A digital twin model incorporating generalized metabolic fluxes to predict chronic kidney disease in type 2 diabetes mellitus

One of the biggest complication in diabetes patients is chronic kidney disease (CKD), making it a tremendous burden on the countrys public healthcare system. We have developed HealthVector Diabetes (HVD), a digital twin model that leverages generalized metabolic fluxes (GMF) to proficiently predict the onset of CKD and facilitate its early detection. Our HVD GMF model utilized commonly available clinical and physiological biomarkers as inputs for identification and prediction of CKD. We employed four diverse multi-ethnic cohorts (n=7072): one Singaporean cohort (EVAS, n=289) and one North American cohort (NHANES, n=1044) for baseline CKD identification, and two multi-center Singaporean cohorts (CDMD, n=2119 and SDR, n=3627) for 3-year CKD prediction. We developed one identification model and two prediction models (with complete or incomplete parameters). The identification model demonstrated strong performance with an AUC ranging from 0.80 to 0.82. For prediction, with incomplete parameters, we achieved an AUC of 0.75, while the complete parameter model achieved an improved AUC of 0.86. Our model also effectively stratifies patients into low, moderate, and high-risk categories, with the high-risk category having the highest proportion (53.3-62.9%) of patients with CKD. Our method also reveals metabolic health profile differences among patient subgroups at baseline, indicating that patient subgroups who develop future CKD exhibit more deteriorated profiles compared to future non-CKD patients. Furthermore, we also show that GMF-based clustering reveals distinct metabolic profile differences that act as drivers for CKD progression, and the distance between different patient clusters can be used to map patient health trajectories. Our HVD GMF digital twin model has the ability to identify patients with baseline CKD and predict future CKD within a 3-year time frame. Furthermore, our approach enables risk stratification, sub-grouping and clustering based on metabolic health profiles, positioning our model as a valuable clinical application tool for healthcare practitioners.