Perinatal & Pediatric
Characterizing pregnancies with unobserved outcomes in administrative claims data Chase D. Latour* Chase Latour Latour Latour Latour Latour Latour Latour Latour Latour Department of Biostatistics, Epidemiology, and Informatics, University of Pennsylvania, Philadelphia, PA, USA; Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
Introduction: Including pregnancies with unobserved outcomes (ie, lost to follow-up [LTFU]) may reduce selection bias but exacerbate measurement error in claims-based studies of prenatal exposures. Evidence is limited on how to best balance these concerns.
Objective: To characterize differences between pregnancies with observed versus unobserved outcomes.
Methods: We identified pregnancies using U.S. commercial insurance claims data based on pregnancy-related diagnosis and procedure codes. We included pregnancies with estimated last menstrual periods (LMPs) from 7/1/2016-2/1/2023 and ≥180 days pre-LMP continuous enrollment. Pregnancy outcomes were classified as observed versus LTFU, and among LTFU, by the presence of ≥1 gestational week-specific diagnosis code (Z3A.08–Z3A.42). We compared demographics, comorbidities, gestational age (GA) at end of pregnancy or LTFU, and timing of disenrollment for observed vs LTFU pregnancies and, among LTFU, by Z3A code presence.
Results: Of 1,818,342 pregnancies, 470,726 (26%) were LTFU. LTFU pregnancies had more comorbidities than observed pregnancies (chronic hypertension: 6% vs 4%). Among LTFU, only 35% had ≥1 Z3A code, and they had fewer comorbidities than those with 0 Z3A codes (chronic hypertension: 3% vs 7%). GA at LTFU differed by strata: 41% of those with ≥1 Z3A code were <20 weeks’ gestation at LTFU, compared to 83% of those with 0 Z3A codes. Cumulative incidence of disenrollment within 30 days after LTFU differed: 60% for ≥1 Z3A code versus 24% for 0 Z3A codes.
Conclusions: Pregnancies LTFU differed from those with observed outcomes and by the presence of Z3A codes. Those with ≥1 Z3A code had fewer comorbidities and disenrolled quickly after LTFU, compared to those with none. Excluding pregnancies with such differing characteristics could result in meaningful selection bias in studies of prenatal exposures. Ongoing manual claim review will clarify LTFU pregnancy validity.
