Purpose: A recent direct comparative trial (FENOC) has demonstrated similar efficacy between rFVIIa and APCC for the treatment of joint bleeds in hemophiliacs with inhibitors. This finding is in contrast to the results of previous non-comparative, single-arm trials that have suggested differences in efficacy between the two agents. To place the results of the comparative trial into the context of earlier non-comparative studies of rFVIIa and APCC, a Bayesian meta-regression survival model with time-dependent covariates was developed.

Methods: A systematic review of the literature indicated that different dosages and regimens (e.g. timing of injection) were applied across single arm trials of the two therapies. Results from the single-arm and direct comparative clinical studies included in the systematic review were pooled in a Bayesian random effects meta-regression survival model, within which a repeating Gompertz hazard function (resembling a shark tooth hazard function) was deemed appropriate to model the hazard of bleeding resolution over time, following sequential infusions as would occur in practice. Model covariates included medication type, and the combination of the time-dependent covariates dose and medication type. It was assumed that each subsequent injection had the same efficacy.

Results: Medication type, alone or combined with dosage, significantly influenced the modelled efficacy of the therapies. Using a standard treatment regimen of one 90 mg/kg injection of rFVIIa every 3 hrs, the model estimated that at the 12, 24 and 36 hr time points there was a probability that 76%, 94%, and 99% of bleeds would be treated effectively. In contrast, using a standard treatment of one 75 IU/kg injection of APCC every 3 hrs, the model estimated that there was a 40%, 60%, and 79% probability that bleeds would be treated effectively at the corresponding time points.

Conclusions: The Bayesian random-effects meta-regression survival model developed is well suited to capture the effect of sequential injections when comparative data is limited. The major limitation of the present study is the exclusion of potential time-independent confounding factors (e.g. efficacy rating scale method, treatment setting, efficacy rater, and severity of bleeds).