RESUMO
Accounting for variability in plasma protein binding of drugs is an essential input to physiologically-based pharmacokinetic (PBPK) models of special populations. Prediction of fraction unbound in plasma (fu) in such populations typically considers changes in plasma protein concentration while assuming that the binding affinity remains unchanged. A good correlation between predicted vs observed fu data reported for various drugs in a given special population is often used as a justification for such predictive methods. However, none of these analyses evaluated the prediction of the fold-change in fu in special populations relative to the reference population. This would be a more appropriate assessment of the predictivity, analogous to drug-drug interactions. In this study, predictive performance of the single protein binding model was assessed by predicting fu for alpha-1-acid glycoprotein and albumin bound drugs in hepatic impairment, renal impairment, paediatric, elderly, patients with inflammatory disease, and in different ethnic groups for a dataset of >200 drugs. For albumin models, the concordance correlation coefficients for predicted fu were >0.90 for 16 out of 17 populations with sub-groups, indicating strong agreement between predicted and observed values. In contrast, concordance correlation coefficients for predicted fold-change in fu for the same dataset were <0.38 for all populations and sub-groups. Trends were similar for alpha-1-acid glycoprotein models. Accordingly, the predictions of fu solely based on changes in protein concentrations in plasma cannot explain the observed values in some special populations. We recommend further consideration of the impact of changes in special populations to endogenous substances that competitively bind to plasma proteins, and changes in albumin structure due to posttranslational modifications. PBPK models of special populations for highly bound drugs should preferably use measured fu data to ensure reliable prediction of drug exposure or compare predicted unbound drug exposure between populations knowing that these will not be sensitive to changes in fu.
RESUMO
BACKGROUND: Routine therapeutic drug monitoring (TDM) relies heavily on measuring trough drug concentrations. Trough concentrations are affected not only by drug bioavailability and clearance, but also by various patient and disease factors and the volume of distribution. This often makes interpreting differences in drug exposure from trough data challenging. This study aimed to combine the advantages of top-down analysis of therapeutic drug monitoring data with bottom-up physiologically-based pharmacokinetic (PBPK) modeling to investigate the effect of declining renal function in chronic kidney disease (CKD) on the nonrenal intrinsic metabolic clearance ( CLint ) of tacrolimus as a case example. METHODS: Data on biochemistry, demographics, and kidney function, along with 1167 tacrolimus trough concentrations for 40 renal transplant patients, were collected from the Salford Royal Hospital's database. A reduced PBPK model was developed to estimate CLint for each patient. Personalized unbound fractions, blood-to-plasma ratios, and drug affinities for various tissues were used as priors to estimate the apparent volume of distribution. Kidney function based on the estimated glomerular filtration rate ( eGFR ) was assessed as a covariate for CLint using the stochastic approximation of expectation and maximization method. RESULTS: At baseline, the median (interquartile range) eGFR was 45 (34.5-55.5) mL/min/1.73 m 2 . A significant but weak correlation was observed between tacrolimus CLint and eGFR (r = 0.2, P < 0.001). The CLint declined gradually (up to 36%) with CKD progression. Tacrolimus CLint did not differ significantly between stable and failing transplant patients. CONCLUSIONS: Kidney function deterioration in CKD can affect nonrenal CLint for drugs that undergo extensive hepatic metabolism, such as tacrolimus, with critical implications in clinical practice. This study demonstrates the advantages of combining prior system information (via PBPK) to investigate covariate effects in sparse real-world datasets.
