Emulating a Target Trial to Estimate the Causal Effect of Antiseizure Medication on Birthweight

Background: Results of previous observational studies on the effect of antiseizure medication on birthweight are conflicting, which might be due to methodological challenges. For example, if comparing monotherapy with two different medications, retrospective assignment to exposure groups does not adequately consider women who change or stop treatment during pregnancy, which may introduce bias if this is not at random. The target trial emulation (TTE) framework helps researchers to design a study that reduces these biases and improves the interpretability of effect estimates.

Objectives: To compare the causal effect of antiseizure medications during pregnancy on the risk of small for gestational age (SGA) newborns using the TTE framework, exemplified by the comparison of lamotrigine vs. levetiracetam monotherapy.

Methods: Using data from the German Pharmacoepidemiological Research Database (GePaRD; claims data covering ~20% of the German population), we emulated a target trial in pregnant women with epilepsy, a continuous insurance of ≥12 months before conception, and a dispensing of lamotrigine or levetiracetam overlapping conception. We included all pregnancies occurring in these women between 2005 and 2020. We compared the strategies A) use of lamotrigine vs. B) use of levetiracetam, both as monotherapy from conception to pregnancy end. We used ordinal and logistical regression and adjusted for age, socioeconomic status (SES), and risk factors for SGA. Inverse probability of censoring weighting was used to account for the fact that a deviation from strategy A or B during pregnancy may not be random. SGA was defined as birthweight < 10th percentile.

Results: Our study included 2520 pregnancies with lamotrigine and 747 pregnancies with levetiracetam monotherapy at conception. In both strategies median age of women at conception was 31 years. The proportion of live births was 79% in lamotrigine exposed pregnancies and 77% in levetiracetam exposed pregnancies. The proportion of SGA live births was 8% in lamotrigine exposed pregnancies (A) and 11% in levetiracetam exposed pregnancies (B). After adjustment for potential confounders, pregnancies exposed to levetiracetam monotherapy had a 1.4 times higher risk of SGA newborns than pregnancies exposed to lamotrigine monotherapy (95% CI: 1.08-1.89).

Conclusions: According to our target trail emulation the risk of SGA children was higher in pregnancies exposed to levetiracetam as compared to pregnancies exposed to lamotrigine monotherapy. Adequate modeling of informative stop of respective monotherapy and use of ordinal regression to account for non-live births requires more advanced statistical methods than usual analyses, but the extra effort leads to results that are less likely to be biased.