External validation with sparse, adaptive-design data for by S5I8Y6


									                        External validation with sparse, adaptive-design
                        data for evaluating the predictive performance
                        of a population pharmacokinetic model of
               Johan E. Wallin1,2, Martin Bergstrand1, Mats O. Karlsson1, Henryk Wilczek3, Christine E. Staatz1,4
            1. Department of Pharmaceutical Biosciences, Uppsala University, Sweden, 2. PK/PD/TS, Eli Lilly, Erl Wood Windlesham, UK, 3. Division of Transplantation
                              Surgery, Karolinska Institute, Stockholm, Sweden 4. School of Pharmacy, University of Queensland, Brisbane, Australia.

Tacrolimus is a potent immunosuppr-
essant used to prevent and treat organ
rejection in paediatric liver transplantation.
Tacrolimus has a narrow therapeutic
window and displays considerable
between and within-subject pharmaco-
kinetic (PK) variability. The PK of
tacrolimus change markedly in the
immediate post-transplant period. We
have previously developed a population                                Prediction corrected visual predictive checks with the three compared models
PK model of tacrolimus with the intent of
capturing this process. This model has                                                                                                                MPE             RMSE
been used to suggest a revised initial                    Objectives:                                                             Wallin                1.1             5.8
dosing schedule and forms the basis for a                 To evaluate the predictive performance of                               Staatz                2.2             7.9
dose adaptation tool.                                     our population model, in comparison to                                  Sam                   2.1             7.7
                                                          two previously published models (2, 3),
                                                                                                                           Mean prediction error and root mean squared error
To validate the model and compare it to                   using data collected from an independent                         with the three compared models
previously      published   models,     an                group of paediatric liver patients and
independent dataset was used. The                         based on model diagnostics suitable for
nature of this dataset, comprising of                     use with TDM data. Accuracy of early                        Results:
sparse      adaptive-type   TDM       data,               measurements as well as avoiding                            Accuracy and precision expressed as
necessitate some caution in model fit                     overprediction was of special concern.                      MPE and RMSE was better for the
evaluation. Population predictions can                                                                                proposed model compared to the Sam
only be used for data prior to                                                                                        and Staatz models. Graphical diagnostics
individualization,      and      individual                                                                           confirmed the increased predictive
predictions does not serve as an unbiased                 Data on the PK of tacrolimus in the first                   capability with the proposed model.
guide in model structure discrimination.                  two weeks following liver transplantation
                                                          was collected retrospectively from the
                                                          medical records of 12 paediatric patients.
Commonly        used simulation    based                  Population predicted drug concentrations
diagnostics are also unsuitable when                      from the three models were compared to
using adaptive design data, but visual                    measured concentrations using samples
evaluation of the predictive performance                  drawn prior to TDM associated dosage
can be performed with prediction                          adaption.     Individual predicted drug
corrected VPC (pcVPC), where observed                     concentrations based on all data were
and      simulated   observations    are                  compared     to     all  the   measured
normalized based on the population                        concentrations.
prediction (1).

       DV                                                 To evaluate the models’ potential for
                                                          Bayesian forecasting in dose adaptation,
                                                          individual predicted drug concentrations
                                                          based on prior samples were compared to
                                                          measured        concentrations.     Model
                                                          predictive performance was compared by
                                                          calculation of MPE and RMSE. Prediction                             Baysian predictions based on only the previously
                                                          corrected     VPC:s     (pcVPC),     were                           measured concentrations, mimicking Bayesian
                                                          constructed using the PsN software and                              forecasting.
                                                          the Xpose graphical analysis toolpack.
  Population prediction of samples drawn prior                                                                          Simulation based diagnotics was a
  to a posteriori dose individualisation                                                                                valuable aid in determining that the
                                                                                                                        proposed PK model predicted the
                                                                                                                        validation data set reasonably well, and
                                                                                                                        performing better than the previously
                                                                                                                        published models in this early post-
                                                                                                                        transplantation phase.

                                                                                                      1.       M Bergstrand, A.C Hooker, J.E Wallin, M.O Karlsson. Prediction Corrected Visual
                                                                                                      Predictive Checks. ACoP (2009) Abstr F7.
                                                                                                      2.       Sam WJ, Aw M, Quak SH, et al. Population pharmacokinetics of tacrolimus in Asian
                                                                                                      paediatric liver transplant patients. Br J Clin Pharmacol 2000; 50 (6): 531.
                                                                                                      3.       Staatz CE, Taylor PJ, Lynch SV, Willis C, Charles BG, Tett SE. Population
                                                                                                      pharmacokinetics of tacrolimus in children who receive cut-down or full liver transplants.
                                                                                                      Transplantation 2001; 72 (6): 1056.
      Posthoc Bayesian individual predictions of the three compared models representing
      the overall fit to data

To top