Koichiro Yoneyama, Tomohisa Saito
Chugai Pharmaceutical Co., Ltd., Tokyo, Japan
Introduction: Dose-ranging clinical studies can encounter imbalanced baseline responses among the groups of subjects testing different dose levels of a drug, especially for the cases of no randomization, sequential enrollment, or small number of subjects across the dose levels (e.g., dose-ascending studies conducted in the early-phase clinical development). This inter-dose baseline imbalance may complicate the characterization of the dose- or exposure-response (E-R) relationship simply based on the observed data. Nonlinear mixed-effect (NLME) modeling is expected in theory to overcome this limitation by appropriately considering the inter-individual variability (IIV) in the data analysis [1]. However, no dedicated investigation on the parameter estimation performance in the E-R analysis in the presence of inter-dose baseline imbalance has been reported.
Objectives: To evaluate how the inter-dose baseline imbalance impacts the parameter estimation performance in the E-R analysis by NLME modeling.
Methods: A stochastic simulation and estimation study was performed using SAS version 9.4 and NONMEM version 7.4.3 for the respective processes. First-order conditional estimation with interaction method was used for the maximum likelihood estimation. Immediate response models with a stimulatory or inhibitory maximum effect (Emax) function of a dose-proportional exposure variable were used for simulating the response data. One baseline and 1 post-baseline time points were set for the response observation, and IIV was considered for the baseline response parameter (RBL). Subjects were evenly allocated to 4 groups of different dose levels including placebo in the following 3 scenarios: 1) a subject is randomly allocated to a dose level regardless of the observed baseline response value; or 2) and 3) the higher the observed baseline response value for a subject is, the higher or lower the allocated dose level is (i.e., positively or negatively correlated inter-dose baseline imbalance). Accuracy and precision of the estimation of the population mean of RBL, Emax, the exposure value to achieve half of Emax (EC50), the IIV of RBL, and the residual variability parameter by the true models were evaluated and compared among the 3 allocation scenarios. Alternative models, which remove the IIV of RBL (i.e., naive pooled data [NPD] modeling), condition the estimation of the individual RBLs by the observed baseline response values with residual variability correction [2], or add a nested level of random effect of inter-group variability (IGV) on the IIV of RBL by the $LEVEL functionality without or with the LEVWT option [3], were tested for the estimation for comparison with the true models. Sensitivity analysis was performed by varying several conditions.
Results: Accuracy and precision of the parameter estimation by the true models were comparable among the 3 allocation scenarios. In the presence of inter-dose baseline imbalance, the NPD modeling resulted in less accurate estimation of Emax and EC50, which corresponded to higher objective function values (OFVs; higher OFVs occurred even in the absence of inter-dose baseline imbalance), as compared to the true models. The conditioned RBL estimation or IGV addition also tended to result in less accurate estimation of Emax and EC50 in the presence of inter-dose baseline imbalance, while the OFVs were lower (lower OFVs occurred even in the absence of inter-dose baseline imbalance for the case of conditioned RBL estimation), as compared to the true models. Regardless of the presence or absence of inter-dose baseline imbalance, the NPD modeling or conditioned RBL estimation resulted in less accurate estimation (and less precise estimation for the case of NPD modeling) of the residual variability parameter, as compared to the true models. Sensitivity analysis supported the validity of these findings.
Conclusions: NLME modeling provides stable accuracy and precision of the estimation of Emax model parameters even in the presence of inter-dose baseline imbalance. Careful consideration, not only the OFV comparison, should be made if implementing a modification to the IIV of RBL in relation to the inter-dose baseline imbalance (e.g., what caused the inter-dose baseline imbalance). Further investigations are needed to evaluate the generalizability of these findings from this simple study to more realistically complicated cases (e.g., more IIVs, delayed response models, and discrete response data).
References:
[1] Aarons L, et al. Eur J Pharm Sci 2001;13:115-22.
[2] Dansirikul C, et al. J Pharmacokinet Pharmacodyn 2008;35:269-83.
[3] Bauer RJ. CPT Pharmacometrics Syst Pharmacol 2019;8:538-56.
Reference: PAGE 32 (2024) Abstr 10799 [www.page-meeting.org/?abstract=10799]
Poster: Methodology - Other topics