Predictive performance of population pharmacokinetic models for Bayesian forecasting of coagulation factor VIII in hemophilia A
David W. Uster (1), Cecilia Diaz Garcia (2), Elsa Aradom (2), Molly Musarara (2), Pratima Chowdary (2), Sebastian G. Wicha (1)
(1) Dept. of Clinical Pharmacy, Institute of Pharmacy, University of Hamburg, Hamburg, Germany (2) Katharine Dormandy Haemophilia Centre and Thrombosis Unit, Royal Free London NHS Foundation Trust, London, UK
Objectives:
Factor VIII is an essential coagulation glycoprotein and commonly substituted in the treatment of hemophilia A, an x-linked inherited bleeding disorder with a deficiency of the aforementioned protein.
Dosing according to instruction inserts only takes into account the body weight of the patient (prophylaxis) and/or the therapeutic indication of factor VIII treatment (i.e. surgery, heavy bleeding). Regarding the prophylactic treatment the time spent below a plasma concentration of 1 IU/dl should be as short as possible as this leads to fewer bleeds and hemarthroses [1].
Due to high interpatient variability in the pharmacokinetics (PK) of factor VIII, the variable response and its high cost, therapeutic drug monitoring (TDM) facilitated by population PK models has gained interest to individually tailor factor VIII dosing. The objective of the present study was to evaluate the predictive performance of population PK models in different settings.
Methods:
For the evaluation, a clinical dataset was available comprising 33 patients with a total of 208 one stage assay observations. Data below the limit of quantification (5.7% of all observations) was not included.
Four published population PK models were recreated and processed using NONMEM® 7.4. The population PK models included in the comparison were all two-compartment models and varied in the covariate inclusion, the handling of endogenous factor VIII levels and the underlying drug product [2–5]. The models accounted for body weight (one in form of lean body weight [3] ), while only two accounted for the patients age.
The following scenarios were evaluated to predict plasma concentrations from a densely sampled (5 samples, 0.25 - 48 h after dose) dosing interval: (i) forecasting using dosing information only excluding covariate information (age, body weight), (ii) forecasting using dosing information incl. covariate information, (iii) forecasting using dosing information, covariates and a measured factor VIII trough concentration from the previous dosing interval.
Model-predicted vs observed values in the different scenarios were graphically examined. Furthermore, forecasting metrics were calculated and included relative bias (rBias) and relative root mean square error (rRMSE) describing the accuracy and precision of the prediction. Forecasting performance was stratified including (a) all samples taken between 0 h and 48 h after dose or (b) trough samples (time after dose >40 h) only.
Results:
The inclusion of covariates (scenario i vs ii) led to an improvement of the predictive performance in all models except of age in the model by Björkman et al., 2009 [2] and the rBias was reduced from 43% to 38% on average across models (rRMSE 113% to 105%).
Inclusion of a measured trough concentration of the previous dosing interval additionally to dosing information and covariate information (scenario iii vs ii) improved the predictive performance of all models (except the model of Björkman et al., 2009), and the rBias was reduced from 38% to 29% on average (rRMSE 105% to 95%). The rBias and rRMSE increased significantly in the scenario (iii) when only trough concentrations were used for their calculation (case b) with a mean increase from 29% to 78% and 95% to 161%, respectively.
The best overall predictive performance was displayed using the model by Zhang et al., 2017 [5] in both cases (a) and (b). There was only an exception in scenario (a) (iii) for patients receiving any of the four drugs or the B domain deleted product only (Refacto AF® Pfizer, n = 16). In this case the PK profiles were described more accurately by the model of Hazendonk et al., 2016 [4]. The outcome seems plausible as only the Hazendonk model accounted for discrepancies in plasma concentration measurements of B-domain deleted products by the one-stage assay [6].
Conclusions:
The studied population PK models varied substantially in their predictive performance taking different clinically relevant scenarios (i.e. information provided or subpopulation type) into account. The value of the main covariate body weight was confirmed for Factor VIII plasma concentrations in this external evaluation. The trough level prediction was not optimal and needs to be improved as this is most critical in prophylaxis treatment [1]. Further models will be evaluated and suitable models will be implemented into the TDMx software (www.TDMx.eu, [7]).
References:
[1] P. W. COLLINS et al., “Break-through bleeding in relation to predicted factor VIII levels in patients receiving prophylactic treatment for severe hemophilia A,” J. Thromb. Haemost., vol. 7, no. 3, pp. 413–420, Mar. 2009.
[2] S. Björkman, A. Folkesson, and S. Jönsson, “Pharmacokinetics and dose requirements of factor VIII over the age range 3–74 years,” Eur. J. Clin. Pharmacol., vol. 65, no. 10, pp. 989–998, Oct. 2009.
[3] D. Garmann, S. McLeay, A. Shah, P. Vis, M. Maas Enriquez, and B. A. Ploeger, “Population pharmacokinetic characterization of BAY 81-8973, a full-length recombinant factor VIII: lessons learned ? importance of including samples with factor VIII levels below the quantitation limit,” Haemophilia, vol. 23, no. 4, pp. 528–537, Jul. 2017.
[4] H. Hazendonk et al., “A population pharmacokinetic model for perioperative dosing of factor VIII in hemophilia a patients,” Haematologica, vol. 101, no. 10, pp. 1159–1169, Oct. 2016.
[5] Y. Zhang et al., “Population pharmacokinetics of recombinant coagulation factor VIII-SingleChain in patients with severe hemophilia A,” J. Thromb. Haemost., vol. 15, no. 6, pp. 1106–1114, Jun. 2017.
[6] A. R. Hubbard, L. J. Weller, and S. A. Bevan, “A survey of one-stage and chromogenic potencies in therapeutic factor VIII concentrates,” Br. J. Haematol., vol. 117, no. 1, pp. 247–247, Apr. 2002.
[7] S. G. Wicha, M. G. Kees, A. Solms, I. K. Minichmayr, A. Kratzer, and C. Kloft, “TDMx: A novel web-based open-access support tool for optimising antimicrobial dosing regimens in clinical routine,” Int. J. Antimicrob. Agents, vol. 45, no. 4, pp. 442–444, Apr. 2015.