In this study, a LSS for the EHL-FVIII concentrate BAX 855 was developed for patients with severe hemophilia A. The best LSSs included sampling time points at 15-30min, 48h and 72h (LSS7) or 15–30 min, 48h (LSS14) for both adults and children. The blood samples of LSS7 can be obtained during two visits to the clinic, when the 72h sample is taken as a pre-dose level. In this way - during the first clinical visit - a sample is taken 72h after BAX 855 is administrated at home, a new BAX 855 dose is administrated and 15–30 min after dose the second sample is collected. Two days later the second clinical visit is planned to take the sample 48h after the dose.
The selected strategy with two samples (peak, 48 hours) resulted in adequate, but lower accuracy than strategies with ≥ 3 samples. We recommend this 2-sample design - which can be collected in one clinical visit - especially for children with difficult venous access or in low income countries. The two-sample design including a peak and 72h sample (LSS15) also demonstrated adequate relative errors in the PK parameters. However, for 9% of the adults and 12% of the children the 72h sample will be BQL and thus undetectable, complicating adequate estimation of the 72h trough level for these individuals when this LSS is applied. Treatment centers can decide to select LSS15, but for 9–12% of the patients, it necessitates taking a new 72h sample when the sample is BQL.
Interestingly, similar LSS for both adults and children were selected, despite a difference in the number of BQL samples. This suggests that the difference in BQL samples is too small (9 vs 12% for the 72h sample) to result in better performance of other sampling schemes. Moreover, in our results, the LSS including a 96h sample, performed less adequate than sample schemes including 48 or 72h samples. This is caused by a high percentage (45%) of patients for who this sample is BQL and thus less informative. Presumably, this sample can be informative for patients when being above BQL. However, with the current approach beforehand it is not known for which patients this will be the case. A possible solution would be to use personalized limited sampling approaches.[26]
In our study, we also illustrated how the LSSs of SHL FVIII concentrate and BAX855 can be combined resulting in six blood samples during four visits. In contrast, when the two PK profiles are not combined and especially when the pre-dose level is not used as trough level, eight samples during six visits to the clinic are needed to construct both PK profiles. Importantly, the approach of combining the sampling of two PK profiles may be applied to other FVIII and FIX concentrates, when limited sampling strategies of the specific factor concentrate products are present, such as for rFVIIIFc (Elocta®). When this combined sample design for two concentrates is performed, individual PK parameters for both concentrates can be obtained. Using Bayesian estimation, knowledge about previous FVIII levels during onset of bleeds and physical activities can be obtained when - in for instance a logbook of a patient - administration times of factor concentrate doses are available. These levels can be informative to set target levels for the PK-guided dosing of the new concentrate. However, it is important to bear three limitations in mind. Firstly, the PK-pharmacodynamics (PD) relation between two factor concentrates may be different, meaning that a similar targeted FVIII level could result in a different effect and thus different bleeding tendency. Previously published studies are too limited to compare the PK-PD relation of factor concentrates, but differences in obtained annual bleeding rate (ABR) values with similar doses could indicate a different PK-PD relation. Secondly, time spent above or below certain factor level may be clinically relevant and differs between for instance SHL and EHL factor concentrates, although an identical trough level is targeted.[5] To overcome this limitation, the time above a certain factor level could also be calculated. Thirdly, differences in one-stage and chromogenic assay can vary per factor concentrate, making it more difficult to compare the factor levels on both concentrates.[27]
In the preferred strategy for BAX 855, the peak, 48h and 72h sample were present. This finding is partly consistent with the LSS for rFVIIIFc (Elocta®) of Mc Enemy-King et al., which selected a 72h trough, 1h peak and 72h as 3-point sample design for adults.[12] Our results cannot be compared to a LSS for BAY 94-9027 (Jivi®) as Solms et al only evaluated the following sampling designs in their study: one sample, two samples (4h, 48h) or three samples (4h, 24h, 48h).[28] The selected LSS for BAX 855 includes two to three samples. This is in accordance with the recommendation of the subcommittee on FVIII, factor XI and rare bleeding disorders of the ISTH that proposed two to four post infusion samples.[13] Furthermore, the study of Blanchette et al compared a 6-sample design to a 2-sample design in a PK study of Advate with 39 patients with hemophilia A, and concluded that the 2-sample design showed sufficient accuracy to be used in routine clinical care.[29] This conclusion was based on an substantial and almost perfect interclass correlation for clearance (0.73) and t1/2 (0.84) between the PK parameters in the 6- and 2-sample design. Though, it should be noted that the PK parameters calculated with a 6-sample design are not the exact (true) individual PK parameters, making the agreement of the 2-sample design with true individual PK parameters smaller.
The results of this study are based on simulations and therefore have not yet been validated with real-life patient data in a clinical setting. Furthermore, the predictive performance of the LSSs is dependent on the population PK model that is used for the simulations and may differ depending on the patient characteristics and sampling design of the data used for model development. The population PK model of BAX 855 used in this analysis included 154 hemophilia A patients from 3–72 years.[17] The results therefore reflect a population that is similar to the population that was used for building this model.