Racym Berrah1, Marc Labriffe1,2,3, Pierre Marquet1,2,3, Caroline Monchaud1,2,3, Jean-Baptiste Woillard1,2,3
1Pharmacology & Transplantation, INSERM U1248, Limoges University, 2Department of Pharmacology, Toxicology and Pharmacovigilance, CHU Limoges, 3Fédération Hospitalo-Universitaire Survival Optimization in Organ Transplantation (FHU SUPORT)
Background: Pharmacokinetic (PK) models combined with Bayesian Maximum A Posteriori Estimation (MAPBE) are widely applied in Model-Informed Precision Dosing (MIPD) to estimate drug exposure. This approach is frequently used for tacrolimus, an immunosuppressant with a narrow therapeutic range that requires precise dosing to balance efficacy and toxicity. Tacrolimus exposure is best quantified by the area under the concentration-time curve (AUC)(1), which MAPBE estimates using prior population models and individual patient data. However, the impact of residual error assumptions on individual parameter estimation and AUC prediction remains insufficiently explored, particularly when PK models are adopted from the literature rather than developed from local data. Since residual error accounts for unexplained variability in drug concentration measurements, its handling may significantly influence dose individualization strategies in clinical practice. Aims: This study investigates how different residual error assumptions affect MAPBE-based AUC estimation and individual PK parameter predictions. Specifically, we assess whether reducing or eliminating residual error alters dose predictions and model reliability. Methods: Pharmacokinetic (PK) data from two clinical studies on tacrolimus were analyzed. MAPBE was conducted using a previously published model(2) developed by our team from one of these studies, implemented with mrgsolve and mapbayR. Three different residual error models were compared: the original model, which included an 11% proportional residual error, a reduced-error model with a 5% proportional residual error, and a zero-error model. The impact of these assumptions was assessed at both individual and population levels. To evaluate individual goodness-of-fit, selected case studies were examined, while population-level analysis included ANOVA, Tukey’s HSD tests, and calculations of absolute bias and root mean square error (RMSE) to quantify differences in AUC estimates across models. Results: At the individual level, the zero-error model occasionally produced extreme parameter estimates and poor fits, particularly in atypical cases where individual concentration profiles deviated from the population trend, despite closely matching observed values—unlike the other error models. In contrast, both the original and reduced-error models yielded more stable and clinically plausible parameter estimates. At the population level, AUC estimates varied significantly depending on the residual error model used, even after excluding patients with aberrant estimations (ANOVA, p = 0.00218). The zero-error model resulted in the highest bias (8.53 vs. 5.45 for the original model), suggesting that disregarding residual error can lead to overfitting and unrealistic predictions. However, despite these differences in bias, absolute bias and RMSE values remained comparable across models, indicating that the overall predictive performance of MAPBE was robust to changes in residual error assumptions at the population level. Conclusions: Residual error assumptions significantly influence AUC calculation using MAPBE for MIPD. While MAPBE remains a robust method for individualized dosing, assuming zero residual error can lead to unrealistic parameter estimates and model overfitting. In practice, careful visual inspection of individual fits is essential to ensure reliable PK modeling and avoid biased dose adjustments. Future research should explore the impact of different residual error structures in larger and more diverse patient populations. Keywords Tacrolimus, Residual error, Kidney transplant patients, MIPD, AUC , MAP-BE
1. Brunet M, Van Gelder T, Åsberg A, Haufroid V, Hesselink DA, Langman L, et al. Therapeutic Drug Monitoring of Tacrolimus-Personalized Therapy: Second Consensus Report. Therapeutic Drug Monitoring. juin 2019;41(3):261-307. 2. Woillard JB, de Winter BCM, Kamar N, Marquet P, Rostaing L, Rousseau A. Population pharmacokinetic model and Bayesian estimator for two tacrolimus formulations–twice daily Prograf and once daily Advagraf. Br J Clin Pharmacol. mars 2011;71(3):391-402.
Reference: PAGE 33 (2025) Abstr 11344 [www.page-meeting.org/?abstract=11344]
Poster: Methodology - Estimation Methods