A new study recently published in Nature suggests that the standard blood sugar test relied on by doctors today to screen for prediabetes may be misleading for many patients, either overlooking those at high risk or causing unnecessary worry for those at low risk.
The study, conducted by researchers at Johns Hopkins Carey Business School, Johns Hopkins School of Medicine, University of California Irvine, and MedStar Health, used a machine learning model trained on real-world data from 10 hospitals and over 300 care locations to analyze the electronic health records of over 14,000 patients. The findings reveal that a person’s risk of developing diabetes within a year depends on much more than just their blood sugar level.
“It is very common for medical professionals to rely solely on blood sugar levels to diagnose prediabetes, but that measurement alone does not fully capture a patient’s progression risk,” said Ritu Agarwal, author of the study and co-director of the Center for Digital Health and Artificial Intelligence at the Johns Hopkins Carey Business School. “Our research points out the importance of using a more holistic approach to identify patients who are likely to progress to diabetes within one year.”
Currently, doctors rely heavily on a test called HbA1c, which measures average blood sugar over three months. With this in mind, the researchers leveraged AI to look at a patient’s entire health picture, including weight, BMI, blood pressure, cholesterol levels, liver health, and lifestyle habits like exercise and smoking. This AI-driven approach predicts a patient's risk for diabetes far more accurately.
The study used a precise data approach, which discarded irrelevant information often found in medical records, such as missing follow-up visits or patients who were already being treated, and relied on complete patient data.
“AI technology has many applications in healthcare beyond prediction, but what makes AI especially valuable to patients is its ability to sort through vast amounts of data to determine which factors are most relevant for individualized care,” said Nawar Shara, co-author of the study and director of the AI CoLab, a joint initiative at MedStar Health and Georgetown University.
The AI model achieved an accuracy rate of 81.6% of diabetes predictions, which outperforms the 74.18 percent accuracy rate seen when relying solely on traditional blood sugar levels. Beyond overall accuracy, the AI model proved far more precise at correctly identifying high-risk patients; it identified 76.47% of patients in the top 10% risk group, compared to only 62.65% identified by HbA1c levels alone.
The model’s ability to predict risk remained consistent across patient data between 2018 and 2022, suggesting the tool is robust and could be ready for clinical use.
”AI has the potential to change the way we fight diabetes,” said Gordon Gao, an author of the study and co-director of the Center for Digital Health and Artificial Intelligence. “This technology can accurately pinpoint who is truly in danger and who isn't, allowing for more effective treatment."
“AI has the potential to change the way we fight diabetes,” said Gordon Gao, an author of the study and co-director of the Center for Digital Health and Artificial Intelligence. “This technology can accurately pinpoint who is truly in danger and who isn't, allowing for more effective treatment.”
Nearly one-third of patients whose blood sugar levels were in the low-risk prediabetes range were actually identified by the AI model as being at medium or high risk for developing diabetes within a year, and those in the high-risk group saw a "steep decline" in their health, with many progressing to diabetes within just a few months. On the other hand, over 40% of patients with higher prediabetes blood sugar levels were found to be at a lower risk than their tests suggested.
“Prediabetes is much more complex than a lab result suggests,” said Gao. “We need to move away from this one-size-fits-all approach and start looking at the individual patient’s health holistically if we want to truly prevent diabetes."
For the millions of people living with prediabetes, the condition is a "window of opportunity,” according to the researchers.
This discovery points to a significant gap in current medical guidelines. By moving away from a stand-alone blood sugar screening to a personalized risk score, doctors can ensure that high-risk patients receive immediate and individualized help, such as structured weight-loss programs or specific medications, while low-risk patients can avoid the stress and cost of unnecessary treatments.
“Knowing the true, personalized risk of diabetes can empower patients and physicians to take the right steps at the right time to delay or even prevent type 2 diabetes from ever starting,” said Agarwal.
