- βEvolving AI outperformed traditional models at predicting future blood glucose levels
- βSystem accounts for complex real-world factors like meal type and exercise style
- βCould give patients earlier warning of hypos and hypers, reducing complications
| Journal | Scientific reports |
| Year | 2025 |
| Authors | Vaughan |
| PMID | 41291130 |
What Was Studied and Why It Matters
Keeping blood sugar in a safe range is one of the biggest daily challenges for people with diabetes. Too high for too long damages organs and nerves. Too low, even briefly, can be dangerous or life-threatening. Researchers wanted to know whether a smarter type of artificial intelligence (AI) could predict where blood sugar is heading β before it gets there.
Published in Scientific Reports in 2025, this study tested a new AI approach called an evolving neural network β a system that essentially teaches and improves itself over time β to see if it could forecast blood glucose levels more accurately than existing computer models.
What the Researchers Did
The team used data collected from continuous glucose monitors (CGMs) β the small wearable sensors that track blood sugar every few minutes. They fed this real-world data into two different AI systems and compared the results:
- A traditional AI model (back-propagation neural network)
- A new self-improving AI model (evolving neural network using neuro-evolution)
The evolving network is unique because it uses a process similar to natural selection β the AI automatically refines itself to get better at the task, rather than being manually programmed with fixed rules.
What They Found
The evolving neural network consistently outperformed the traditional model at predicting future blood glucose levels. It was better at accounting for the many factors that make blood sugar unpredictable, including:
- Meals β carbohydrates cause glucose to rise
- Aerobic exercise (like walking or cycling) β tends to lower blood sugar
- Anaerobic exercise (like weightlifting) β can actually raise blood sugar temporarily
- Individual variation β every person responds differently
By learning patterns from a person's own CGM history, the AI could give advance warning of an approaching high or low blood sugar episode.
What This Means for Patients
Right now, most CGM devices show you what your blood sugar is β and some show a trend arrow for the next 15β30 minutes. This research points toward a future where your device could predict changes further ahead and with greater accuracy, giving you more time to act β eat a snack, adjust insulin, or rest after exercise.
If you already use a CGM, pairing it with smarter predictive software could make managing diabetes significantly less stressful. At MDS Diabetes, we supply a wide range of CGM sensors and diabetes monitoring supplies to help you stay on top of your glucose data every day β the foundation any AI prediction system depends on.
Bottom Line
A self-improving AI system can predict blood sugar changes more accurately than older computer models by learning from continuous glucose monitor data. While this technology isn't in your pocket yet, it represents a meaningful step toward smarter, more personalised diabetes management tools that could one day help prevent dangerous highs and lows before they happen.