Individuals with early prediabetes often present with slight elevations or even normal glucose level. However, their insulin profiles are often dysregulated with either hyper- or hypoinsulinemia, suggesting insulin as a more sensitive biomarker for prediabetes detection. We developed a human digital twin that can predict an individual’s insulin time course and glycaemic status from an oral glucose tolerance test (OGTT) time-course. The model incorporates homeostatic processes such as insulin secretion and glucose uptake in adipose and muscles. The model was fed with individualized data such as age, BMI and OGTT glucose time-course to predict insulin profiles. Overlay of clinical and simulated glucose and insulin profiles can be seen in Figure 1. Insulin time-course, area-under-the-curve (AUC) and Matsuda Index were accurately predicted for three Chinese males, one healthy (subject 1), one exhibiting insulin insufficiency (subject 2) and one insulin resistance (subject 3). Fold-change of predicted to actual insulin AUC and Matsuda Index were all within 1.5 times. We demonstrated high accuracy of our model in predicting difficult-to-obtain insulin data using easily obtained glucose data. We envision more validation and use of our model to predict insulin-related metrics in different ethnicities and glycaemic status.
K. Hor Cheng: None.
