Methods
Studies. Pharmacokinetic data were pooled from three studies where oral 40-mg esomeprazole was given. From PIE, day one concentrations were available from pregnant patients treated with esomeprazole capsules (Nexium brand).13 Day one and day five concentrations obtained from healthy, non-pregnant individuals were included from two studies: those from Hunfeld et al . were treated with MUPS tablets while those from Helgadóttir et al.were treated with Actavis tablets.14,15 Meal times differed between the studies: 2 hours post-dose for PIE and Helgadóttiret al . and 5 minutes post-dose for Hunfeld et al . CYP2C19 genotype information was available for the study by Hunfeld et al .14 In all the studies, esomeprazole concentrations were quantified using validated LC-MS/MS. For Helgadóttir et al.,esomeprazole concentrations were determined at Actavis pharmaceutical company in Iceland. For Hunfeld et al. , the methods used for lab analysis of esomeprazole concentrations have been previously published,16 and analysis was done at the laboratory of the Central Hospital Pharmacy, Netherlands. For PIE, esomeprazole concentrations were determined at the Tygerberg Hospital laboratory, the details of which have been included in the original publication.13 The lower limit of quantifications (LLOQ) were 0.001, 0.0260 mg/l, and 0.00503, PIE, Hunfeld et al . and Helgadóttir et al ., respectively.13–15
Pharmacokinetic analysis. Population pharmacokinetic analysis was performed using non-linear mixed-effects software NONMEM (v7.4.3, Icon®), PsN v4.9.0, Pirana v2.9.8, and R v3.6.1 for data processing.17 The first-order conditional estimation method with interaction was used for model runs. Concentrations below the limit of quantification were handled similarly to the M6 method by Beal, with additive error inflated by LLOQ/2.18
The clearance and bioavailability after a single dose in non-pregnant individuals were used as a reference to report the percentage change in pregnancy or repeated dosing. The pregnancy and the repeated-dosing effects were then tested as a categorical covariate on clearance and bioavailability for each genotype subgroup (CYP2C19 extensive and poor metabolizers). For individuals where genetic information was not available, a mixture model was used for imputation, as suggested by Keizer et al .19 We used proportions fixed to 95% and 5%, for extensive and poor metabolisers, respectively, based on literature.20 Study effects were also tested as covariates on all absorption parameters to account for differences in formulation and mealtimes between the studies.
A non-parametric bootstrap (n=500) was run on the final model to obtain 95% confidence intervals (CIs) for the parameter estimates.