Population Pharmacokinetic Modelling and Bayesian Estimation of Tacrolimus Exposure: Is this Clinically Useful for Dosage Prediction Yet?

Abstract

This review summarises the available data on the population pharmacokinetics of tacrolimus and use of Maximum A Posteriori (MAP) Bayesian estimation to predict tacrolimus exposure and subsequent drug dosage requirements in solid organ transplant recipients. A literature search was conducted which identified 56 studies that assessed the population pharmacokinetics of tacrolimus based on non-linear mixed effects modelling and 14 studies that assessed the predictive performance of MAP Bayesian estimation of tacrolimus area under the plasma concentration-time curve (AUC) from time zero to the end of the dosing interval. Studies were most commonly undertaken in adult kidney transplant recipients and investigated the immediate-release formulation. The pharmacokinetics of tacrolimus were described using one- and two-compartment disposition models with first-order elimination in 61 and 39 % of population pharmacokinetic studies, respectively. Variability in tacrolimus whole blood apparent clearance amongst transplant recipients was most commonly related to cytochrome P450 (CYP) 3A5 genotype (rs776746), patient haematocrit, patient weight, post-operative day and hepatic function (aspartate aminotransferase). Bias, as calculated using estimation of the mean predictive error (MPE) or mean percentage predictive error (MPPE) associated with prediction of the tacrolimus AUC, ranged from -15 to 9.95 %. Imprecision, as calculated through estimation of the root mean squared error (RMSE) or mean absolute prediction error (MAPE), was generally much poorer overall, ranging from 0.81 to 40. r ² values ranged from 0.27 to 0.99 %. Of the Bayesian forecasting strategies that used two or more tacrolimus concentrations, 71 % showed bias of 10 % or less; however, only 39 % showed imprecision of 10 % or less. The combination of sampling times at 0, 1 and 3 h post-dose consistently showed bias and imprecision values of less than 15 %. No studies to date have examined how closely MAP Bayesian dosage predictions of tacrolimus actually achieve target AUC by comparing dosage prediction from one occasion with a future measured AUC. Further research involving larger prospective studies including more diverse transplant groups and the extended-release formulation of tacrolimus is needed. Several questions require further examination, including the following. Do Bayesian forecasting methods currently use the most appropriate population pharmacokinetic models and optimal sampling times for dosage prediction? Does Bayesian forecasting perform well when applied to make dosage predictions on a subsequent occasion? How can Bayesian forecasting be simplified for use in the clinical setting? And, are patient outcomes improved with dosage prediction based on Bayesian forecasting compared with trough concentration monitoring?