Quantitative Systems Pharmacology Approach to the Treatment of CKD Metabolic Bone Disorder (CKD-MBD) Using Deep Learning

Adam E. Gawęda, Michael E. Brier, Eleanor DeLand Lederer · Journal of the American Society of Nephrology · 2020

Background: CKD-MBD is a common comorbidity that leads to serious skeletal and cardiovascular complications. Using a Systems Biology model of CKD-MBD and Artificial Intelligence (AI) -guided precision dosing approach, we tested the hypothesis that this approach can effectively achieve recommendations for Ca, P, and PTH by balancing the administration of vitamin D and a calcimimetic when faced with varying adherance to phosphate binder dosing. Methods: Using a Quantitative Systems Pharmacology (QSP) approach to model the disease trajectory, we trained a Deep Neural Network AI-agent to adjust doses of a P binder, vitamin D, and a calcimimetic to drive P, Ca, and PTH to recommended targets. We evaluated the agent through treatment simulation in a cohort of 100 virtual patients (defined by dietary P and sensitivity of the Ca receptor) under 3 experimental conditions: 100%, 50%, and 0% adherence to P binder prescription. Using model derived doses of vitamin D and calcimimetic, we analyzed the effect of P binder adherence on achieving the recommended Ca, P, and PTH target ranges. Drug doses were determined by simultaneously maximizing the percent in range of Ca, P, and PTH while minimizing the changes in bone Ca efflux and vascular Ca influx. Simulations were performed over 18 months. Results: Results are shown in Table 1. Conclusions: Using a QSP model of CKD-MBD, we trained an AI agent to guide precision dosing of P binder, vitamin D, and calcimimetic. We validated the agent under three simulated scenarios of P binder adherence. Simulation results show that control of intestinal P absorption is paramount in treatment of CKD-MBD. Failure to control P level severely limits ability to control vascular tissue calcification even when Ca and PTH are controlled pharmacologically. Funding: Veterans Affairs Support

Read the paper · More papers on PaperTik