Global Modeling and Simulation Scientist Astellas Pharma, United States
Objectives: Organ impairment studies require enrolling matched healthy subjects based on key covariates, like age and sex, to that of organ impaired subjects. Leveraging the use of virtual populations (VPs) to replace the matched healthy control (HC) in the organ impairment studies is gaining attention. Currently, there are no best practices available to create VPs. An overarching aim of this work is to implement and evaluate different approaches for creation of VPs for renal impairment study of ASPXXX.
Methods: 6 datasets were prepared. Datasets 1 and 4 were created by applying filters based on the inclusion and exclusion criteria from the renal impairment study protocol, utilizing data from historic phase-1 studies (viz, hepatic impairment, drug-drug interactions, bioavailability, and food on ASPXXX in healthy subjects) and NHANES database, respectively. Datasets 2 and 5 consisted of 1000 replicates of 8 subjects sampled with replacement from datasets 1 and 4, respectively. Datasets 3 and 6 were sampled from datasets 1 and 4, respectively, using propensity score matching (PSM) to the HC group from renal impairment study for the covariates age, weight, sex, Eastern Cooperative Oncology Group score, and baseline albumin, creatine phosphokinase and aspartate aminotransferase. Using a validated POPPK model, ASPXXX concentrations were simulated following 20 mg single oral dose in the HC from the renal impairment study (reference group), and in the VP datasets created. Geometric mean (GM) and 95% CI of exposure metrics for Cmax (ng/mL) and AUC0-last (ng.h/mL) were computed for each group. Additionally, the same metrics were computed for dose normalized (to 20 mg) concentrations from datasets 1 and 3. Each set of the exposure metrics was then compared to the reference group.
Results: A total of 106 and 1717 subjects were in the historic and NHANES datasets, respectively. GM (5th & 95th % CI) of exposure metrics shown below. A. For PSM datasets based on • observed concentrations from dataset-3: Cmax 14.8 (11.8 & 18.5); AUC0-last 885.6 (742.2 & 1056.7) • simulated concentrations from dataset-3: Cmax 12.9 (10.6 & 15.8); AUC0-last 836.8 (722.7 & 968.9) • simulated concentrations from dataset-6: Cmax 12.6 (10.8 & 14.8); AUC0-last 759.4 (698.7 & 1195.4) B. For reference group based on • observed concentrations: Cmax 13.8 (9.9 & 19.2); AUC0-last 927.4 (694.7 & 1238.1) • simulated concentrations: Cmax 13.2 (9.2 & 18.8); AUC0-last 913.9 (698.7 & 1195.4)
Conclusions: Only the PSM datasets yielded exposures concordant with the reference group. Exposure metrics for historic and NHANES datasets that created using similar inclusion-exclusion criteria and sampling with replacement were discordant compared to observed results. Further analysis revealed variability in covariate distribution as the primary cause of discordance. Hence, study design aspects are pivotal to avoid selection bias in VP creation, warranting careful consideration on a case-by-case basis.
Citations: Younis, I.R., Wang, F. and Othman, A.A. (2024), Feasibility of Using Population Pharmacokinetics-Based Virtual Control Groups in Organ Impairment Studies. J Clin Pharm. https://doi.org/10.1002/jcph.2410
