep to estimate the impact of work schedules on alertness levels. 

“AlertSafe generates rosters using artificial intelligence-based optimization, which infers the consequences of each assignment of a shift to a person who can and cannot be assigned to other shifts. The platform then determines smarter ways to improve a roster time until it meets the preference needs of the roster and the people working within it,” says Mark Wallace, PhD, of the Faculty of Information Technology, in a release. 

AlertSafe was developed alongside a personalized sleep schedule app, called Zest, which optimizes individual sleep to improve mental health outcomes. The mobile phone app, which is currently in the testing phase, has already shown improvements among shift workers who have reported improved sleep and overall mental well-being.

While some people fret that more AI in sleep medicine will push sleep medicine professionals out of their jobs, I think artificial intelligence applications will always need humans behind them. 

“The best systems are supplemented by human access at some point,” Cerebra’s Crampton says. He cites diabetes management company Livongo as an example. Livongo gathers user input, such as weight, in a patient app to drive behavior change. But, ultimately, when patients’  blood sugar readings are out of range, a Livongo expert coach reaches out to them. Comparably in sleep medicine, a sleep coach must be on hand to speak to patients. 

Ultimately, having AI assist on the therapy side of sleep medicine would be advantageous for patients and professionals. As Crampton notes, “That’s a huge area of opportunity for sleep in the future.” I, for one, am looking forward to a sleep revolution.

Sree Roy is editor of Sleep Review.

Illustration 172909805 © Vs1489 | Dreamstime.com