Richard Dimelow, Herbert Struemper
GlaxoSmithKline
Objectives:
Belimumab (BENLYSTA) is a monoclonal antibody that binds to the B-lymphocyte stimulator protein and is currently approved for the treatment of active systemic lupus erythematosus (SLE) in adults. To address unmet needs in childhood-onset SLE (cSLE), a study in a paediatric population with active cSLE is planned to investigate the safety, pharmacokinetics and pharmacodynamics following sub-cutaneous (SC) administration (study 200908). It is a regulatory requirement that the study must be prospectively powered to target a 95% confidence interval (CI) within 60% to 140% of the geometric mean estimate for clearance (CL) and volume of distribution (V) with at least 80% power [1]. The objective of this analysis was to confirm that study 200908, with proposed subject number (N=24) and sampling times, was sufficiently powered to meet this condition.
Methods:
A simulation / re-estimation approach was performed to investigate the precision to which CL and V can be estimated. The precision estimation for the PK focused only on data from the mandatory Part A of study 200908 (12 weeks Part A plus 40-week extension phase with additional PK samples) and therefore provide a conservative estimate of the precision. In the simulation scenario 24 subjects between 5 and 17 years will receive 200 mg belimumab SC every week for 12 weeks, with subjects less than 30 kg receiving 200 mg once every two weeks. Pre-dose PK samples will be taken on weeks 1, 2, 4, 8 and 12, followed by an 8-week washout sample. A paediatric population PK model previously developed for IV administration (study BEL114055) was combined with the SC absorption component in adults [2], and the resulting model used to simulate 2000 trial outcomes. Several 1-compartmental models, all with fat-free mass (FFM) as the body size covariate on CL and V, were fitted to each trial simulation. In each case the covariance matrix for parameter precision was used to calculate the 95% CI in CL and V [1]. The probability that the 95% CI was within 60% to 140% of the geometric mean or median estimate (defined as the power) was calculated. The impact of a 25% drop-out rate was investigated by repeating the model-fitting on simulated datasets containing only 18 subjects. The results were benchmarked against a theoretical “best-case” design with rich PK sampling scheme, enabling accurate individual PK parameters to be obtained through non-compartmental analysis (NCA).
Results:
Both CL and V can be estimated with 100% power (CI = 60% to 140%) providing the FFM exponents on CL and V are fixed at their allometric theoretical values 0.75 and 1.0 respectively for all subjects. The power remains high (95% for both CL and V) when measured against the more stringent definition of precision (CI = 80% to 120%). The power (CI = 60% to 140%) drops off significantly when the FFM exponent on CL (power = 36%) and V (power = 22%) are estimated. Applying a prior probability distribution on the FFM exponents (0.75 ± 0.1 for CL, 1.0 ± 0.1 for V) as part of the parameter estimation reverses this drop in performance, recovering 100% power (CI = 60% to 140%) in both the CL and V estimates, although the power is still somewhat below 80% when benchmarked against the higher level of precision (CI = 80% to 120%). For a rich sampling scheme with NCA derived CL and V for each subject, at least 10 subjects per trial would be required to estimate CL and V with 80% power (CI = 60% to 140%). This is a theoretical “best-case” outcome but given the sparse PK sampling requirements for study 200908 (to minimize patient burden) 15 subjects (but not 12) are sufficient to estimate CL and V with a population PK approach to the required level of precision. The study design should therefore be robust against a drop-out rate of at least 25% (N=18).
Conclusions:
- The paediatric study 200908 (N=24, sparse PK sampling) is sufficiently powered to estimate CL and V directly form the data providing the FFM exponent is fixed at 0.75 (on CL) and 1.0 (on V) or providing a prior likelihood about these is included.
- A 25% drop-out rate (N=18) can be accommodated with minimal loss of power.
- By comparison, a rich PK sampling scheme enabling NCA derived PK parameters for each subject would only require 10 subjects to estimate CL and V with 80% power. This is the smallest possible sample size to meet the precision criteria for the paediatric population if patient burden were not of concern.
References:
[1] Wang Y, et al. Clarification on Precision Criteria to Derive Sample Size When Designing Pediatric Pharmacokinetic Studies. J Clin Pharmacol. 2012; 52(10):1601-1606.
[2] Struemper H, et al. Population pharmacokinetic and pharmacodynamic analysis of belimumab administered subcutaneously in healthy volunteers and patients with systemic lupus erythematosus. Clin Pharmacokinet. 2018; 57(6):717-28.
Reference: PAGE 28 (2019) Abstr 8948 [www.page-meeting.org/?abstract=8948]
Poster: Methodology - New Modelling Approaches