2G-06 - Benzodiazepines and spontaneous abortions: sequential target trial emulation design and other methodological considerations

Background: Previous studies have suggested an approximately 2-fold increased risk of spontaneous abortions (SABs) associated with benzodiazepine (BZD) exposure during early pregnancy. However, interpreting these findings is challenging due to methodological limitations, including immortal time bias arising from misalignment of time zero, and reverse causation bias resulting from confounding by prodromal symptoms before SABs occur.

Objectives: To evaluate the association between BZD treatment in early pregnancy and risk of SAB.

Methods: We conducted a cohort study nested in two nationwide US databases, Medicaid (2000-18, publicly insured) and MarketScan (2003-20, privately insured). To mitigate immortal time bias, we used a sequential target trial emulation approach and assigned treatment status on a weekly basis. For example, for the week 1 trial, pregnancies with a BZD prescription filled between the date of last menstrual period [LMP] and LMP+7 days were classified as treated, while those without a BZD prescription were classified as untreated. Fine and Gray subdistribution hazards models were used to compare the risk of SABs, with terminations considered as competing events. Analyses were conducted at four progressively adjusted levels: (0) crude analysis; (1) excluding pregnancies with only short-term BZD prescriptions ( < 3 days' supply), which are more likely to be prescribed preoperatively for abortion procedures, to reduce reverse causation bias; (2) applying a lag-time approach by shifting the event date three days earlier to exclude BZDs prescribed for unmeasured prodromal symptoms of SABs; and (3) using propensity score overlap weighting to control for potential confounders.

Results: Among 5,079,190 pregnancies in Medicaid and 3,535,136 pregnancies in MarketScan, 116,738 (2.3%) and 69,250 (2.0%), respectively, filled a BZD prescription during early pregnancy. In the Medicaid cohort, the pooled hazard ratios (HRs) for SABs from the sequential trials were 1.68 (95% CI 1.66-1.70) in the crude analysis, attenuating to 1.67 (1.65-1.69), 1.65 (1.63-1.67), and 1.19 (1.18-1.21) in the three adjusted analyses. A similar attenuation was observed in the MarketScan cohort, with HRs of 1.52 (95% CI 1.50-1.54), 1.49 (1.46-1.51), 1.43 (1.41-1.46), and 1.23 (1.21-1.25), respectively.

Conclusions: Careful attention to immortal time bias, reverse causation, and confounding led to a marked attenuation of the association between BZD treatment and the risk of SABs. However, profile review suggests that concerns about residual bias remain. Future research will explore additional approaches to further address these concerns before drawing causal conclusions.