Marquez-Megias, S1; Ramon-Lopez, A1,2; Dokoumetzidis, A3; Mas-Serrano, P1,2,4; Nalda-Molina, R1,2
1: Miguel Hernández University, Department of Engineering, Division of Pharmacy and Pharmaceutics, School of Pharmacy, San Juan de Alicante, Alicante, Spain 2: Alicante Institute for Health and Biomedical Research (ISABIAL-FISABIO Foundation), Alicante, Spain 3: Department of Pharmacy, National and Kapodistrian University of Athens, Athens, Greece. 4: General University Hospital of Alicante, Pharmacy Department, Clinical Pharmacokinetics Unit, Alicante, Spain.
Introduction and objectives:
In the clinical setting, one of the limitations of developing population pharmacokinetic (PopPK) models is the sparseness of the available data, usually limited to trough levels. In such cases, there may not be enough information to develop a PopPK model from scratch and it is necessary to rely on PopPK models from the literature. Using the values of the PK parameters as priors with a given weight, from previous studies not only can stabilize the PK estimation and reduce the interindividual variability (IIV) but also allows for a more accurate fit to new data compared to fixing parameters. However, the PK parameters of the studied population may differ from those used in the literature model (prior population).
The objectives of this study were to assess, through simulations based on a biased reference model, whether the estimation of PK parameters was improved by using a model with informative priors derived from a population with different PK parameters, and to develop a methodology to detect which parameter should include priors and which parameters should not.
Methods:
The reference model used in this study consisted of a one-compartment model with first-order absorption and linear elimination with a proportional residual error model. The reference model parameters were set as CL/F = 5 L/h, V/F= 50 L, Ka=0.5 h-1, with IIV=0.2 in all parameters.
Based on the reference model, three scenarios were stochastically simulated with SIMULX, modifying the CL, V, and both CL and V from the reference model. Each patient had two available concentrations at the steady state, one peak and one trough level.
Two models were developed for each scenario using MONOLIX: one estimating the PK parameters without priors (flat-prior model) and the other including informative priors in all parameters with the Maximum a Posteriori Estimation option (informative prior model). The values of the PK parameters and relative standard error of the reference model were used to define the prior. The accuracy and robustness of the estimates were assessed through a bootstrap with the 500 datasets, evaluating mean and 95% CI from individual replicates.
To compare the flat-prior models and the informative-prior models, the estimated PK parameter values of each model were compared to the “true” parameter values. Forest plots of the mean and 95% CI of the differences between estimated and true parameters obtained from the bootstrap replicates were used for graphical and numerical validation.
The methodology to detect which parameter was biased compared to the reference model (and therefore, which parameter should not include prior), consisted of conducting a sensitivity analysis by estimating the PK parameters with the same model with different weights of the priors. Subsequently, the 95% confidence intervals for each estimation of the PK parameters for each weight were calculated. The criterion to determine the inclusion of the informative prior in a parameter was whether the confidence interval included the value of the parameter from the reference model for all the weights of the prior evaluated. If not, the prior was considered biased for the studied population in that parameter and should be removed from the parameter.
Results:
In all scenarios, including priors in all parameters (even those that were biased), resulted in a better estimation of all parameters when compared to the true parameter values. However, the ability to estimate the typical value of Ka was limited, and the ability to estimate the typical value of V/F, as well as the interindividual variability in ka and V/F, was also compromised. Nevertheless, by using informative priors in all parameters, the estimations for each parameter became more stabilized and closer to the real parameters of different populations.
The sensitivity analysis developed to detect which parameters were biased allowed the adequate detection of biased parameters in the different studied populations in each scenario and helped to decide in which parameters the priors should be kept.
Conclusions: In the studied scenarios, the developed models using informative prior showed better results, even when priors come from a different population. The sensitivity analysis allows for the adequate detection of parameters where informative priors can be eliminated without additional information.
References:
[1] Karlsson MO, Sheiner LB. The importance of modeling interoccasion variability in population pharmacokinetic analyses. J Pharmacokinet Biopharm. 1993;21(6):735-50. DOI: 10.1007/BF01113502.
[2] Chan Kwong AHP, Calvier EAM, Fabre D, Gattacceca F, Khier S. Prior information for population pharmacokinetic and pharmacokinetic/pharmacodynamic analysis: overview and guidance with a focus on the NONMEM PRIOR subroutine. J Pharmacokinet Pharmacodyn. 2020;47(5):431-46. DOI: 10.1007/s10928-020-09695-z
Reference: PAGE 32 (2024) Abstr 11265 [www.page-meeting.org/?abstract=11265]
Poster: Methodology - Estimation Methods