Identification and dating of pregnancies in administrative healthcare claims databases: An updated, transparent, and transportable algorithm.

Abstract

Studies using routinely collected healthcare data are essential for understanding the benefit-risk profile of medications in pregnancy. However, reproducibility is limited by the lack of shared pregnancy identification algorithms, and generalizability is constrained by their development primarily in commercial insurance databases. A previously developed pregnancy identification algorithm was updated for ICD-10-CM and translated into a standardized format to support transport across diverse databases. Algorithm implementation characteristics were evaluated in three US health insurance claims databases (2000-2025), including commercially and publicly insured populations. Overall, 5.8 million and 3.0 million pregnancies were identified in two commercial databases (Merative™ and Optum) and 2.3 million pregnancies in Merative™ MarketScan® Multi-State Medicaid Database (MDCD). Across databases, live birth prevalence ranged from 76-83%, miscarriage/abortion 15-23%, ectopic pregnancy 1%, and stillbirth ≤0.5%, with a higher proportion of live births in MDCD. Pregnancy dating methods varied substantially across coding eras: use of gestational age records increased from 4-9% in the ICD-9-CM era to 87-92% during the ICD-10-CM era, with greater use in MDCD. Using a standardized algorithm, this study demonstrates how database characteristics and coding transitions influence pregnancy identification and dating in healthcare data, highlighting methodological considerations for reproducible pregnancy research across data sources.