DISCUSSION
In this study, the predictive performance of a published rFIX-Fc population PK model was evaluated using independent real world data16. The published model was based on patients ≥12 years whereas in this study children with age <12 years were included as well. The published model significantly underpredicted the observed FIX activity levels in all patients, especially for children <12 years of age. Consequently, a new population PK model was developed which should preferably be used to perform PK guided dosing in young children.
Compared to the previously published model, our newly developed model better describes the PK profiles of children <12 years of age that were included. These improvements are not surprising as weight normalized CL and V1 are generally larger in children compared to adults12. This phenomenon has also been reported for recombinant factor VIII-Fc fusion protein (rFVIII-Fc)33. For children <12 years specifically, the novel model shows adequate characterization of CL and V1 (Fig. 2C and 2D).
Observed inter-patient variability of CL and V1, and within-patient variability of CL were somewhat increased in comparison to reported values (Table 2). As real word data is obtained from a highly heterogenic population, a larger variability is imminent compared to selected clinical study populations. This also explains why the residual proportional error in the novel model (16.3%) was slightly higher compared to the published model (10.6%) (Table 2). Real world clinical data may contain more noise due to variability in assay precision, variability in administration and sample times.
Surprisingly, this study found a near two-fold lower typical clearance than reported by Diao et al.16 (Table 2). A possible explanation for this may be related to the neonatal Fc receptor (FcRn), to which the Fc domain of the IgG1 molecule in rFIX-Fc binds. FcRn concentrations are negatively correlated with body weight34. Consequently, children have higher concentrations of weight-adjusted FcRn, possibly resulting in lower CL. This is in contrast to the expected higher FIX CL in children, as is found in factor VIII (FVIII)35,36. As half of our population was paediatric (<18 years) and 38% was <12 years of age, the age related effect on FcRn may have influenced CL estimation.
Our real world clinical data was best described by a two-compartment model and not by a three-compartment model as previously constructed by Diao et al.16 This is due to differences in sampling times during PK profiling between both study populations. More specifically, the published model was constructed based on a rich sampling schedule during a 10-day period, whereas the current study used a maximum of six FIX activity levels sampled during a 7-day period. In the present study, less FIX activity levels were sampled at early time points. This could explain why we were not able to describe a third compartment that characterizes the rapid distribution phase of rFIX-Fc occurring within 2-3 hours after the end of administration37. Notwithstanding these limitations, our model adequately described the terminal elimination phase which determines the trough concentration on which doses are generally adjusted for in clinical practice.
The observed difference in terminal t1/2 between the models is due to the difference in the estimated PK parameters. Nevertheless, the t1/2 of the novel model (70, 76 and 88 h for <12 years, ≥12 and <18 years and adults, respectively) are closer to the reported t1/2 in the Alprolix® SmPC21 (70, 82 and 82 h) than those calculated for the published model (88, 99 and 101 h).
In this study, we have illustrated the clinical impact of underlying population PK models on dosing advice when personalizing treatment. In general, a population PK model should be applied that is representative for the patients for which individual PK are characterized. In our study, however, we did not observe a difference in dose for patients <12 years of age which could be due to the limited number of patients. When considering data from all patients, a significant dose difference was observed, probably caused by the difference in population PK parameters.
In this context, it is important to realize that individual PK parameters are calculated by combining information from both the population and the individual. When more samples are available (5 or more) per individual, individual PK parameters are mainly determined by information from this individual. In the present study, an intermediate clinically representative (3) number of samples was available, hence individual PK parameters were mostly determined by the individual observations. It should however be realized that large differences in dose predictions may occur when less samples are available for an individual patient.
The strength of the present study is that it contains real world data reflecting clinical variability. A study limitation is the relatively sparse sampling method with aforementioned consequences at early time points. The impact of FIX extravascular distribution is recognized by a growing body of literature and should be incorporated in future models38–40. Investigation of extravascular binding of FIX could be of clinical importance, as studies in mice suggest a haemostatic function of extravascular FIX41,42. We, carefully, advocate the use of other techniques, like physiology-based pharmacokinetic (PBPK) models, to investigate an estimation of this extravascular compartment.