AI model detects hidden diabetes risk by analyzing glucose spikes before symptoms appear

An AI model has shown promise in identifying individuals at high risk of developing type 2 diabetes or prediabetes earlier than traditional methods, even before symptoms emerge. While clinicians commonly use the HbA1c test—a measure of average blood glucose levels over the past two to three months—to diagnose diabetes and prediabetes, this metric has limitations. It cannot reliably predict who is most likely to progress from normal glucose levels to prediabetes or from prediabetes to full diabetes. The new AI model analyzes patterns in glucose spikes following meals, capturing subtle fluctuations that may not be evident in standard HbA1c readings. By examining continuous glucose monitoring data from wearable devices, the model identifies individuals with abnormal glucose responses, indicating early metabolic dysfunction. These patterns often precede changes in HbA1c, offering a window for earlier intervention. Researchers trained the AI on data from thousands of individuals, including those with and without diabetes, to detect early warning signs such as delayed glucose recovery after eating or exaggerated spikes. The model demonstrated high accuracy in predicting future diabetes development, outperforming traditional risk assessment tools. This approach could transform early detection, enabling preventive strategies like lifestyle changes or targeted medical interventions before irreversible damage occurs. Unlike HbA1c, which reflects long-term averages, the AI’s focus on real-time glucose dynamics provides a more sensitive indicator of metabolic health. As wearable glucose monitors become more accessible, this AI-driven method could become a routine part of preventive care, helping to curb the rising global burden of diabetes.