AI Model Detects Hidden Diabetes Risk by Reading Glucose Spikes

However, this test cannot predict who is most at risk of progressing from a healthy state to pre-diabetes, or from pre-diabetes to diabetes. While some fluctuations in blood sugar are normal—particularly after eating—frequent or exaggerated glucose spikes may indicate an inability to regulate sugar effectively. In at-risk individuals, these spikes may become more pronounced or resolve more slowly, often before traditional tests like HbA1c detect any abnormalities. Researchers have now developed a new solution that uses wearable sensors and artificial intelligence (AI) to detect early metabolic changes and flag individuals with a heightened risk of developing diabetes.

The new model, developed by researchers at Scripps Research (La Jolla, CA, USA), uses continuous glucose monitor (CGM) data alongside gut microbiome, diet, physical activity and genetic information to flag early signs of diabetes risk missed by standard HbA1c tests. The researchers used data from a multi-year, digital research program called the PRediction Of Glycemic RESponse Study, or PROGRESS, to train the AI model to distinguish people with type 2 diabetes from healthy individuals. The study enrolled more than 1,000 remote participants across the U.S. using social media outreach. The trial included individuals with pre-diabetes, type 2 diabetes, and healthy glucose regulation. For ten days, participants wore Dexcom G6 CGMs, tracked their food intake and physical activity, and submitted blood, saliva, and stool samples. Researchers also reviewed participants’ electronic health records. The study showed that tracking day-to-day dynamics provides a much more detailed view of a person’s metabolic health, and might help identify trouble earlier.

The AI model developed in this study examined patterns such as how long glucose spikes took to normalize and overnight glucose behavior to distinguish individuals on a faster path to diabetes. To validate the model, researchers trained it to differentiate between individuals with type 2 diabetes and those without the disease. One of the clearest indicators of diabetes risk was the time taken for blood sugar to return to normal after a spike—diabetic individuals often required over 100 minutes. In contrast, healthier individuals returned to baseline more quickly. The study also found that people with a more diverse gut microbiome and higher activity level tended to have better glucose control, while a higher resting heart rate was linked to diabetes. The results, published in Nature Medicine, show that the model revealed that even among people with similar HbA1c scores, some showed high metabolic risk while others resembled healthy individuals. This AI-powered approach could help tailor early interventions, guide lifestyle changes, and eventually be adopted for personal use in CGM-based monitoring. Researchers are continuing to follow participants to assess long-term predictive accuracy and expand clinical utility.

“Ultimately, this is about giving people more insight and control. Diabetes doesn’t just appear one day—it builds slowly, and we now have the tools to detect it earlier and intervene smarter,” said co-senior author Giorgio Quer, the director of artificial intelligence and assistant professor of Digital Medicine at Scripps Research.

Related Links:
Scripps Research