Ronald Kadada Karyaburo1,2, Dr. Mohammed Wulgo Ali1,3, Dr. Islam R. Younis4, Dr. Dan Li4, Dr. Roeland E. Wasmann5
1APT-Africa Fellowship Program, c/o Pharmacometrics Africa NPC, 2Dept. of Pharmacology and Therapeutics, College of Health Sciences, Makerere University, 3Division of Clinical Pharmacology, Stellenbosch University, Cape Town South Africa., 4Quantitative Pharmacology and Pharmacometrics, Merck & Co., Inc., NJ, 5Certara, Princeton, NJ
Introduction: Organ impairment pharmacokinetic (PK) studies typically require a healthy control group. However, enrolling healthy participants in clinical studies raises ethical concerns, such as exposure to potential adverse effects and invasive procedures without medical benefits [1,2]. Simulations from population pharmacokinetic (PopPK) models offer a promising alternative by generating virtual control groups to reduce these risks while maintaining study reliability[3,4]. However, it is unclear how much data or at which stage of drug development a model can reliably predict the PK of healthy control participants. Objectives: •To explore best practices for the development of virtual control populations in organ impairment studies using PopPK modelling approach at various milestones in the drug development spectrum by leveraging data available at the time. Method: The study will include five drugs to optimally explore the methodological design of studies that will generate a virtual control population. The selection of drugs is based on the availability of data and the complexity of the developed model. For each drug, data were categorized into five scenarios representing different stages in drug development. The scenarios include data from: Scenario 1: Single ascending dose (SAD) and multiple ascending dose (MAD) study Scenario 2: SAD, MAD and all other Phase 1 studies Scenario 3: SAD MAD study and the proof of concept (Phase 2) studies Scenario 4: All Phase 1 and Phase 2 studies Scenario 5: All data from Phase 1 to 3 studies A PopPK model was developed and validated for each drug and scenario. Models were selected using goodness-of-fit plots and visual predictive checks. Covariates included based on biological plausibility and their significance (objective function value of -3.84; p-value of 0.05 for 1 degree of freedom). Using these models, we performed simulations in two populations: 1) healthy participants from the organ impairment study, and 2) a population from the National Health and Nutrition Examination Survey (NHANES) database with matched demographic characteristics. The NHANES database was used to generate a larger population with participant characteristics similar to those in the reference study. The area under the curve (AUC) and maximum concentration (Cmax) for both simulated and observed healthy participants were computed using PKNCA R package (Version 0.9.5). Geometric mean ratios (GMRs) and their confidence interval (CI) of AUC and Cmax were calculated for simulated versus observed data across all five scenarios. Additionally, GMRs comparing observed organ impairment participants with both observed and simulated healthy participants were also computed. Modelling and simulation were conducted using NONMEM (Version 7.44 and Perl-speaks-NONMEM (PsN),[5] with post-processing performed in R. Result: In this preliminary analysis, we report the results of the first two drugs. The GMR between the observed and simulated PK parameters were generally close to 1 for all scenarios. Overall, the best predictions were observed with Scenario 4 for both drugs. The AUC / Cmax GMRs (CI) for the respective drugs in each scenario using healthy participants from the organ impairment study were; Scenario 1 [0.871 (0.697–1.12), 1.11 (0.77–1.55) / 1.14 (0.961–1.42), 1.45 (1.28–1.84)], Scenario 2 [0.997 (0.787–1.30), 1.02 (0.702–1.42) / 1.13 (0.927–1.44), 1.42 (1.24–1.79)], Scenario 3 [1.07 (0.476–2.45), 0.991 (0.420–1.960) / 1.01 (0.43-2.37), 0.934 (0.34-2.0)], Scenario 4 [1.04 (0.484–2.26), 1.00 (0.443–1.89) / 1.02 (0.450–2.30), 0.959 (0.367–1.99)], Scenario 5 [1.10 (0.448–2.74), 1.01 (0.439–2.73) / 1.06 (0.434–2.55), 0.963 (0.352–2.08)]. Similarly, for the NHANES database Scenario 1 [0.885 (0.722–1.08), 1.21 (0.817–1.66) / 1.15 (0.808–1.52), 1.57 (1.00–2.26)], Scenario 2 [0.897 (0.700–1.15), 1.20 (0.804–1.166) / 1.11 (0.75–1.52), 1.53 (0.971–2.21)], Scenario 3 [1.08 (0.486–2.35), 1.06 (0.429–2.59) / 1.02 (0.425-1.97), 0.973 (0.332-2.2.24)], Scenario 4 [1.04 (0.496–2.16), 1.07 (0.447–2.52) / 1.03 (0.450–1.91), 0.999 (0.362–2.25)], Scenario 5 [1.11 (0.461–2.61), 1.12 (0.436–2.82) / 1.04 (0.444–1.95), 1.01 (0.343–2.36)]. These predictions were compared to the observed values, with AUC GMRs of 1.01 (0.819–1.23) and 0.987 (0.788–1.23), and Cmax GMRs of 0.973 (0.769–1.36) and 0.973 (0.803–1.38). Generally, there was higher variability in all simulated data than the observed data. However, greater variability was observed in Scenarios 3, 4, and 5 compared to Scenarios 1 and 2. The GMR of the renal impaired category observed / simulated AUC were similar for both data generated from the Reference study and the NHANES database. The GMR ratios and their CI for both drugs were similar for both NHANES simulated participants and the organ-impaired reference study participants. Conclusion: The study contributes to the growing body of knowledge by demonstrating the potential of leveraging existing data to replace healthy control groups in organ impairment studies. The finding highlights that using simulated virtual control participants, incorporating data from phase 2, and 3 trials yields more accurate and predictive outcomes.
[1] Moore KT et al. J Clin Pharmacol. (2022) 62(12), 1465-1467. [2] Walker RL et al. J Med Philos. (2018) 43(1), 83-114. [3] Younis IR et al. J Clin Pharmacol. (2024) 64(6), 713-718. [4] Prybylski JP et al. AAPS J. (2024) 26(4)65. [5] Lindbom L et al. Comput Methods Programs Biomed. (2004) 75(2), 85-94.
Reference: PAGE 33 (2025) Abstr 11493 [www.page-meeting.org/?abstract=11493]
Poster: Methodology - Study Design