Prediction tacrolimus blood levels based on the Bayesian method in adult kidney transplant patients

The use of tacrolimus is complicated by its narrow therapeutic index and wide intra- and interpatient variability. We have previously described a tacrolimus population pharmacokinetics model obtained in an adult kidney transplant cohort. The aims of the present study were (1) to validate that model using an external dataset and (2) to evaluate the prediction using a Bayesian method. Data were retrospectively collected from 34 adult patients receiving kidney transplantation. Trough blood concentrations of tacrolimus were predicted using the empirical Bayesian method with sparse samples obtained during the previous week. The system performance was evaluated by the mean prediction error (ME), mean absolute prediction error (MAE). and root mean square error (RMSE). Patients were administrated oral or intravenous tacrolimus as part of a triple immunosuppressive regimen with mycophenolate mofetil and corticosteroids. Subsequent doses were adjusted on the basis of clinical evidence of efficacy and toxicity and by routine therapeutic drug monitoring. In our previous model, clearance increased with post transplantation days and with prednisone dosage. Concentrations predicted by the population mean pharmacokinetic parameter values match well with observed concentrations during oral therapy. Bayesian prediction using trough concentrations obtained after 21 days of treatment significantly decreased ME, MAE, and RMSE compared with predictions from data including this period. After 21 days of treatment, there was an insignificant bias ME (0.22 ± 2.59 ng/ml), a reasonable precision MAE (1.97 ± 1.69 ng/ml) and RMSE (1.28 ± 0.58 ng/ml). The present study demonstrates the suitability of the Bayesian method for the prediction of trough blood concentrations of tacrolimus using only few samples in adult kidney transplantation recipients.