Longitudinal Methods for Modeling Exposures in Pharmacoepidemiologic Studies in Pregnancy

Area of Science:

• Pharmacoepidemiology
• Perinatal Health
• Biostatistics

Background:

• Perinatal pharmacoepidemiologic studies often use simplified "ever vs. never" medication exposure classifications.
• These methods fail to capture real-world medication use complexity, including dosage, timing, and duration.
• This can lead to exposure misclassification and biased safety assessments.

Purpose of the Study:

• To review advanced exposure modeling methods for medication use in pregnancy.
• To highlight the strengths and limitations of techniques for capturing complex, time-varying medication exposures.
• To propose integrated approaches for robust medication safety evaluations.

Main Methods:

• Overview of unsupervised clustering methods (k-means, group-based trajectory models, hierarchical clustering) for visualizing medication use trajectories.
• Discussion of time-varying exposure analytical techniques (extended Cox models, Robins' generalized methods).
• Exploration of combining clustering with causal modeling.

Main Results:

• Unsupervised clustering aids in understanding medication use patterns, comedication, and drug switching.
• Time-varying methods are crucial for non-static medication exposure during pregnancy.
• Integrated approaches offer a powerful framework for medication safety research.

Conclusions:

• Refined exposure modeling is essential for accurate perinatal medication safety studies.
• Combining unsupervised clustering with causal inference provides a more comprehensive understanding of medication effects.
• This framework has broader applicability in epidemiological research.