(W-085) The Influence of Systematic and Technical Errors on Population Pharmacokinetic / Pharmacodynamic Model Parameters: Nonlinear Mixed-Effect Approach
Graduate Assistant University of Florida Orlando, Florida, United States
Disclosure(s):
Parsshava Mehta, PharmD: No financial relationships to disclose
Objectives: It is a common yet optimistic assumption that pharmacokinetic (PK) and pharmacodynamic (PD) studies are error-free when employing a nonlinear mixed-effect approach. However, errors can arise at any phase of clinical and drug development studies, potentially influencing the residual unexplained variability (RUV). Previous research highlighted that neglecting these errors could lead to biased PK estimates [1]. We aim 1) to assess the impact of manufacturing(ME), dosing time(DE), sampling time(SE), and technical(TE) errors on the accuracy and precision of PK/PD parameter estimation; 2) to explore the differences in error propagation between sequential and simultaneous fitting approaches.
Methods: We utilized a two-compartment PK model with first-order absorption and elimination, linked to an indirect response PD model inhibiting the rate of production, for a hypothetical biomarker. The study was simulated in-silico, perturbing PK (pPK) and PD (pPD) systems independently. Individual parameters were log-normally distributed. A combined error model was used for PK and a proportional error model for PD. The best 9-time points were selected using an A-optimal design to cover both PK and PD profiles comprehensively. Model fitting was performed using both methods and relative bias (rBias) and relative root mean square errors (rRMSE) from the true parameter values were calculated.
Scenarios:
1. pPK/pPD: Simulate and perturb 20 PK/PD datasets of 500 subjects each (independent PK and PD errors). 2. L2: Simulate and perturb 20 PK/PD datasets of 500 subjects each (dependent PK/PD errors).
Results: ME showed minimal deviation in both pPK and pPD scenarios with the highest impact of 32% on ωIC50. %. DE resulted in a 122% rBias in the additive PK RUV for the pPK run, with no similar increase in the perturbed PD (pPD) run. TE (PK model misfit) inflated the rBias in tvKa by 146%. Disproportionate relationship in rBias/rRMSE between population and variability magnitude values has been observed (e.g., rBias-tvVc =11.14%, rBIas-BSV_Vc = -70%). When all the above errors are combined (CE) population estimates of PD parameters were not significantly impacted (rBias < 50%). For the combined pPK and pPD runs L2 remained low (< = 6%) in CE runs, suggesting a limited effect on L2. However, single system perturbations showed higher L2 with an average of 21.45% in pPK and 15.38% in pPD. Sequential runs inflated the RUV more than BSV, while simultaneous runs had a greater impact on BSV than RUV.
Conclusions: Our findings underscore the importance of accurately fitting the compartmental PK model to minimize the impact of increased uncertainty on PD parameters. For projects focused on dose optimization, employing a simultaneous analysis approach is advantageous and will be supported in follow up work. Additionally, models exhibiting significant increases in RUV might not be reliable for model-informed drug development when variability is important in decision-making.
