411-P: Extended Generalized Metabolic Fluxes (GMF) Based Digital Twins for Predicting CKD

Introduction: CKD is a prominent complication in diabetes patients. Our regulatory-approved digital twin model, HealthVector Diabetes (HVD), leverages generalized metabolic fluxes (GMF) to predict CKD within a 3-year timeframe using common biochemical and physiological parameters. HVD achieved an accuracy of 0.86 in our clinical validation studies. In this study, we aim to explore the expanded use case of HVD when incorporating specialized input parameters related to vascular dysfunction for the prediction of CKD.

Methods: A preliminary investigation involved collecting follow-up records from a previous T2DM clinical trial. Eligible patients had to be CKD-negative at baseline according to eGFR and ACR criteria and have either eGFR or ACR follow-up readings within 3 years from baseline. A total of 20 input parameters, including specialized ones (carotid intima-media thickness (CIMT), logarithmic reactive hyperemia index (lnRHI), pulse wave velocity (PWV), oxidized low density lipoprotein (ox-LDL), haptoglobin, ferritin, reactive oxygen metabolites (ROM), biological antioxidant potential (BAP), high sensitivity C-reactive protein (hs-CRP)), were gathered at baseline to calculate the GMF for predicting future CKD within 3 years.

Results: We identified 193 eligible patients based on our inclusion and exclusion criteria, with 22 developing CKD within 3 years. Extending HVD for prediction resulted in an AUC-ROC of 0.90.

Conclusion: The extension of our regulatory-approved software as a medical device digital twin model, HealthVector Diabetes continues to perform well. It accurately predicts CKD within 3 years with 90% accuracy using specialized inputs assessing vascular dysfunction.