ARTICLE TYPE : RESEARCH ARTICLE
Published on : 08 Aug 2026,
Volume - 2
Journal Title :
WebLog Journal of Obstetrics and Gynecology
| WebLog J Obstet Gynecol
| WJOG
Journal ISSN: 3142-922X
Source URL:
https://weblogoa.com/articles/wjog.2026.h0811
Permanent Identifier (DOI) :
Testing the Limits: Can Federated EHR Networks Detect Teratogenic Signals?
2Department of Internal Medicine, Howard University, Washington, DC, USA
3Department of Internal Medicine, University of Maryland School of Medicine, Baltimore, USA
Abstract
Background: First-trimester topiramate exposure carries an approximately 2-fold increased risk of Major Congenital Malformations (MCMs) compared with levetiracetam, which has among the lowest MCM rates of Anti-Seizure Medications (ASMs). Federated Electronic Health Record (EHR) networks are increasingly used for drug safety research; however, their ability to replicate known teratogenic signals has not been extensively evaluated.
Objective: To examine whether the TriNetX Global Collaborative Network can replicate the established differential teratogenic risk between topiramate and levetiracetam.
Methods: This retrospective cohort study used the TriNetX Global Collaborative Network (173 healthcare organizations). Women aged 15-45 with first-trimester gestational age codes prescribed topiramate without levetiracetam (n=40,990) or levetiracetam without topiramate (n=14,429) were identified. Propensity score matching (PSM; 1:1) on 28 covariates yielded 8,112 matched pairs. Outcomes ascertained from maternal records included overall congenital anomalies (ICD-10-CM Q00-QA0), congenital heart defects (CHDs; Q20-Q28), and clubfoot (Q66). Oral clefts, the most specific topiramate-associated anomaly, could not be reliably assessed because infant diagnoses were not linkable to maternal records. Hypospadias (Q54) was included as a negative control outcome.
Results: After matching, no significant differences were detected for overall congenital anomalies (4.7% vs. 5.1%; odds ratio [OR] 0.92, 95% confidence interval [CI] 0.79-1.07; p=0.265), CHDs (1.7% vs. 1.4%; OR 1.21, 95% CI 0.94–1.57; p=0.144), or clubfoot (0.2% vs.0.3%; OR 0.61, 95% CI 0.33 1.14; p=0.120). Hypospadias yielded zero events in both cohorts, confirming the absence of infant level outcome capture. Kaplan-Meier analysis for CHDs paradoxically showed a higher event rate in the levetiracetam cohort (hazard ratio 1.41, 95% CI 1.10-1.83; p=0.008), likely reflecting differential diagnostic ascertainment rather than a true biological signal.
Conclusions: Despite 8,112 matched pairs, this analysis did not replicate the expected topiramate associated teratogenic patterns. These findings highlight critical limitations of unlinked maternal EHR data for pregnancy medication safety research, including absent infant outcome capture, imprecise exposure timing, and insufficient power for rare anomalies. Federated EHR platforms without validated mother-infant linkage should not be used to generate evidence on medication teratogenicity.
Keywords: Topiramate; Levetiracetam; Teratogenicity; Congenital Malformations; Electronic Health Records; Federated Data Networks; Pregnancy Safety; Pharmacoepidemiology; TriNetX; Mother-Infant Linkage
Citation
Tawfik F, Sienkianiec J, Deverapalli M, Rameau A, Ehikioya O, Taiwo S, et al. Testing the Limits: Can Federated EHR Networks Detect Teratogenic Signals?. WebLog J Obstet Gynecol. wjog.2026. h0811. https://doi.org/10.5281/zenodo.21860752