A linked maternal-infant cohort spanning 2,929,600 pregnancies from the Epic Cosmos electronic health record system was constructed and validated using 2023–2024 birth data. Variable completeness exceeded 90% for most perinatal characteristics, though racial/ethnic group (76%) and delivery BMI (77%) lagged. Cohort demographics closely matched U.S. birth certificate population data. Restricting to longitudinal records — from first-trimester prenatal care through 12-week post-discharge infant follow-up — reduced the sample to 586,841 pregnancies, yet cohort characteristics remained stable, suggesting modest selection bias. A validation analysis confirmed that first-trimester systolic blood pressure ≥140 mmHg in women with chronic hypertension raised preeclampsia risk by 26% (aRR 1.26; 95% CI 1.23–1.30).
This is a methods and infrastructure paper rather than a novel clinical discovery, but its implications for maternal-infant research are substantial. Epic Cosmos represents one of the largest centralized U.S. health record repositories, and rigorous pipelines for extracting longitudinal perinatal cohorts have been lacking. The 80% attrition from full to longitudinal cohort is a meaningful limitation researchers must navigate carefully to avoid survivorship bias in long-term outcome studies. Incompleteness in race/ethnicity data also limits health-equity analyses — a critical gap given persistent maternal mortality disparities. As a preprint not yet peer-reviewed, the methodological choices here warrant independent scrutiny before wide adoption. Still, this framework could accelerate pharmacoepidemiology, obstetric safety, and neonatal outcomes research at population scale.