Results

Study profile

Bedaquiline pharmacokinetic samples were available from 13 women in the third trimester of pregnancy, at 30 (IQR: 25 - 37) weeks gestation, six of whom returned for postpartum sampling at 7 (IQR: 6.5 – 8) weeks. Seventy-one plasma samples of bedaquiline parent and metabolite concentration were available for analysis. Participant characteristics are shown in Table 1. All participants were living with HIV were treated with antiretroviral therapy (ART), most commonly on nevirapine-based ART (n=10, 83.3%), but one woman received lopinavir/ritonavir. Serial human milk samples at the same time-points that plasma was sampled, were available in two breasfeeding participants. A single random plasma bedaquiline concentration was available from four infants on the postpartum PK sampling day, of whom one was breastfed. The serial post-dose bedaquiline concentrations at each time point are shown in Table 2.

Pharmacokinetic model

When we used the published model (14) to predict the expected exposures in these patients (thus using the original population parameter estimates and assuming no effect of pregnancy), the model overpredicted both bedaquiline and M2 concentration on both antepartum and postpartum visits, as presented in the visual predictive check (VPC) inFigure 1 . The VPC shows that the PK terminal elimination phase of the participant not on lopinavir/ritonavir were approximately 50% lower that the model prediction (for both the metabolite and parent) as illustrated by the deviation of the 50th percentiles of the observations (red line) from the median of the model predicted confidence interval (black line). If the pharmacokinetic parameters in this study were in line with the previous report, we would have expected to observe higher bedaquiline concentrations. Only the data from the participant co-administered lopinavir/ritonavir, who had higher bedaquiline concentrations due to a drug-interaction, was in line with the model prediction.
When encountered several challenges when attempting to fit the original model to the current data by re-estimating the parameter values. The model structure is complex, with multiple disposition compartments, and the current data did not reliably support the re-estimation of all parameters - some of the parameter estimates obtained when attempting to re-fit were unstable and/or implausible. In other words, while the model could be adapted to fit the study data, this could be achieved in multiple different ways, e.g. assuming a larger clearance or lower bioavailability (both pre- and postpartum), but also a larger peripheral volume of distribution. We experienced further complications when trying to estimate a significant difference between the two pharmacokinetic sampling visits, i.e. possibly due to pregnancy status. All the scenarios were nearly equivalent in terms of goodness of fit, and there was no meaningful difference in terms of statistical significance, thus leaving the choice largely in the domain of speculation. Choosing a different scenario (on which a difference is ascribed to) would imply a different interpretation of the results, and if the different options for the model were to be used to predict concentrations and suggest dose adjustments they could come to very different conclusions. We also attempted to use a frequentist prior approach (20) to try and stabilise the parameter estimates, but the results became highly dependent on the assumptions on the prior precision of each parameter, thus not solving the problem. For this reason, we decided to simply use the model as originally published and acknowledge that the concentrations we observed are lower than expected, assuming that the pharmacokinetics are the same as non-pregnant patients.
A graphical overview of the infant and human milk data is provided in Figures 2 and 3, together with the plasma concentrations in the respective mothers. The PK profiles for bedaquiline and M2 are shown: maternal plasma concentrations ante- and postpartum, human milk and infant concentrations. To calculate the human milk to maternal plasma (M:P) ratio, we used matched milk and plasma samples captured more than 24 hours after the last bedaquiline dose. This was done because during drug absorption the ratio was rapidly varying and the later samples better reflect the steady-state equilibrium of bedaquiline. We observed an average M:P ratio of 24:1 and 5.9:1 for bedaquiline and M2, respectively. Therefore, assuming a maternal average steady-state concentration of 0.3 mg/L, the infant bedaquiline dose would be 1.08 mg/kg/day. Similarly, assuming the M2 average steady-state concentration is 0.1, the infant M2 dose would be 0.09 mg/kg/day. In comparison, a 70-kg individual administered the standard dose of 200 mg bedaquiline three times a week would result in approximately 1.22 mg/kg/day dose of bedaquiline.
Bedaquiline and M2 concentrations in the infant who was breastfed were similar to maternal plasma concentrations, while for the three infants who were not breastfed, bedaquiline and M2 concentrations were detectable but lower than maternal plasma values (see Figures 2 and 3).