Prenatal antiseizure medication exposure and offspring structural and neurodevelopmental outcomes: an umbrella review of systematic reviews and meta-analyses

Review-level evidence on prenatal antiseizure medication (ASM) exposure is difficult to interpret because outcome domains are often synthesized separately, primary studies overlap, and comparator frameworks and certainty vary. We aimed to calibrate evidence across structural and neurodevelopmental outcomes. We conducted an anchor-based umbrella review of systematic reviews and meta-analyses (PROSPERO: CRD420251142541), searching four databases through December 31, 2025. One prespecified anchor review represented each drug–outcome–comparator question; overlapping reviews were not combined into second-order pooled estimates. AMSTAR 2, corrected covered area, GRADE, and directional umbrella-credibility classes were applied, with NS assigned when the 95% confidence interval included the null. Fourteen reviews were included: nine structural and five neurodevelopmental. All had Low or Critically Low AMSTAR 2 confidence, and ASD/ADHD review overlap was very high (CCA, 28.1%). Valproate showed the most consistent retained pattern, including Class I credibility for combined major congenital anomalies and the most credible retained neurodevelopmental signal for autism spectrum disorder (adjusted hazard ratio, 3.10; 95% CI, 2.24–4.28; Low certainty; Class II). Topiramate showed an oral-cleft signal with Low certainty and Class III credibility. The lamotrigine estimate for major congenital anomalies was NS, whereas the orofacial-cleft estimate excluded the null but had Very Low certainty (OR, 1.42; 95% CI, 1.05–1.92; Class IV). Other retained neurodevelopmental anchors had Very Low certainty and were Class IV or NS; no independent pairwise neurodevelopmental anchor was available for carbamazepine, oxcarbazepine, topiramate, or levetiracetam. Review-level evidence was most consistent for valproate and aligned with its established structural and neurodevelopmental risks. NS or Very Low-certainty findings for non-valproate ASMs should not be interpreted as evidence of no risk. Comparative safety requires individualized risk–benefit assessment rather than binary safety classifications.