Generating Realistic Albumin Concentrations in Virtual Subjects Across A Spectrum of Renal Function to Account for Variability in Protein Binding Within PBPK Models

Use of physiologically-based pharmacokinetic (PBPK) modelling for extrapolation to organ impairment populations requires successful prediction for physiological changes. For drugs bound to human serum albumin (HSA), prediction of albumin concentrations is crucial to predict population differences in fraction unbound in plasma (fu). In this study, a multi-variable model was developed for prediction of HSA concentrations in renal impairment, using easily accessible variables (BMI, eGFR, age, sex, race and ethnicity) as predictors. An increase of eGFR from 15 to 90 mL/min/1.73m² was predicted to elevate HSA concentration by 0.30-0.32 g/dL regardless of subjects' characteristics. Data from obese patients undergoing mini-gastric bypass surgery was used for external validation (observed BMI from 44.5 to 27.3 kg/m², leading to predicted HSA concentration change of 0.3 versus 0.1-0.3 g/dL), highlighting the model's potential to enhance PBPK simulations for a broader population. Application of the new albumin model for predicting fu in renal impairment was evaluated with the single binding protein model. Consideration of inter-individual variability predicted by the albumin model could explain some variability in the observed fu data between different drugs and studies (54% observed records within 2.5th-97.5th percentile range of prediction). However, overall underprediction of fold-change in fu between healthy and severe renal impairment (45% observed data exceeded 97.5th percentile of prediction) was noted. Although accounting for changes in binding affinity in predictive models of fu remains a challenge, the newly developed albumin model can support generation of realistic virtual subjects to support PBPK predictions of plasma protein binding.