A real time optimal design for model discrimination and parameter estimation for itraconazole population pharmacokinetics
Duffull SB(1), Waterhouse TH(2), Redman S(1), Eccleston JA(2)
1. School of Pharmacy, University of Queensland, Australia. 2. School of Physical Sciences, University of Queensland, Australia.
Introduction: Itraconazole is used in the treatment of allergic bronchopulmonary aspergillosis in patients with cystic fibrosis. Recent evidence suggests that itraconazole has a better absorption profile when administered as a solution compared to capsule. Itraconazole has an active metabolite hydroxyitraconazole.
Aim: To develop an optimal design for estimation of the population pharmacokinetics for itraconazole and hydroxyitraconazole following administration by solution and capsule.
Methods: The clinical study investigators required the design to have a maximum of 30 patients and a maximum of 4 blood samples taken on each of 2 occasions. Itraconazole is to be administered on two periods, one occasion as the capsule and one occasion as the solution. The time frame for the design was 2 months. Elicitation of prior information from the literature revealed that itraconazole and its metabolite could be adequately described by a 2 compartment model for the parent and a 1 compartment model for the metabolite. It was, however, unclear if the parent displayed linear or non-linear elimination from the central compartment. We developed an optimal population design (using POPT«) for two competing multiple response (parent and metabolite) repeated measures models. It was assumed that capsule and solution would follow the same structural model but have different input parameter values. We optimised the product criterion of the linear and non-linear models using simulated annealing. We assessed the performance of the optimal design using simulation and estimation in NONMEM (ver 5).
Results: The optimal population design consisted of 3 elementary designs for both capsules and solution. Joint sampling windows were provided using Monte Carlo simulation. Due to the constraints on the maximum number of blood samples per patient, it was required that we fix some of the parameters, thereby reducing the dimensionality of the model. These parameters were included within the information matrix to preserve their interaction with other parameters but the corresponding row and column were deleted from the information matrix prior to computation of the determinant in each step in the optimisation algorithm. The design took 7 days to optimise on a P4 2.8 GHz PC. Simulation from the optimal design revealed that the design was able to support the estimation of the expected and the alternative model in 100% of the simulated data sets. We also confirmed that the design was able to correctly discriminate between the two models on 74% of occasions when the correct model was linear and 100% of occasions when the correct model was non-linear. We are waiting on the results from the actual study.
Conclusion: We developed a real-time design for a complex population pharmacokinetic model which had multiple response types and where there was uncertainty in the structural pharmacokinetic model. The use of optimal design techniques made an otherwise difficult to construct sparse design manageable in a clinical setting.