A meta-review from the University at Buffalo, published in NPJ Digital Medicine, assesses the value and limits of AI‑enhanced wearable devices for people with Type 2 diabetes and prediabetes. The authors screened about 5,000 peer‑reviewed studies and selected 60 that test how AI and wearables work together in diabetes management. Continuous glucose monitors (CGMs) provide frequent glucose readings that AI models can use to recognise patterns and predict changes before they occur.
The review highlights tangible benefits: AI-enhanced wearables can predict glucose changes up to one to two hours in advance, helping users keep steadier control and receive personalised guidance tied to daily routines, activity and sleep. These systems can also reduce clinical workload by filtering large streams of data and flagging items that require attention. For people with prediabetes, early use of wearables combined with AI could support lifestyle changes and possibly delay or prevent progression to diabetes. Raphael Fraser, the corresponding author, says AI can turn CGMs from a “rear-view mirror into a heads-up display.”
At the same time, the review identifies significant shortcomings. Research has been uneven and often focused on a few device types, data forms and AI models. Many models act as "black boxes," which undermines trust and clinical use. Limited sample sizes, narrow demographic representation, the absence of standard benchmark datasets and inconsistent data quality make it hard to compare studies. Practical barriers include limited integration into clinical workflows and the cost and accessibility of devices. Different AI approaches suit different tasks: models that learn patterns over time, such as long short‑term memory networks, often handle continuous glucose data well, while newer transformer models can combine multiple data types. Simpler models may be easier for clinicians to interpret, so the right AI must balance performance and explainability.
The authors conclude that larger studies, stronger validation and more transparent models are needed before AI-enabled wearables become routine in clinical care. The research received support from the American Diabetes Association, the National Institute of Diabetes and Digestive Kidney Disease, and the National Institute for Minority Health and Health Disparities.
Difficult words
- meta-review — a study that summarises many other studies
- continuous glucose monitor — a wearable device that measures blood sugar continuouslyContinuous glucose monitors, CGMs
- predict — to say or estimate a future event
- black box — a system whose internal process is not visibleblack boxes
- benchmark dataset — a standard set of data for comparing modelsbenchmark datasets
- transformer model — an AI model that combines different data typestransformer models
- integration — the process of including something into a system
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Discussion questions
- What benefits and limits of AI-enhanced wearables seem most important for people with prediabetes or Type 2 diabetes?
- How should researchers balance performance and explainability when choosing AI models for clinical use?
- What practical steps could improve integration of these devices into clinical workflows and increase accessibility?
Related articles
Sudan turns to AI as health system struggles
Sudan’s health system is under severe strain after an almost two‑year civil war. A senior health official says the country is using artificial intelligence to help provide care where normal services no longer reach, while shortages and attacks worsen the crisis.
Clock disruption speeds breast cancer in models
New research finds that repeating disruption of the body’s 24-hour clock changes breast tissue, weakens immune defenses and leads to faster, more aggressive breast cancer in experimental models. The study links altered light–dark schedules to worse outcomes.