Associate Director, Clinical Pharmacology Arcus Biosciences Brisbane, California, United States
Objectives: The objective of this analysis is to demonstrate a general simulation-based methodology to assess sample size and power for clinical studies with exposure metric based end points.
Methods: A two-compartment model following intravenous administration with moderate between subject variability on pharmacokinetic parameters was used for pharmacokinetic (PK) simulations. An example drug development scenario of infusion time change from 1-hour to 0.5-hour was used. PK profiles of reference and test populations were simulated following single dose and steady-state using 1-hour and 0.5-hour infusion times, respectively. Non-compartmental analysis was conducted to derive test and reference population-based exposure metrics, including maximum concentration [Cmax] and average concentration [Cavg] from individual predictions (IPRED). Random sampling was conducted from the test and reference populations to represent a parallel design PK evaluation trial with various sample sizes (n=6, 12, 20, 30, 40 or 50). The geometric mean ratios (GMR) of test to reference Cmax and Cavg were calculated for each trial. The process was repeated 1000 times to generate virtual clinical trials for each sample size scenario. Power to detect a nominal 20% difference between test and reference population was calculated as percentage of trials with GMR within 0.8-1.2. In addition, the entire power calculation process was repeated to evaluate scenarios with an assumed true difference of 25% lower exposure in test compared to reference population.
Results: Sample size of ≥6 can achieve >80% probability of correctly concluding that Cmax and Cavg for 0.5-h co-admin are within ±20% of those for 1-h infusion when no true differences between test and reference populations are assumed. Sample size of ≥40 patients would provide >80% probability of correctly concluding that Cmax and Cavg for 0.5-h co-admin are outside ±20% of those for 1-h infusion when the true difference is 25% between test and reference population.
Conclusions: Clinical trial simulation using population PK models can provide an assessment of sample size and power for studies with PK exposure metric driven endpoints. It is more advantageous than traditional power calculation as it is more intuitive, adaptable to different study designs, and incorporates known PK variability related to intrinsic and extrinsic factors as well as differences in dosing. The framework presented can be generally applied in various scenarios in drug development with drug specific models of varying complexities with simple simulation workflows.
