PAGE. Abstracts of the Annual Meeting of the Population Approach Group in Europe.
PAGE 27 (2018) Abstr 8451 [www.page-meeting.org/?abstract=8451]
Poster: Drug/Disease modelling - Other topics
F. Jonsson (1), F. Mercier (2), N.H. Prins (1), C. Schmitt (2), S. Retout (2)
(1) qPharmetra, Andover MA, (2) Roche Pharma Research and Early Development, Pharmaceutical Sciences, Clinical Pharmacology, Roche Innovation Center, Basel, Switzerland
Objectives: Emicizumab has recently been approved in the US for routine prophylaxis to prevent or reduce the frequency of bleeding episodes in patients with hemophilia A with factor VIII (FVIII) inhibitors. Emicizumab is a recombinant, humanized, bispecific, immunoglobulin G4 (IgG4) monoclonal antibody that binds with moderate affinity to activated factor IX (FIXa) and factor X (FX) and has co-factor activity that substitutes for activated FVIII. Therefore it is capable of promoting the activation of FX by FIXa. Emicizumab activates downstream hemostasis at the site of bleeding in hemophilia A patients who have hypofunctional levels or entirely lack FVIII . Up to 30% of patients with hemophilia A develop neutralizing alloantibodies against FVIII (inhibitors). In patients with hemophilia A, hemostasis can be restored irrespective of the presence of FVIII inhibitors, as emicizumab shares no sequence homology with FVIII.
Our main objective is to assess and quantify the relationship between exposure to emicizumab and reduction in the annualized bleeding rate (ABR) among hemophilia A patients. In addition, we want to perform simulations exploring the effect on bleeding count at different doses in order to support the intended dose regimen
Methods: Bleeding event data were pooled from 301 hemophilia A patients participating in four clinical studies [2-6], including one non-intervention study where no emicizumab was given . Emicizumab exposure was predicted using a previously developed population pharmacokinetic model. Bleeding event data were transformed into daily bleed frequencies and analyzed as count data [7,8] using a PKPD model implemented in NONMEM version 7.3 (ICON Development Solutions), with a small run-in METHOD=SAEM to obtain a solid starting position for the subsequent main estimation run using MCMC Bayesian. Models with Poisson, generalized Poisson, and negative binomial (NB) distributions were fit to the data and assessed. Several forms of exposure-response relationships were tested: linear, all or nothing, power and Emax models. After selection of an appropriate structural model, the following covariates were tested on all exposure-response model parameters: adjunctive hemophilic treatment (episodic/prophylactic), FVIII inhibitor status (yes/no), FIX and FX concentration at baseline, baseline ABR, as well as demographic covariates. A stepwise procedure was implemented to screen the covariates. Candidate covariate models were then compared by Objective Function Value (OFV), Akaike Information Criterion (AIC), and precision of parameter estimates. The final model was subjected to a Visual Predictive Check (VPC) and subsequently used to simulate the progression of ABR over time. The simulation results were summarized in the form of change from baseline and fraction of bleed free patients.
Results: The generalized Poisson distribution (characterized by two parameters λ and ω) provided the best description of the data, although low differences were seen with the NB distribution. The best fit model included an Emax relationship for the effect of emicizumab concentrations on the parameter λ, which described the mean daily bleed frequency. The bleed frequency was estimated to be mildly over-dispersed between patients, as indicated by the dispersion factor (ω) estimate. The mean values (%Relative Standard Error) for λ, ω, maximum treatment effect, and EC50 were 0.0356 (8.4%), 0.0244 (19%), 0.979 (1.1%), 2.72 ug/mL (20%), respectively. The estimated Emax corresponds to 97.9% reduction of bleed frequency. The inter-individual exponential variability of λ was 108%. No significant covariates were found. The VPC showed good agreement with the data.
The model was then used to simulate bleeding count over 1 year under different treatment regimens for a large Phase 3 trial with 300 patients. The observed daily mean concentrations at the intended therapeutic dose of 1.5 mg/kg were around 50 µg/ml. The simulated mean bleeding count at 50 µg/ml was 2, which corresponds to a 91% reduction compared to no emicizumab prophylaxis.
Conclusions: Occurrence of bleeding events was described well by a generalized Poisson model, and showed important reduction of the endpoint with increasing emicizumab concentrations following an Emax relationship. The suggested therapeutic dose of 1.5 mg/kg is predicted to provide a 91% mean reduction of ABR compared to no emicizumab prophylaxis, which is close to the estimated maximum effect.