Researchers from Weizmann Institute of Science, Pheno.AI, and NVIDIA have developed GluFormer, a transformer model that can forecast individual glucose levels and other health metrics based on continuous glucose monitoring data.
The model was trained on data from over 10,000 non-diabetic participants, monitoring glucose levels every 15 minutes over 14 days through wearable devices.
The research addresses a growing health challenge, as diabetes currently affects approximately 10% of the world's adults and could impact over 1.3 billion people by 2050. The economic impact of diabetes is projected to reach $2.5 trillion globally by 2030.
GluFormer's capabilities extend beyond glucose prediction to include forecasting other health metrics such as visceral adipose tissue, systolic blood pressure, and apnea-hypopnea index. The model has been validated across 15 different datasets, demonstrating effectiveness for various conditions including prediabetes, type 1 and type 2 diabetes, gestational diabetes, and obesity.