Measurement of factor VIII pharmacokinetics in routine clinical practice

Authors

  • S. BJÖRKMAN,

    1. Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden
    Search for more papers by this author
  • P. COLLINS,

    1. School of Medicine, Cardiff University, and University Hospital of Wales, Cardiff, UK
    Search for more papers by this author
  • for the Project on Factor VIII/Factor IX Pharmacokinetics of the Factor VIII/Factor IX Scientific and Standardization Committee of the ISTH

    Search for more papers by this author
    • The members of the Project on Factor VIII/Factor IX Pharmacokinetics are: P. Collins (chair), S. Björkman, V. Blanchette, K. Fischer, M. Morfini, and E. Tuddenham.


Sven Björkman, Department of Pharmaceutical Biosciences, Box 591, Uppsala University, Uppsala S-751 24, Sweden.
Tel.: 46 18 471 4995; fax: 46 18 471 4003.
E-mail: sven.bjorkman@fambio.uu.se

Introduction

Prophylaxis for hemophilia A is effective but expensive. Knowledge of a patient’s pharmacokinetic (PK) response to infusion of factor VIII is likely to be useful in clinical management [1]. However, the pharmacokinetics of FVIII cannot be predicted from patient characteristics such as age and body weight (BW), but must be determined empirically [2,3]. According to guidelines for PK studies on new FVIII concentrates [4], 10 or 11 blood samples should be taken over a period of 32–48 h. However, for dose tailoring during ongoing treatment, useful PK parameters can be estimated from only a few (well-timed) determinations of drug plasma levels, by means of Bayesian analysis [5]. This has also been shown for FVIII [6–12]. During prophylaxis, the trough level, or the time per week spent below a certain level, is important [1,8,9,13]. The aim of this communication is to describe PK methodology that is suitable for the prediction of individual trough levels, and that can be applied in clinical practice.

Blood sampling and Bayesian analysis

An analysis [11] of data from 41 FVIII PK determinations demonstrated that blood sampling at 4, 24 and 48 h gave practically the same information as a conventional (7–10 samples) study for the design of alternate-day dosing schedules. Sampling at 8 and 30 h, or at 24 h alone, gave useful, albeit less accurate, results. Because of the irregularities in the early part of an FVIII coagulant activity (FVIII:C) vs. time curve [3,8,14], blood sampling too early after the infusion is not meaningful (Fig. 1). As a rule of thumb, FVIII:C levels obtained < 4 h post-infusion should not be used in the Bayesian analysis described here. The sampling should be well spread within the ‘B’ region and the times accurately recorded.

Figure 1.

 Representative curve of factor VIII coagulant activity (FVIII:C; data from one patient in [14]) vs. time. Levels at 48 h (C) may be close to the assay limit and/or the endogenous baseline, so a measurement here may be inaccurate. Region B defines the terminal half-life of FVIII:C, and levels can be measured accurately. In region A, FVIII:C levels often lie above the terminal half-life regression line, owing to ‘post-infusion activation’ and/or the ‘two-compartment’ pharmacokinetic behavior of FVIII:C. Thus, the best samples for estimating trough levels are obtained within region B.

FVIII washout is not needed for estimating pharmacokinetics. The Bayesian analysis can be performed on data from practically any dosing schedule. Doses and times of preceding infusions must be known for at least five half-lives (after which < 3% of a dose remains in the body) before the study infusion. Five FVIII half-lives would correspond to up to 5 days in prophylaxis patients.

The Bayesian analysis is based on PK information in the relevant population of patients, as described by a population PK model. The central part of the population model is the structural (compartmental) model, which is defined by the shape of the FVIII:C vs. time curve. Concomitantly, a covariate model describes relationships between PK parameters and patient characteristics, and a statistical model describes the variance between individuals as well as residual (random) variance. For example, from a model based on plasma-derived FVIII in mainly adolescent and adult patients [2]:

  • 1 Structural model: two-compartment.
  • 2 Covariate model for clearance (CL; relationships with BW and age):
    image
  • 3 Coefficient of variation (%CV) of CL between individuals: 28%.

Several similar population PK models for FVIII have been published [2,11,15–17].

Then, according to the Bayesian principle, the first assumption about an individual’s pharmacokinetics, before any FVIII:C data have been obtained, is that it corresponds to the values calculated from the covariate models. The most likely CL is thus calculated from BW and age. Acquisition of data shifts the estimate towards the individual’s actual value. As biological measurements are never exact, a balance is maintained: with few measurements, the estimate of pharmacokinetics is a compromise between the model prediction and the best fit to the data; with more measurements, reliance on the data increases. This balance is regulated by the %CV of PK parameters between individuals in comparison with the residual variance. It should be noted that minor differences in pharmacokinetics between types of FVIII concentrate (this excludes new, long-acting ones) may be treated as general interindividual variance.

Application in clinical practice

If adjusting prophylaxis to an appropriate trough level based on individual pharmacokinetics, in addition to monitoring bleed pattern, is useful, then the introduction of limited blood sampling for the determination of pharmacokinetics has major benefits. Different trough levels may be targeted, depending on circumstances: higher levels may be desired to manage target joints, highly active patients, or those more prone to bleeding; alternatively, lower levels may be allowed in a patient who has not bled for a long time. Because pharmacokinetics changes with growth in young children [2,3,17], and breakthrough bleeds are potentially more damaging, PK information is likely to be more useful at this age.

Because of the inevitable random variance in pharmacokinetics over time, as well as assay uncertainty, it is probably better to perform limited sampling studies on several occasions (and to continuously update the Bayesian estimates of pharmacokinetics) than to perform one multisample study on a single occasion. However, no comparative clinical trial for these approaches has been performed.

Conclusion

Proof of concept has been presented for the use of sparse blood sampling and Bayesian analysis for measuring pharmacokinetics, and for the use of this information to tailor doses in prophylaxis. Validation of these methods for patients with inhibitors, or during surgery or bleeding, is lacking. The procedure could allow routine determination of individual pharmacokinetics in the clinic, potentially making prophylaxis more cost-effective.

Disclosure of Conflict of Interests

The authors state that they have no conflict of interest.

Ancillary