External evaluation of tacrolimus population pharmacokinetic models in adult lung transplant patients: How to enhance the predictive ability of the model?

Objective: Tacrolimus is the cornerstone of current immunosuppressive strategies after lung transplantation. However, its narrow therapeutic range and considerable pharmacokinetic variability pose challenges for individualized treatment. Several tacrolimus population pharmacokinetic (popPK) models have been developed for precision dosing in adult lung transplant patients. However, their applicability across different clinical settings remains uncertain. The aim of this study was to evaluate the external predictability of these models and identify influential factors.

Methods: Published models were systematically retrieved and assessed based on an external dataset of 39 patients (1240 tacrolimus trough concentrations) using three approaches: (1) prediction-based diagnosis using dosing records and patient characteristics; (2) simulation-based diagnosis, with prediction- and variability-corrected visual predictive checks (pvcVPC) and normalized prediction distribution error tests (NPDE); and (3) Bayesian forecasting using one to four observations for posterior predictions. We also investigated the impact of model structure and covariates on predictability.

Results: The predictive performance of six published models was externally evaluated, but none demonstrated satisfactory accuracy in prediction- and simulation-based diagnosis. Bayesian forecasting yielded satisfactory results with only one prior observation and optimal predictive performance with 2-3 priors for all included models. The structural model parameterized on plasma tacrolimus concentration outperformed others. Significant correlations were observed between prediction-error and daily tacrolimus dose, postoperative day, and voriconazole co-administration.

Conclusions: The overall predictive performance of all published models was unsatisfactory, making direct extrapolation inappropriate. However, Bayesian forecasting significantly improves predictive performance. Utilizing plasma tacrolimus concentration for parameter estimation can improve the predictive ability of tacrolimus popPK models.