Pharmacoepidemiology
Systematic Surveillance of Maternal and Neonatal Outcomes Following Antipsychotic Use in Pregnancy Using Tree-Based Scan Statistics Motohiko Adomi* Motohiko Adomi Adomi Adomi Adomi Adomi Adomi Harvard T.H. Chan School of Public Health
Background: Antipsychotics are used to treat severe psychiatric disorders, and their use has increased in recent years. Despite guideline recommendations for continuation during pregnancy, patients often discontinue treatment due to safety concerns. Prior studies have mainly focused on congenital malformations and prematurity-related outcomes, leaving limited evidence for other maternal/neonatal outcomes.
Objectives: To conduct hypothesis-free safety surveillance of atypical antipsychotic use in pregnancy using tree-based scan statistics.
Methods: Using Medicaid (2000–2018) and MarketScan (2003–2021) databases, we identified a cohort of pregnant individuals linked to live-born infants. Three exposure patterns during pregnancy were evaluated: (i) early exposure (regardless of late exposure), (ii) both early and late exposure, and (iii) late exposure (without early exposure), each compared with no exposure during the respective window. Outcomes included all ICD-9/10 diagnosis codes, excluding congenital malformations, and were organized into a hierarchical tree comprising 9,151 nodes. Exposed pregnancies were propensity score–matched to unexposed pregnancies on the first dispensing date of interest, and follow-up began the next day. Tree-based scan statistics using a Poisson model were implemented, with p-values obtained via permutation tests.
Results: The early (N=17,653), early and late (N=6,458), and late (N=3,375) exposed pregnancy comparisons identified 36, 21, and 0 statistical alerts based on p < 0.05, respectively; 40–60% of alerts corresponded to outcomes previously reported in the literature (e.g., preterm birth, neonatal respiratory distress syndrome).
Conclusions: Tree-based scan statistics enabled systematic surveillance of maternal and neonatal outcomes, identifying known safety signals and additional potential signals. This approach could serve as a scalable screening tool to prioritize outcomes for targeted follow-up studies using refined study designs.
