Clinical Pharmacology Modeling and Simulation Scientist GlaxoSmithKline, Japan
Disclosure(s):
Misako Takenaka, PhD: No financial relationships to disclose
Objectives: Characterization of pediatric PK in drug development using popPK can be challenging due to limited information from pediatric studies, in addition that there is lack of consensus on the preferable analysis method [1]. This case study compares two potential popPK approaches: 1) Pooled analysis (PA) of pediatric and adult PK, including covariates to explain PK differences and 2) Bayesian analysis (BA) of pediatric PK with prior information from adult parameters. Bias and precision of estimated PK parameters are presented for different scenarios to determine in what cases these approaches may be optimal.
Methods: A 2-compartment popPK model for an existing GSK compound [2] was used to simulate an adult dataset and various potential scenarios in the pediatric population. Simulated dataset reflects clinical program. The adult dataset includes 550 adults (intensive sampling in FTIH, sparse sampling in Ph2 and Ph3), and the pediatric dataset includes 50 pediatrics (6 samples each). The following two potential clinical scenarios [3] were simulated: 1) PK differences between adults and pediatrics explained by an allometric effect of body weight (BW) on CL and V, 2) as 1) plus age related differences in SC bioavailability and Ka. For both approaches the popPK analysis explored the impact of BW and age on CL, V, Ka, and bioavailability. Model selection was based on likelihood ratio test in PA or Pareto smoothed importance sampling leave-one-out cross-validation in BA [4], as well as parameter uncertainty and plausibility. Analyses were conducted by NONMEM 7.5 using FOCE-I in PA and MCMC (NUTS) in BA. Sensitivity to sample size and sampling design in the pediatric study was investigated. 200 trials were repeated for each scenario and sensitivity analysis. Bias and precision of the population parameter estimates and the Cmax and AUC derived from the individual parameters were calculated.
Results: [Precision] In both scenarios, the 95% CI of CL and V were within 80% to 120% of the estimated values > 90% of trials in both approaches. PA estimated Ka with similar precision as CL and V, while the 95% credible interval of Ka in BA was 60% to 200% of the posterior mean. [Bias] In scenario 1, PA and BA approaches estimated CL and V for the pediatric population with comparable accuracy (Mean absolute percentage error (MAPE) < 10%). In scenario 2, CL and V estimates in PA had more bias than in BA (MAPE for CL: 3.1 - 30% (PA) and 4.6 - 12% (BA), V: 7.6 - 38% (PA) and 4.9 - 23% (BA)), however the bias in individual Cmax and AUCinf estimates was similar with the 2 approaches (MAPE: 15% for Cmax and 10% for AUCinf).
Conclusions: This work demonstrated that the estimated bias in the PK parameter values for the pediatric population can be affected by the analysis method. In this case study, the difference in bias due to the choice of popPK approach was larger when the underlying PK difference between adults and pediatrics was a function of both age and BW.
Citations: [1] Knebel W et al., J Clin Pharmacol. 2013;53:505–516. [2] Assessment report EMA/CHMP/672504/2015 rev. 1 [3] Temrikar Z et al., Paediatr Drugs. 2020; 22:199–216. [4] Johnston CK et al., CPT Pharmacometrics Syst Pharmacol. 2024;13:192-207.
