AI models use glucose spikes to reveal hidden risks of diabetes before symptoms appear

To diagnose either type 2 diabetes or prediabetics, clinicians usually rely on lab values known as HBA1C. This test captures a person's average blood sugar level over the past few months. However, HBA1C cannot predict who is at the highest risk of progressing from health to prediabetes or from diabetes to full-scale diabetes.

Currently, scientists at Scripps Research have discovered that artificial intelligence can use a combination of other data, including real-time glucose levels from wearable monitors, to provide a more nuanced view of diabetes risk.

The new model explained in Natural Medicineuses continuous glucose monitor (CGM) data along with gut microbiota, diet, physical activity and genetic information. Flag early signs of diabetes risk that standard HBA1C tests may miss.

“We have shown that two people with the same HBA1C score can have very different underlying risk profiles,” said Giorgio Quer, director of artificial intelligence at Scripps Research and assistant professor of digital medicine. “By bringing in more data, we can see who is on and who is on a fast trajectory of diabetes and who is not,” he said, “what glucose spikes take to resolve, what happens overnight with glucose, what food intake is, and even what happens in the intestines.”

“Ultimately, the goal of this work is to better understand what is driving diabetes progression and how to intervene in the clinic faster,” adds Ed Ramos, senior director of digital clinical trials at Scripps Research.

Some degree of fluctuations in blood sugar levels are completely normal, but especially after eating – frequent or exaggerated glucose spikes are a sign that your body is struggling to effectively manage sugar. In healthy people, blood sugar levels usually rise and drop smoothly. However, in people at risk of diabetes, these spikes can be sharper, more frequently, or slower to resolve, even before routine lab tests like HBA1C acquire the problem. New research shows that tracking these daily dynamics could provide a more detailed perspective on human metabolic health and help identify troubles beforehand.

The findings are the results of a multi-year digital research program called Predictions (Progress) of Glycemic Response Research. The study used social media outreach to enroll more than 1,000 people across the United States in fully remote clinical trials. Participants included those who had either diabetes or diagnosed with diabetes, and those who were healthy.

For 10 days, they wore Dexcom G6 CGMs, followed their diet and exercise, and sent blood, saliva and stool samples for testing. Researchers also had access to the participants' electronic health records. This included lab values and diagnosis made by previous practitioners.

“This has been a pioneering effort in the field of remote clinical trials,” says Ramos. “We had to design a study where participants could complete the entire process, from sensor application to biological samples collection and delivery without visiting the clinic. That level of self-directed participation required a completely different kind of infrastructure.”

Using the data, researchers trained AI models to distinguish people with type 2 diabetes from healthy people.

One of the clearest signals of diabetes was the risks that researchers found was the time it took for blood sugar spikes to return to normal. In people with type 2 diabetes, it took more than 100 minutes for blood sugar to drop after spikes, while healthier people returned to baseline faster. The study also found that people with a more diverse gut microbiota and higher activity levels tend to have better glucose controls, while higher resting heart rates are associated with diabetes.

Importantly, the AI model did not only detect risks in people with already elevated HBA1C. When applied to prediabetic individuals, some people have metabolically similarities to diabetic patients, while others resemble healthy people despite the value of similar labs. This level of granularity can help clinicians personalize their treatment. This helps to focus on lifestyle changes and early treatment of patients at the highest risk of disease progression.

Although the current study was a snapshot in time, researchers continue to follow participants to see if model predictions translate to actual disease progression. They also validated the model using a separate set of patient data from Israel, enhancing the possibilities for broader clinical use.

The team envisions a future version of the model being used by clinicians to assess metabolic risks and monitor how daily choices affect diabetes.

“Ultimately, this is about giving people more insight and control,” Quer says. “Diabetes not only will come one day, it's slowly built, so there are tools to detect before and intervene smarter.”