Development of a sufficient design for parameter estimation of fluconazole population pharmacokinetics in HIV patients

Posted on by Juliana Roos

Introduction: Fluconazole (FCZ) is an antifungal drug commonly used in HIV patients. A clinician suspected low concentrations of FCZ in HIV patients, which could be due to saturable input (low bioavailability) or nonlinear disposition of the drug. In contrast, the literature suggests that FCZ concentrations are increased in HIV patients. A study found in the literature was undertaken to address this question [1]. However, this study protocol is characterized by a very intensive sampling schedule with a total of 1440 samples from 20 HIV patients.

Aim: 1) To obtain a sufficient design capable of estimating PK parameters for fluconazole in patients with HIV; 2) To test the performance of the optimal design.

Methods: The prior population pharmacokinetic model arose from a healthy study population (denoted S1) [2] and was a 1 compartment model with first-order input and elimination and
lag-time. Three possible structural models were considered to may explain the altered pharmacokinetics in HIV patients (1) Linear absorption with linear elimination (M11 and M12), (2) nonlinear absorption with linear elimination (M21 and M22) and (3) linear absorption with nonlinear elimination (M31). For structural models 1 and 2 parameter sets that provided a high (M11 and M21) and low (M12 and M22) exposure were considered, representing 40% increase and decrease in FCZ concentrations. The design was optimised using POPT (Version 2.0; www.winpopt.com) with the actual HIV study (denoted S2) used to provide the constraints on the design space [1]. The authors were blinded to the results of S2. The S2 protocol included one IV and 3 oral doses with 18 blood samples taken from each dose level, totalling 72 samples per patient with 20 patients in the study.  Assessment of the design was by the following processes:

  1. The full design from S2 (denoted D3) with 72 samples per patient was used to assess the loss of information associated with more parsimonious but optimised designs (D1 & D2).
  2. Empirical standard errors (SEs) from simulation-estimation using NONMEM were obtained and compared to the SEs predicted by POPT.
  3. A power analysis was undertaken by simulation to assess the ability of the design to discriminate between the competing models.
  4. The performance of the best model developed from the optimised design was compared to best the model developed from the actual data that arose from S2.

Results: The optimised design contained only 1 IV and 2 oral doses without significant loss of information. A final balanced design (D1) was obtained with 4 samples from
the IV dose and 4 samples from each oral dose (a total of 12 samples per patient). An unbalanced design (D2) with 10 samples per patient was unstable and not considered further.  The results of evaluation of the design are shown:

  1. The precision of the parameter estimates from D1 were only minimally different from those from D3.
  2. The predicted SEs from POPT were close to those obtained by
  3. simulation-estimation under NONMEM.
  4. The power to discriminate between models was high (generally > 80%), except when fitting the nonlinear disposition model (M3) to data arising from the linear (M1) model where the power was 50%. However since M3 can collapse to M1 then this finding is not unexpected.
  5. The model developed from the optimised design (D2) showed very close predictive performance compared to the model developed from the actual data (S2); the bias and precision were very similar with bias being less than the assay detection limit (0.2 mg/L).

Conclusions: The optimised design (D1) presented in this study illustrates that cost-effective studies that provide acceptable parameter precision are achievable even when there is considerable uncertainty in the model and parameter space.

References:

  1. Tett S, Moore S, Ray J. Pharmacokinetics and bioavailability of fluconazole in two groups of males with human immunodeficiency virus (HIV) infection compared with those in a group of males without HIV infection. Antimicrob Agents Chemother 1995;39(8):1835-41.
  2. Gross AS, McLachlan AJ, Minns I, Beal JB, Tett SE. Simultaneous administration of a cocktail of markers to measure renal drug elimination pathways: absence of a pharmacokinetic interaction between fluconazole and sinistrin, p-aminohippuric acid and pindolol. Br J Clin Pharmacol2001;51(6):547-55.
2007