Simultaneous Prediction of Area Under the Curves of Mycophenolic Acid and Its Metabolites and Enterohepatic Recirculation in Kidney Transplant Recipients

Background: Therapeutic drug monitoring of mycophenolic acid (MPA) is limited due to the requirement for intensive pharmacokinetic sampling to assess the area under the curve (AUC). Limited sampling strategies (LSS) offer a practical alternative; however, enterohepatic recirculation (EHR) affects prediction accuracy and precision. This study is the first to develop LSS models capable of simultaneously predicting the AUC of MPA, its metabolites [mycophenolic acid glucuronide (MPAG) and acyl mycophenolic acid glucuronide (Acyl-MPAG)], and MPA EHR in kidney transplant recipients (KTRs). Methods: Intensive pharmacokinetic sampling was conducted in 84 adult KTRs receiving mycophenolate mofetil. MPA AUC_0-12 was calculated, and MPA EHR was determined. During the development of the LSS models, a balanced representation of patients with high and low EHR was ensured. Multiple linear regression was used to develop AUC prediction models for MPA, MPAG, and Acyl-MPAG, as well as an EHR prediction model. The best models were selected based on prediction performance, the highest prediction concordance, and the shortest interval between the first and last samples. Results: Three models for AUC_0-12 prediction were identified, incorporating 4, 5, and 6 concentration timepoints. The LSS model with 6 concentrations demonstrated the best performance, with excellent prediction concordance (100% for MPA and MPAG, and 93% for Acyl-MPAG). The EHR prediction model included 4 concentrations and exhibited an;80% prediction concordance. An online calculator was developed for these models. Conclusions: The developed LSS models simultaneously predict MPA, MPAG, and Acyl-MPAG AUC_0-12 using the same timepoints with high accuracy and precision. MPA EHR can be predicted using 4 concentration timepoints. The inclusion of late concentration timepoints is essential for the high predictive performance of LSS models.