Abstract
Background/Objective: Physiologically based pharmacokinetic (PBPK) modeling is a powerful tool for predicting pharmacokinetics (PK) to support drug development and precision medicine. However, it has not been established for non-renal clearance pathways in patients with end-stage renal disease (ESRD), a population that bears heavy medication burden and is thereby at high risk for drug-drug-disease interactions (DDDIs). Furthermore, the pronounced inter-individual variability in PK observed in ESRD patients highlights the urgent need for individualized PBPK models. Methods: In this study, we developed a PBPK population model for ESRD patients, incorporating functional changes in key drug-metabolizing enzymes and transporters (DMETs), including CYP3A4, OATP1B1/3, P-gp, and BCRP. The model was initially constructed using the recalibrated demographic and physiological parameters of ESRD patients. Then, we used five well-validated substrates (midazolam, dabigatran etexilate, pitavastatin, rosuvastatin, and atorvastatin) and their corresponding PK profiles from ESRD patients taking a microdose cocktail regimen to simultaneously estimate the abundance of all these DMETs. Lastly, machine learning was employed to identify potential factors influencing individual clearance. Results: Our study suggested a significant reduction in hepatic OATP1B1/3 (75%) and intestinal P-gp abundance (34%) in ESRD patients. Ileum BCRP abundance was estimated to increase by 100%, while change in hepatic CYP3A4 abundance is minimal. Notably, simulations of drug combinations revealed potential DDDI risks that were not observed in healthy volunteers. Machine learning further identified Clostridium XVIII and Escherichia genus abundances as significant factors influencing dabigatran clearance. For rosuvastatin, aspartate aminotransferase, total bilirubin, Bacteroides, and Megamonas genus abundances were key influencers. No significant factors were identified for midazolam, pitavastatin, or atorvastatin. Conclusions: Our study proposes a feasible strategy for individualized PK prediction by integrating PBPK modeling with machine learning to support the development and precise use of the aforementioned DMET substrates in ESRD patients.