A Bayesian analysis of published pharmacokinetic data ‐ a ketoconazole example

Purpose In drug‐drug interaction research, an inhibitor/inducer's pharmacokinetic model is usually neither directly available from clinical studies nor published literature, and sometimes published results were inconsistent. A well known CYP3A inhibitor, ketoconazole, was reported to follow either a one‐ or two‐compartment model. In order to establish its optimal PK model from published data, an innovative Bayesian meta‐analysis was proposed. Methods Published sample mean ketoconazole plasma concentrations with standard errors from five papers were described by a three‐level Bayesian model: study‐specific sample mean concentration profiles were modeled at level‐1; study‐specific PK parameters were defined at level‐2; and PK parameter priors were specified at level‐3. Monte Carlo Markov chain (MCMC) estimation procedure was implemented. Optimal PK model was selected through the Bayesian factor (BF). Both data analysis and simulation study were performed. Results Bayesian factor analysis choose a two‐ over a one‐compartment model for ketoconazole(BF= 45.1); PK parameters are Vsys=24±0.7, Vperi=7.3±1.5, CL12=1.2±0.3, ka(solution)=1.8±0.2, Vmax=53.5±4.0, and Km=2.8±0.4. Their relative biases were between 0.3% and 6.5%, and their 90% CI coverage rates were between 81% and 89%. Conclusions This general Bayesian method efficiently summarized an optimal PK model from published literature. Its PK parameter estimates had low biases. (see Table) Clinical Pharmacology & Therapeutics (2005) 77 , P87–P87; doi: 10.1016/j.clpt.2004.12.227PK Parameter Estimates Parameter Estimate SE Relative Bias 90% CI CoverageVsys 24.0 0.7 0.3% 89% Vperi 7.3 1.5 6.5% 81% CL12 1.2 0.3 0.6% 87% ka (solution) 1.8 0.2 1.2% 85% Vmax 53.5 4.0 2.2% 85% Km 2.8 0.4 4.5% 82%