Bayesian forecasting of recombinant factor IX-Fc fusion protein concentrate dosing in haemophilia B patients: Comparison of the performance of two population pharmacokinetic models
Sjoerd F. Koopman (1)*, Tine M.H.J. Goedhart (2)*, Laura H. Bukkems (1), Trevor M. Mulders (1), Frank W.G. Leebeek (3), Karin Fijnvandraat (4), Michiel Coppens (5,6), Mary Mathias (7), Peter W. Collins (8), R. Campbell Tait (9), Catherine Bagot (9), Nicola Curry (10), Jeanette Payne (11), Pratima Chowdary (12), Marjon H. Cnossen (2)* & Ron A.A. Mathôt (1)*, for the OPTI-CLOT study group and SYMPHONY consortium. *shared first and last authorship
(1) Hospital Pharmacy-Clinical Pharmacology, Amsterdam University Medical Centers, Amsterdam, The Netherlands; (2) Department of Pediatric Hematology and Oncology, Erasmus MC Sophia Children’s Hospital, University Medical Center Rotterdam; Rotterdam, The Netherlands; (3) Department of Hematology, Erasmus MC, University Medical Center Rotterdam, Rotterdam, The Netherlands; (4) Amsterdam UMC, University of Amsterdam, Emma Children’s Hospital, Pediatric Hematology, Amsterdam, The Netherlands; (5) Vascular Medicine, Amsterdam UMC, University of Amsterdam, Amsterdam, The Netherlands; (6) Amsterdam Cardiovascular Sciences, Pulmonary Hypertension & Thrombosis, Amsterdam, The Netherlands; (7) Haemophilia Comprehensive Care Centre, Great Ormond Street Hospital for Children NHS Foundation Trust, London, United Kingdom; (8) Arthur Bloom Haemophilia Centre, School of Medicine, Cardiff University Hospital, Cardiff, United Kingdom; (9) Department of Haematology, Royal Infirmary, Glasgow, United Kingdom; (10) Oxford Haemophilia and Thrombosis Centre and Oxford NIHR BRC, Nuffield Orthopaedic Hospital, Oxford, United Kingdom; (11) Department of Paediatric Haematology, Sheffield Children’s NHS Foundation Trust, Sheffield, United Kingdom; (12) Katharine Dormandy Haemophilia Centre and Thrombosis Unit, Royal Free London NHS Foundation Trust, London, United Kingdom
Introduction: Recombinant factor IX Fc fusion protein (rFIX-Fc) is an extended half-life (EHL)-clotting factor concentrate administered to haemophilia B patients. For prophylaxis, therapeutic drug monitoring using Bayesian forecasting is recommended[1]. Generally, doses are adjusted to obtain specific factor target activity levels. So far, two population models are available for the description of its pharmacokinetics (PK). Model A is constructed with rich data from registration trials including adults and children ≥12 years[2]. Model B is based on real world data from adults and children ≥2 years (unpublished; manuscript in preparation). Both models differ significantly from each other. Model A comprises three compartments, whereas model B uses two to describe the concentration vs time profiles. Interestingly, the models also give different values for typical PK parameters. For instance, clearance is 42% lower in model B than in model A for similar patients. Furthermore, model A contains more estimates of inter-individual and inter-occasional variability, whereas model B only contains a limited description of this variability.
Objectives: To compare dose recommendations following PK-guided dosing by application of Bayesian forecasting using the population model based on clinical trial data (model A) and the population model constructed with real world data (model B).
Methods: Data was collected from haemophilia B patients (FIX activity level <5 IU/dL) treated with rFIX-Fc in the OPTI-CLOT TARGET study (NTR7523) and United Kingdom (UK)-EHL registry (NCT02938156). FIX measurements for PK assessment were obtained at 15-30 min, 4, 24, 72, 120 and 168 h after infusion. Additional FIX activity levels were sampled during regular visits between 10 days and 18 months after rFIX-Fc treatment initiation. Doses were calculated by Bayesian forecasting. We calculated individual dose – required to yield a FIX activity level of 3 IU/dL one week after administration during steady-state (Dose3%) – based on the following four situations: 1) all available PK measurements, 2) PK profile assessment measurements only, 3) three clinically relevant measurements in PK assessment (peak, trough and random mid FIX activity level) and 4) three random measurements in PK assessment. The four situations were evaluated for all patients and patients <12 years separately. Differences were analysed with the permutation test.
Results: We included 36 patients (median age: 17 years, range 2-71) of whom 13 patients <12 years. Situation (sit.) 1 and 2 included a median of 7 (range 4-12) and 4.5 (range 3-7) measurements, respectively. Population predicted Dose3% when comparing model A to model B, was significantly higher for each situation with respective median values of 3500 vs 1375 IU (p<0.001). For individually predicted Dose3%, no significant difference was found when all available PK measurements were evaluated (sit.1). Median doses were 1500 IU (range 250-4250) and 1500 IU (range 250-2750) for model A and B, respectively. For the three remaining situations, significant differences were found when all patients were considered. Median doses were 1750 vs 1500 IU (p=0.004, sit.2), 1750 vs 1500 IU (p=0.041, sit.3) and 1750 vs 1500 IU (p=0.001, sit.4) for model A and B, respectively. Surprisingly, however, when focusing on children <12 years no significant differences in Dose3% were found. Median doses were 1000 vs 1000 IU (p=0.984, sit.1), 1000 vs 1250 IU (p=1.000, sit.2), 750 vs 1250 IU (p=0.328, sit.3) and 1000 vs 1000 IU (p=0.750, sit.4) for model A and B, respectively.
Conclusion: This study demonstrates that both models a priori recommend different doses when no individual measurements are considered in dose calculations. This difference is explained by a dissimilarity in predicted typical clearance of included patients. When performing maximum a posteriori Bayesian analysis, this difference is reduced when more samples per patient are available. Nevertheless, in a clinical setting when only sparse samples would be available (reflecting sit.2, 3 and 4) the recommended dose would depend on the model used. For children <12 years no differences were detected, which could be due to the limited number of patients. This study exemplifies that Bayesian forecasting combines information from both the population and individual. Hence, in situations of sparse individual measurements, it is necessary to strive for a representative population PK model.
References:
[1] M. V. Ragni, S. E. Croteau, M. Morfini, M. H. Cnossen, and A. Iorio, “Pharmacokinetics and the transition to extended half-life factor concentrates: communication from the SSC of the ISTH,” J. Thromb. Haemost., vol. 16, no. 7, pp. 1437–1441, Jul. 2018, doi: 10.1111/jth.14153.
[2] L. Diao, S. Li, T. Ludden, J. Gobburu, I. Nestorov, and H. Jiang, “Population pharmacokinetic modelling of recombinant factor IX Fc fusion protein (rFIXFc) in patients with haemophilia B,” Clin. Pharmacokinet., vol. 53, no. 5, pp. 467–477, 2014, doi: 10.1007/s40262-013-0129-7.