(A-279) BOOST-HP DrugScan: Identifying Medications Associated with Pregnancy Loss

Background: While common, causes of pregnancy loss are largely unknown. Developing knowledge on drug exposures that may contribute to pregnancy loss is critically important.

Objectives: To identify drug exposures associated with pregnancy losses, compared to live births, when using a case-control study design and tree-based scan statistics.

Methods: We included pregnancies in Medicaid and commercial claims data from 2000-2023 that were not exposed to known teratogens in their first trimester, nor had evidence of chromosomal abnormalities or single-gene conditions, or cancer. We compared pregnancy losses (spontaneous abortions (SAB) and stillbirths (SB), separately) to live birth control pregnancies and created comparable risk sets by stratifying on maternal age groups, diagnosis coding era, race, and deciles of a pregnancy loss risk score measured in the 90 days prior to conception. We sequentially established risk sets at each gestational week, scanning antecedent drug exposures prior to that week. We used the Anatomic Therapeutic Classification system with both medication administrations and pharmacy dispensings to allow scanning at the drug and class level, excluding treatments for excessive bleeding and hormonal products used in assisted reproduction. To prevent identifying mediators, we censored drug ascertainment once symptoms of pregnancy loss began and excluded drugs observed in the 7 days prior to pregnancy end. Using conditional Poisson tree-based scan statistics, we considered statistical signals to be drugs with an odds ratio (OR) ≥2 and p-value ≤0.05 after adjusting for multiplicity.

Results: We identified 657,893 and 527,307 SABs, 25,091 and 10,920 SBs, and 2,538,338 and 1,828,584 live births in Medicaid and commercial data, respectively. Before adjustment, pregnancies resulting in loss were older and more likely to have prior pregnancy complications and cardiovascular disease. SBs were more frequent among black patients and those with pre-existing diabetes and hypertension. Study pregnancies were exposed to >1000 unique drug products. Both SAB and SB signals included anti-inflammatory medicines (e.g., ibuprofen OR=2.2, ketorolac OR=2.0); central-acting muscle relaxants OR=2.5; opioids and opioid combinations OR=2.8-3.3; fluoroquinolones (class OR=2.1 and drug-specific, e.g., ciprofloxacin OR=2.0); and antihypertensives (class OR=2.0-2.3 and drug-specific, e.g. lisinopril OR=2.4).

Conclusions: BOOST-HP DrugScan is a useful tool for screening potential associations that can be evaluated in fully controlled pharmacoepidemiologic studies with active comparators. Triage is ongoing for the identified signals. Even with robust confounding control strategies, protopathic bias or residual confounding may still be present.