Mehdi El Hassani, Amélie Marsot
Faculté de Pharmacie, Université de Montréal
Introduction External evaluations are the most stringent methods for evaluating the predictive performance of population pharmacokinetic (PK) models. Our previous work has shown that we were unable to externally validate a previously published tobramycin population PK model due to unacceptable bias and imprecision of population-predicted concentrations. We hypothesize that since the evaluated model was not developed using the same sampling times as in the validation dataset, concentrations at the divergent sampling times could not be well predicted. The aim of this study was to investigate the effects of 8h post-dose sampling times on the performance of PK parameter estimation using models developed with (Crass et al. [1]) and without (Koloskoff et al., not yet published) 8h post-dose sampling times.
Methods Two previously developed two-compartment models for tobramycin were used to simulate 24 scenarios according to different sample sizes, infusion times, and sampling times. Datasets with and without 8h post-dose sampling times were used to fit the simulated data with NONMEM® v7.5. The estimated parameters, as well as the population-predicted concentrations, were compared with the observed values.
Results Population PK parameters relating to the first compartment (i.e., clearance (CL) and central volume of distribution (V1)) estimated with the model by Crass et al. yielded low bias values ranging from -5.5% to 3.5% with no impact of the 8h post-dose sampling time. Estimation of intercompartmental clearance (Q) without 8h post-dose sampling times led to large bias values ranging from 25.3% to 81.5% (0.1 L/h to 0.3 L/h). Estimation of Q with datasets containing 8h post-dose sampling times resulted in more acceptable bias ranging from -37.1% to -2.9% (-0.1 L/h to -0.01 L/h). Estimation of peripheral volume of distribution (V2) resulted in large bias values ranging from -50.2% to 48.9%, but estimation was not affected by 8h post-dose sampling times. Population PK parameters relating to the first compartment (i.e., CL, V1) estimated with the model by Koloskoff et al. yielded low bias values ranging from -5.8% to 11.1%, with no impact of 8h post-dose sampling times. Estimation of Q with 8h post-dose sampling times resulted in high bias at lower sample sizes (n=50 and 100) ranging from -82.4% to 126.4% (-0.2 L/h to 0.4 L/h), but low bias values were observed at a larger sample size (n=500). However, estimation of Q without 8h post-dose sampling times led to lower bias values ranging from -46.8% to 22.2% (-0.1 L/h to 0.1 L/h). Estimation of V2 with a large sample size (n=500) resulted in high bias ranging from -95.3% to 68.4% (-6.4 L to 4.6 L), but no impact of 8h post-dose sampling times was observed. However, bias for this parameter was higher with 8h post-dose sampling times when estimated with smaller sample sizes (n=100 and 50). In general, bias tended to decrease with increasing sample size, except for Q estimated with the model by Crass et al. (without 8h post-dose sampling times) which performed worse with increasing sampling size. Parameter estimation was not affected by the tested infusion times. Population-predicted concentrations for both models performed well with bias values below 1% and imprecision values ranging from 29 to 35%.
Conclusion In conclusion, our study shows that PK parameter estimates relating to the second compartment are biased when fitting concentration data with divergent sampling times (i.e., 8h post-dose) to the evaluated model. Some factors should be considered (e.g., sampling times, sample sizes) prior to externally evaluating a published model, depending on its applications.
References:
[1] Crass, R.L. and M.P. Pai, Optimizing Estimated Glomerular Filtration Rate to Support Adult to Pediatric Pharmacokinetic Bridging Studies in Patients with Cystic Fibrosis. Clin Pharmacokinet, 2019. 58(10): p. 1323-1332.
Reference: PAGE 29 (2021) Abstr 9685 [www.page-meeting.org/?abstract=9685]
Poster: Methodology - Model Evaluation