Cohort population analysis of sparse data: Dexamethasone pharmacokinetics in mother and fetus based on blood sampling...

We introduced an estimation technique for mixed effect models that stems from non-parametric methods of estimation and fully parametric mixture models with cohort population as a central concept. The cohort population partitions data variability into between-cohorts and residual variabilities. This permits maximum likelihood estimates of model parameters from sparse data such as one observation per subject. Our second objective was to test robustness of cohort population estimation technique using previously published pharmacokinetic data on dexamethasone in parturient women. We performed two studies. In Study I we generated 100 dense and sparse datasets and refitted the data to quantify the difference between parameter estimates obtained by the cohort population model and the standard estimation methods. In Study II we evaluated robustness of cohort population model using dexamethasone maternal and fetal plasma concentrations obtained at childbirth after intramuscular injection of two 8 mg antenatal doses of dexamethasone phosphate in pregnant women. We implemented in NONMEM 7.6 the 3-cohort population as a mixture population model and applied first order conditional estimation, stochastic approximation, and importance sampling methods for parameter estimation. We showed that cohort estimation can quantify the variability in (log)normal populations where between-cohort variability is a substitute for between-subject variability. The cohort estimates of structural model parameters were equally accurate than the standard estimates and the accuracy and precision of estimates of inter-individual variability parameters for sparse data were better than for the standard maximum likelihood methods. Finally, we showed that the cohort population estimates provided meaningful quantification of one compartment model parameters for dexamethasone in parturient women where data were limited only to one blood sample per subject drawn after childbirth.