Purpose: A simulation study was conducted to assess the use of a Phase 1 population PK model to estimate Bayesian parameters in Phase 2/3 patients. Specifically, the bias and precision of the Bayesian parameter estimates was examined after applying a Phase 1 model to sparse patient data with varying estimates of mean parameter values and interindividual variability (IIV).
Methods: A 1-CMT model with 1st-order absorption and elimination was used to simulate full-profile steady-state data in Phase 1 subjects for a drug with an elimination half-life (t1/2) of 20 hr. Twelve patient populations were simulated (and replicated 100 times) using combinations of population mean t1/2 values (ranging from 20-40 hr), increased magnitudes of IIV for CL and V (10-60% higher than Phase 1), and sparse sampling schemes (2-6 samples/ patient). The model was fit to the Phase 1 data using NONMEM’s FOCE method with interaction, and applied to each patient dataset to obtain Bayesian PK parameters using the Phase 1 estimates as priors. Bias and precision (median PE% and |PE|%, respectively) were assessed for CL, V and Cmax.
Results: Reasonable Bayesian parameter estimates (PE% ± 15 and |PE|% < 30) were obtained for patient populations with a t1/2 within 25% and magnitudes of IIV for CL and V within 35% of priors. Although bias and imprecision increased as patient PK diverged from priors (PE% ± 45%), this was minimized when there were at least 4 strategic samples per patient.
Conclusions: Reasonably unbiased and precise Bayesian PK estimates can be obtained from sparse patient data using Phase 1 model estimates as priors under the conditions described above. The impact of PK differences from the priors can be mitigated by attending to the quality and quantity of the sparse patient data.
American Association of Pharmaceutical Scientists (AAPS), Baltimore, Maryland, November 2004
By S Van Wart, Luann Phillips, and Thaddeus H. Grasela.