- About every 1 in 9 adults globally is living with diabetes, and over 90% of those cases are type 2 diabetes.
- Type 2 diabetes can be hard to diagnose for several reasons.
- A study discussed an AI model capable of identifying people at a high risk of both diabetes and death from cardiovascular disease more effectively than the HbA1c test.
- Another recent study found small molecules in the blood may help doctors predict a person’s future type 2 diabetes risk beyond traditional risk factors.
Because symptoms take a long time to develop, or may not show themselves at all, it can be very hard for doctors to diagnose type 2 diabetes.
“Type 2 diabetes develops slowly — by the time of diagnosis, adverse changes to the heart, kidneys, or blood vessels may have already begun,” Jun Li, MD, PhD, an assistant professor of medicine and associate epidemiologist in the Department of Medicine at Mass General Brigham, told Medical News Today.
“Current risk evaluation tools rely largely on risk factors such as age, body weight, family history, and blood sugar levels. Although helpful, these measures do not capture the underlying biological changes that lead to diabetes, and many people who eventually develop the disease are not flagged as high risk early enough.”
Because of these issues, researchers are focusing on finding new ways of diagnosing type 2 diabetes earlier.
The study proposes that the AI model is more effective than the standard HbA1c test, which measures a person’s average blood sugar levels over 3 months.
Additionally, Li is both the first and co-corresponding author of the study recently published in
The first study revolves around an AI platform called GluFormer, a generative foundation model for continuous glucose monitoring (CGM). The AI platform was data trained through self-supervised learning, using more than 10 million glucose measurements from almost 11,000 adults, most of whom did not have diabetes.
In a study with 580 participants, scientists reported GluFormer identified participants at an increased risk for diabetes and death from cardiovascular disease more effectively than the HbA1c test.
Over a median follow-up period of 11 years, researchers found that 66% of participants considered in the highest-risk category by GluFormer went on to develop diabetes, while only 7% of those in the lowest-risk category developed the condition.
When looking at cardiovascular death risk, 69% of those categorized at high risk died from heart-related conditions, while no deaths occurred among participants in the lowest-risk group.
Additionally, researchers reported that the GluFormer platform was able to pick out participants with prediabetes who were most likely to experience significant increases in their HbA1c readings over a 2-year period better than baseline HbA1c and common CGM metrics.
MNT had the opportunity to speak with David Cutler, MD, a board certified family medicine physician at Providence Saint John’s Health Center in Santa Monica, CA — who was not involved in this study — about this new AI model.
“It is welcome news about a new AI application of blood sugar measurements which is a better predictor than our traditional tools of both future diabetes and cardiovascular deaths,” Cutler commented.
“Traditionally, we have simply used serial hemoglobin A1c measurements to see who was at risk of developing diabetes. We would combine this with measurements of cholesterol, weight, kidney function, coronary calcium, age, and smoking history to estimate future risks of cardiovascular events.”
“This new AI application of continuous glucose measurements seems to give more accurate estimations of future risk than our current tools.”
“The question remains whether better risk prediction with the GluFormer model of CGM data will lead to better outcomes,” he continued.
“Once patients are told they are more likely to develop diabetes or have a heart attack, will they take the medication, change their behaviors, and undergo procedures which will treat diabetes and prevent heart attacks and strokes?”
“Will the cost for the CGM devices and data interpretation be acceptable to payors, whether that be individuals or insurance companies? And what additional steps will need to be taken to transition healthcare providers from using traditional risk assessment tools to a new, better, but less familiar technology?”
“In the past it has often taken a decade or more once a new test, treatment or technology has shown proven benefit for it to be incorporated into routine practice,” Cutler added.
“This has certainly been the case for blood pressure, cholesterol and diabetes control measures. Next steps should include not only assurance of the benefits of the GluFormer technology, but the process for implementation which will yield significant beneficial results.”
In the second study, Li and her team identified metabolites that may help predict a person’s future risk of developing type 2 diabetes.
“Metabolites are small molecules found in our blood that are produced during our bodies’ daily activities, such as natural biological processes, to maintain function, when we eat, store or use energy, and respond to everyday activities like exercise,” Li explained.
“They are chemical ‘footprints’ that can reflect how well the body’s metabolism is working at a given moment.”
“In this study, we found that certain metabolites begin to change years before type 2 diabetes develops,” she continued.
“These changes reflect early dysfunctions in terms of how the body process sugar, fats, energy, and nutrients, before blood sugar levels become high enough for a person to be diagnosed as having type 2 diabetes. We also found that a combination of metabolites in the blood can better predict future risk of type 2 diabetes.”
At the study’s conclusion, researchers found that diet and lifestyle factors may have a stronger impact on metabolites correlated with type 2 diabetes than on metabolites not linked to the condition.
They also found that the metabolites associated with type 2 diabetes were also genetically linked to clinical traits and tissue types that are connected to the condition.
“This genetic evidence allows researchers to move beyond identifying risk