Using Tree-Based Scan Statistics for Signal Identification: Screening for Elevated Congenital Malformation Risk Following Prenatal Antipsychotic Exposure

Background: Drug-safety studies in pregnancy typically focus on selected or composite outcomes (e.g., any congenital malformation). Tree-based scan statistic (TBSS) approaches allow for simultaneous screening across a broad range of potential adverse pregnancy outcomes (e.g., individual malformations), while controlling the overall rate of false-positive signals.

Objectives: To evaluate the risk of 1st trimester (T1) antipsychotic (AP) exposure with respect to congenital malformations, using a TBSS approach.

Methods: Using a US-nationwide cohort of >4.2 million mother-child dyads nested within the Medicaid data 2000-2018 and the MarketScan Research Database 2003-2020, pregnancies with ≥1 AP dispensing during T1 were compared to those not exposed to any AP from 3 months before pregnancy to the end of T1. Exposure was identified at the individual drug level. After restricting to drugs with ≥50 exposed pregnancies, 15 individual APs were studied. The “tree” in the TBSS approach was created using the hierarchical structure of the International Classification of Diseases, focused on malformations codes, augmented with groupings of malformations that were found through the use of artificial intelligence with human supervision. Two outcome definitions were used: (1) ≥1 malformation code and – in order to increase the specificity of the outcome – (2) multiple records for the same malformation. Outcomes were assessed within 3 months after birth. We used propensity score fine-stratification to control for treatment indication and other confounders and an unconditional Poisson scan statistic to estimate relative risks (RR). Outcomes with >3 exposed cases and p< 0.05 were considered statistical alerts. P values were used as a means to rank and prioritize alerts for further investigation rather than to decide on the presence of a causal association.

Results: The number of exposed pregnancies ranged from 61 for cariprazine to 18,366 for prochlorperazine. Based on the presence of ≥1 code, alerts for an increased risk were observed for: skin anomalies associated with aripiprazole (RR=1.6), haloperidol (RR=2.1), and quetiapine (RR=1.6); and polydactyly of fingers associated with ziprasidone (RR=3.9). When requiring multiple codes, the alert strengthened for polydactyly associated with ziprasidone (RR=6.8), but alerts were no longer present for skin anomalies as most children do not have multiple records for these malformations.

Conclusions: Using TBBS to evaluate a comprehensive range of individual malformations and malformation clusters, we identified some potential alerts associated with prenatal AP exposure that have not previously been reported. As a next step, these alerts will need to be evaluated in follow-up studies.