Can We Predict Individual Concentrations of Tacrolimus After Liver Transplantation? Application and Tweaking of a Published Population Pharmacokinetic Model in Clinical Practice

Background: Various population pharmacokinetic models have been developed to describe the pharmacokinetics of tacrolimus in adult liver transplantation. However, their extrapolated predictive performance remains unclear in clinical practice. The purpose of this study was to predict concentrations using a selected literature model and to improve these predictions by tweaking the model with a subset of the target population.Methods: A literature review was conducted to select an adequate population pharmacokinetic model (L). Pharmacokinetic data from therapeutic drug monitoring of tacrolimus in liver-transplanted adults were retrospectively collected. A subset of these data (70%) was exploited to tweak the L-model using the PRIOR subroutine of the NONMEM software, with 2 strategies to weight the prior information: full informative (F) and optimized (O). An external evaluation was performed on the remaining data; bias and imprecision were evaluated for predictions a priori and Bayesian forecasting.Results: Seventy-nine patients (851 concentrations) were enrolled in the study. The predictive performance of L-model was insufficient for a priori predictions, whereas it was acceptable with Bayesian forecasting, from the third prediction (ie, with $\ge$2 previously observed concentrations), corresponding to 1 week after transplantation. Overall, the tweaked models showed a better predictive ability than the L-model. The bias of a priori predictions was --41% with the literature model versus --28.5% and --8.73% with tweaked F and O models, respectively. The imprecision was 45.4% with the literature model versus 38.0% and 39.2% with tweaked F and O models, respectively. For Bayesian predictions, whatever the forecasting state, the tweaked models tend to obtain better results.Conclusions: A pharmacokinetic model can be used, and to improve the predictive performance, tweaking the literature model with the PRIOR approach allows to obtain better predictions.