Predictive performance of the published model
The predictive performance of the published population PK model for
rFIX-Fc was evaluated by comparing the model-predicted and observed FIX
activity levels. Figure 1A and Figure 1B present the population
predictions goodness-of-fit (GOF) plots for all patients and children
<12 years of age separately. Observed FIX activity levels are
higher than their respective predictions (Fig. 1A and 1B) and a clear
deviation of trend lines from identity lines can be seen in all patients
(Figure 1A), but especially in children <12 years of age
(Figure 1B). These observations indicate structural bias
(underprediction) of the published model. This is also illustrated by
the median PE of -48.8% (IQR: -29.9 – -63.9) for all patients and
-54.1% (IQR: -43.3 – -65.8) for children <12 years of age
(Supplementary Table 2A). The RMSE is shown in Supplementary Table 2B.
Furthermore, deviations were observed in plots of conditional weighted
residuals (CWRES) versus population predictions (PRED, Sup. Fig. 1A) and
time after dose (TAD, Sup. Fig. 1B).
Bayesian analysis was performed to obtain individual PK parameter
estimates. For children <12 years of age, Figure 2A and 2B
show the deviation from the individual PK estimate from the typical
population value over the weight range. For the evaluation of these
graphs, it is important to realize that an adequate population model
would have random inter-patient variability with an average of zero and
no trend with weight. Figure 2A and 2B clearly demonstrate that
children’s CL and V1 are lower than would typically be expected over the
studied weight range and advocate the development of a new model.