(M-097) Accounting for Temporal Differences in Fetal Exposure during Pregnancy using Masking

Yuhan Long – Experimental and Clinical Pharmacology – University of Minnesota; Xintian Lyu – Experimental and Clinical Pharmacology – University of Minnesota; Catherine Sherwin – Internal Medicine – The University of Western Australia; Angela Birnbaum – Experimental and Clinical Pharmacology – University of Minnesota

Objectives: Teratogenicity is a major concern for drugs used during pregnancy. However, the same numerical fetal drug exposure during different periods of pregnancy may not have equivalent effects. For example, some neurodevelopmental outcomes are expected to arise during brain development of the third trimester. Here, we examine bias of misspecified exposure models and introduce temporal masking (i.e., selective weighting of exposure by gestational age) to address this bias.

Methods: For simulations, a one-compartment pharmacokinetic (PK) model with individual variability (n>1000) and an oral dosing strategy of twice daily with two increases at randomized times was used to simulate plasma concentrations during pregnancy. Increases in clearance and volume during pregnancy were modeled using the Emax function. Postpartum pharmacodynamic outcome was modeled using an Emax function with the exposure-response relation defined by exposure linearly decreasing maximal effect. Exposure was weighted using a Gaussian curve (mean 32 weeks, standard deviation 6 weeks) to model temporal variation. The final measured response was calculated as the value at year 6 in the child.
For fitting, PK parameters were fixed to individual-level simulation parameters to remove fitting error due to PK. Exposure was computed as an area under the curve (AUC) during pregnancy. Exposure was controlled via temporal masking using square and Gaussian curves with center and spread. Ground truth exposure, no mask, and trimester masks were also considered for comparison. Linear regression was used to fit the measured response to the computed exposures. Adjusted R^2 was used to compare methods. Sensitivity analyses were completed on the center and spread of the masks.
Simulation and fitting were performed using R and mrgsolve.

Results: Using a temporal mask resulted with both square (0.993) and Gaussian masks (0.994) achieving adjusted R^2 close to the ground truth exposure (0.994). No mask resulted in a lower adjusted R^2 of 0.983. 1st, 2nd, and 3rd trimester masks obtained values of 0.936, 0.984, and 0.993, respectively. The lower value for 1st trimester is expected due to misspecification of exposure timing. This is seen again in the sensitivity analyses on the mask center. Adjusted R^2 worsened the further the center was from the ground truth, reaching 0.942 at gestational week 10. Sensitivity on spread showed expected results with increasing mask width leading to exposure-response fits approaching the adjusted R^2 of no mask. Introducing variability to the measured response significantly decreased the adjusted R^2, though similar sensitivity trends of the exposure methods surfaced.

Conclusions: Temporal weighting of exposure can improve pregnancy-specific exposure-response models. While exact timing of exposure may not always be known, model assumptions that align with biological mechanism (e.g., third-trimester vulnerability) outperform both unweighted and misspecified models. Furthermore, identifying exposure timing can enable optimized study design and reduced participant time commitment. This framework supports more accurate pharmacodynamic modeling in pregnancy by accounting for gestation-specific susceptibility to drug effects.

Citations: None

Keywords: Pregnancy, Exposure, Masking