Effect of pharmacokinetic model misspecification on antibiotic probability of target attainment predicted by Monte...

The first aim of this study was to compare the predictability of efficacy by Monte Carlo simulation between a true one-compartment model and a true two-compartment model for doripenem. The second aim was to explore how we can identify the usefulness of a one-compartment model when the pharmacokinetic/pharmacodynamic (PK/PD) indices between three misspecified one-compartment models and a true two-compartment model are compared.

The reported two-compartment model parameters of two doripenem studies and a vancomycin study were used to generate 200 virtual concentration-time profiles for each study. Sparse and dense sampling designs were selected to build the one- and two-compartment models, respectively. The probability of target attainment (PTA) for the PK/PD indices were compared between the one- and two-compartment models of the same drug, applying the clinical breakpoint distribution of minimum inhibitory concentrations (MICs).

The simulated concentration-time profiles reproduced the original data well. In addition, PTAs were similar between the one- and two-compartment models when infusion time and MIC were the same in the doripenem studies. For vancomycin simulations, the maximum difference was 65.9% between a misspecified one-compartment model and the true two-compartment model.

When a misspecified one-compartment model was established with sparse sampling data, the PTA was significantly different from that of the two-compartment model. Thus, a useful PK model must be verified through diagnostic plots and visual predictive checks and the range of sampling time should be sufficient to explain the PK of a drug.