(T-103) A Fisher Information Matrix-based Analysis to Determine Precision of Estimated PK Parameters for Reduced Sample Sizes in Pediatric Studies in Rare Diseases

Corey Bishop – Johnson & Johnson; Edwin Lam – Johnson & Johnson; Jocelyn Leu – Johnson & Johnson; Sophia Liva – Johnson & Johnson; Mahesh Samtani – Johnson & Johnson; Wangda Zhou – Johnson & Johnson

Objectives: The objectives of this study were to validate data from a pediatric population with an adult and pediatric (AP) model (1-compartment model with first order absorption and fixed allometric exponents) in order to substantiate the precision of the estimated PK parameters (i.e. KA, V, & CL) in the scenario of a reduced sample size. The objective of this analysis was to justify the reduction in sample size from 30 to 12 patients and explore the impact on the precision of the estimated PK parameters.

Methods: The Fisher Information Matrix (FIM)-based optimization process follows design (D)-optimality principles to maximize the information matrix, or maximize the variance of the scores. It is assumed that the structural PK model is known and that the estimate and distribution of the parameters are reliable. A PopED database was created using the create.poped.database() and to evaluate the proposed designs (number of participants), the PopED::evaluate_design() function was used. The evaluate_design() function calculates the FIM, the objective function value (OFV(FIM)), and relative standard error (%RSE). Based on the OFV(FIM) and %RSE, decisions regarding the various designs of interest; namely, the number of participants and sampling schedules, can be made. In this report, OFV(FIM) is used interchangeably with OFV when referring to the PopED output. A previously developed population PK model was implemented within the PopED package (version 0.6.0) and used to calculate the FIM and %RSE for various designs of interest.

Results: The % RSE of the PK parameters decreased slowly with each increasing number of participants included. With 12, 15, 27, and 30 participants all % RSEs for KA, CL, and V are below 15%, 14%, 10%, and 10%, respectively. Thus, decreasing the number of participants from 30 to 12 participants did not substantially lower the precision. These results above indicate that 3 serum concentration samples at the specified times and windows (0.1 (0-14) days, 4±2 weeks, and 12±2 weeks) are sufficient to estimate key PK parameters with acceptable precision. Additionally, given that the Wang et al.1 method of NCA-based estimation of sample sizes using a power analysis results in 12 subjects with a %CV of 47. This suggests that the above finding is reasonable.

Conclusions: Because we were able to validate the data from the pediatric population using the AP model, we are able to use the model of interest for our FIM analysis. According to the various exercises, the FIM analysis demonstrates with 3 serum samples at the planned times and windows, we are able to estimate key PK parameters after lowering the number of subjects to 12 subjects. In conclusion, we were able to 1) validate data from the pediatric population with the AP model and 2) substantiate the precision of the estimated PK parameters (i.e. KA, V, & CL) in the scenario of a reduced sample size.

Citations: [1] Wang et al., Clarification on precision criteria to derive sample size when designing pediatric pharmacokinetic studies. J. Clin. Pharmacol. 52(10):1601-6.

Keywords: Fisher Information Matrix, R, Sample size estimation