Background:
Drugs taken during pregnancy can influence a child’s development, such as neurodevelopmental outcomes, that may only manifest years after birth. This temporal disconnect of short-term prenatal exposure with delayed outcomes presents challenges in quantifying exposure-response (ER) relationships. To address this, we developed a population pharmacokinetic-pharmacodynamic (PK/PD) modeling framework that characterizes in utero drug exposure and delayed longitudinal pharmacodynamic response within a unified disease progression model.
Methods:
A one-compartment maternal PK model with first-order oral absorption was used to simulate drug (levetiracetam, Cl = 3.6 L/h, V = 55 L) disposition during pregnancy. Gestational changes were incorporated via an Emax function on clearance (max change at 2 L/h and 50% change at 14 weeks) and a linear change in volume over 40 weeks. Dose escalations reflected therapeutic adjustments (1000 to 1250 to 1400 mg BID). Fetal exposure was assumed equal to maternal levels (ratio = 1.0). For the PD component, a disease-progression model was implemented: a response compartment captured the child’s neurodevelopmental growth trajectory (rate = 1e-3 [1/h]), and prenatal drug exposure inhibited the growth rate in proportion to fetal exposure (inhibition = 1e-5 [mL/ug]).
Simulations included drug-exposed (n=150) and unexposed control (n=100) cohorts, reflecting parameters of the Maternal Outcomes and Neurodevelopmental Effects of Antiepileptic Drugs (MONEAD) study design. Maternal plasma samples were generated for each trimester with uniformly randomized timing with a minimum of 1 month between each sample. Neurodevelopmental milestones were generated at ages 3, 4.5, and 6 (with SD of 0.25) to reflect collection of neurodevelopmental outcomes collected in MONEAD. Inter-individual variability was included on key PK and PD parameters to mimic population heterogeneity. The simulated data were fitted using nonlinear mixed-effects modeling to assess whether the known ER relationship could be identified.
Simulations were completed in mrgsolve and R, and fitting was completed using NONMEM via PsN.
Results:
The simulation was able to represent PK exposure and its reductive impact on neurodevelopmental growth curves. Simulated levetiracetam concentrations was distributed 21.8±8.6 ug/mL [mean±sd]. The model could distinguish the neurodevelopmental profiles of drug-exposed (mean at 6 years = 14.4±4.6) vs unexposed cohorts (mean at 6 years = 52.1±6.1), demonstrating its ability to link short-term in utero exposure to later childhood outcomes. Due to the sparse sampling strategy, the model could not fit all set parameters, so the fitted model was predicted on a reduced parameter set. The reduced population PK/PD model described the simulated longitudinal data and captured the ER relationship between prenatal drug levels and neurodevelopmental outcomes. The fit consistently returned clearance parameters (3.7 L/h with a max change of 2.0 L/h), but had difficulties in predicting volume (71 L) due to confounding by PK variability. The model fit successfully recovered drug PD effect parameters, confirming that the framework can detect the impact of prenatal exposure on developmental growth. The growth parameter was predicted at 9.1e-4 [1/h], while the inhibitory factor was overpredicted at 2.6e-5 [mL/ug]. This discrepancy is possibly due to variability in exposure from sparse sampling. Removing the unexposed cohort reduced model identifiability and caused growth rate and inhibitory factor to be confounded on each other see in increased inhibitory factor and decreased growth rate.
Conclusions:
We present a proof-of-concept population PK/PD model that links a brief prenatal drug exposure to the delayed measurement of neurodevelopmental outcomes in children. This framework provides a quantitative approach to evaluate longitudinal long-term developmental effects within a pharmacometric context, supporting both retrospective analysis and prospective study design in pregnant populations. The model, while requiring certain assumptions about neurodevelopmental growth, can be extended by integrating mechanistic brain development trajectories. Overall, this approach bridges the gap between short-term in utero exposure and long-term developmental outcomes, supporting more informed risk-benefit assessments of medication use during pregnancy.