Imagine a world where a simple wearable gadget, powered by artificial intelligence, doesn’t just monitor your blood sugar but foresees potential issues hours in advance, empowering you to take control of your health before problems even arise. For the millions battling Type 2 diabetes or teetering on the edge with prediabetes, this isn’t science fiction—it’s the promising frontier of modern medicine that’s finally getting the deep dive it deserves.
In a groundbreaking effort coming out of Buffalo, New York, experts from the University at Buffalo have just released what they call the first-ever comprehensive meta-review—a thorough analysis of dozens of existing studies—on how AI-boosted wearable tech can revolutionize care for folks with prediabetes and Type 2 diabetes. Published in the respected journal NPJ Digital Medicine (check it out here: https://www.nature.com/articles/s41746-025-02036-9.epdf?sharingtoken=1ZG6YU0FnDog6aGGI4GIFdRgN0jAjWel9jnR3ZoTv0MVO8SvZ899ldT91kqIUPfeOCWjGdwoNSDV3URkq6WAmh3JHiUvWbDrZ-HfBBzZ0v7utKRKhp-swTTMr6mHT4nO4byVHljEn4iw6Ykm_zCV4LMgKISsthWpmcORKJzDoE%3D), this work spotlights the huge untapped potential of these devices. The big takeaway? They’re game-changers waiting to happen, but only if we tackle a few key hurdles head-on.
What really sparked this research for lead author Raphael Fraser, PhD—an associate professor of medicine at the University at Buffalo’s Jacobs School of Medicine and Biomedical Sciences (learn more about him here: https://medicine.buffalo.edu/faculty/profile.html?ubit=rfraser)—was the sheer power of continuous glucose monitors, or CGMs. These aren’t your old-school finger-prick tests that give you just a handful of readings a day. No, CGMs deliver real-time data every few minutes, painting a vivid, ongoing picture of your glucose levels. For beginners, think of a CGM as a smart patch or wristband that measures blood sugar non-stop without needles, sending alerts to your phone or device.
‘From looking back to looking ahead’
Fraser recalls the ‘aha’ moment vividly: As advanced AI algorithms started spotting patterns in glucose data and forecasting shifts before they even occurred, it hit him that diabetes management could evolve from a reactive scramble—dealing with highs and lows after the fact—to a proactive strategy that nips issues in the bud. ‘It was like upgrading from a rear-view mirror that shows you what just happened to a heads-up display that warns you about what’s coming next,’ he shares. This shift could truly overhaul everyday routines and lead to better health over the long haul.
For those navigating life with diabetes, these AI-smart wearables promise on-the-spot, tailored advice that keeps glucose levels stable and builds confidence in handling meals, exercise, and stress. Clinicians, meanwhile, stand to gain from early risk detection and streamlined patient care. And here’s a teaser for where early intervention shines: While we need bigger trials to confirm benefits for prediabetes specifically, imagine starting with a CGM and AI combo right away—it could nudge people toward healthier habits like balanced eating and regular movement, potentially staving off full-blown diabetes for years.
The field is exploding with new research, but up until now, it’s been a bit of a patchwork quilt: studies scattered across various gadgets, data sources, and AI techniques, leaving the overall landscape fuzzy and hard to grasp.
Piecing together the puzzle and spotlighting the blind spots
That’s why Fraser and his team set out to consolidate it all. ‘We aimed to sift through the noise, pinpoint what the solid evidence tells us, highlight reliable trends, and flag the areas where we still need more clarity,’ he explains. Their mission? To spotlight the top-performing strategies, uncover roadblocks, and map out the research gaps that must be filled before these tools slide seamlessly into everyday doctor visits and patient lives.
Working within UB’s Division of Population Health in the Department of Medicine, the researchers combed through over 5,000 peer-reviewed papers and zeroed in on 60 that truly delved into blending AI with wearable tech for diabetes care. The results were encouraging on many fronts. ‘These AI-enhanced devices can forecast glucose fluctuations up to one or two hours ahead, allowing users to stay on an even keel and get advice customized to their unique lifestyles—think factoring in your workout intensity, sleep quality, or even how busy your day gets,’ Fraser notes. For example, if you’re someone who loves evening jogs, the AI might predict how that activity stabilizes your levels and suggest tweaks to your pre-run snack.
Plus, these systems could lighten the load for healthcare pros by automatically sifting through mountains of data and flagging only the urgent bits, freeing up time for meaningful patient interactions rather than endless number-crunching.
But here’s where it gets controversial—and the part most people miss… The team also uncovered some serious sticking points that could make or break widespread use. For starters, these wearables rely on diverse AI models, and Fraser stresses that they all need to be open, tested rigorously, and proven reliable before going mainstream. Many current models are like mysterious black boxes: They spit out predictions, but no one can easily see inside to understand the reasoning. This opacity erodes trust, especially for doctors and patients who need to rely on these insights for critical choices.
Take a practical scenario: Your AI-powered glucose app buzzes with a warning that your blood sugar might spike in the next half-hour. Helpful? Sure, but if it doesn’t explain why—maybe it’s because of that carb-heavy lunch, a skipped walk, mounting work stress, a rough night’s sleep, or just natural daily ups and downs—how do you know what to do? ‘Without the ‘why,’ users are left guessing, which turns a potentially lifesaving tool into something frustrating and underused in the real world,’ Fraser points out. And this black-box issue? It’s sparking debates in the medical community—some argue it’s a necessary trade-off for accuracy, while others say transparency should never be compromised. What do you think: Is the promise of prediction worth the mystery?
Digging deeper, the studies often suffered from small participant groups and a lack of diversity in ages, ethnicities, and backgrounds, which means the results might not translate equally to everyone. There’s also no universal set of test data to compare outcomes fairly, leading to apples-to-oranges evaluations. On the practical side, issues like spotty data quality in trials, tricky ways to weave these tools into busy clinic routines, and the steep price tags or limited availability of the devices themselves are all holding things back from reaching more people.
Picking the perfect AI engine
One more layer to consider: The type of AI model powering your CGM can make a world of difference in how effectively it works for both patients and providers. ‘Not all AI is created equal—some excel at certain jobs,’ Fraser breaks it down. For continuous streams like glucose readings, models that specialize in time-based patterns, such as long short-term memory (LSTM) networks, shine because they remember past trends to predict what’s next. (Simply put, LSTMs are like a smart diary that connects yesterday’s entries to tomorrow’s forecasts.) Newer kids on the block, like transformer models—the tech behind tools like ChatGPT—handle multifaceted data beautifully, weaving in glucose with heart rate, sleep cycles, and movement to offer a well-rounded view of your body’s inner workings. For instance, if stress from a late night shows up in your heart rate data, the transformer could link it to a potential glucose dip the next morning.
That said, Fraser cautions that fancier isn’t always better. ‘Sometimes, straightforward models are more clinician-friendly because they’re easier to explain during a check-up,’ he says. The real trick is finding that sweet spot: An AI that’s powerful, interpretable, and reliable enough to discuss confidently in the exam room. In other words, the ideal model matches the data at hand and demystifies its logic for everyone involved.
Joining Fraser on this project were fellow UB Jacobs School experts: Rebekah J. Walker, PhD, associate professor; Jennifer A. Campbell, PhD, associate professor; Obinna Ekwunife, assistant professor; and Leonard E. Egede, MD, the Charles and Mary Bauer Endowed Chair of Medicine.
Funding for this vital work came from heavy hitters like the American Diabetes Association, the National Institute of Diabetes and Digestive and Kidney Diseases, and the National Institute on Minority Health and Health Disparities—proving that tackling diabetes inequities is a top priority.
So, as we stand on the brink of this AI-wearable revolution, one burning question lingers: Will the push for more transparent, accessible tech win out over the allure of complex black-box predictions, or are we risking unequal access that leaves some behind? Drop your thoughts in the comments—do you agree these devices could transform lives, or do the barriers seem too steep? Let’s spark a conversation!