Pregnancy and Lactation

6B-04 - Algorithms for identifying pregnancies and their outcomes in a US Claims Database

Tuesday, August 26, 2025

2:15 PM - 2:30 PM ET

Presenting Author(s)

BZ

Baijun Zhou

Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, United States

Co-Author(s)

CL

Chase D. Latour

Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, United States
Elizabeth Simmons, PhD

Cecil G. Sheps Center for Health Services Research, United States
OO

Oluwasolape Olawore

Gillings School of Global Public Health, University of North Carolina at Chapel Hill, United States
MD

Megan Delgado, MPH

PhD Student
Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, North Carolina, United States
Jacob C. Kahrs, MPH

Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, United States
LK

Lauren M. Kucirka, MD, PhD

Assistant Professor
University of North Carolina School of Medicine, North Carolina, United States
Elizabeth A. Suarez, PhD

Center for Pharmacoepidemiology and Treatment Science, Rutgers Institute for Health, Health Care Policy and Aging Research, New Brunswick, NJ, USA, United States
SP

Sharon Peacock-Hinton

Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, United States
MW

Mollie E. Wood, PhD

Assistant Professor
Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, United States

Background: While multiple algorithms have been developed to identify pregnancies in administrative data, there has been little attention to differences in samples identified by different algorithms.

Objectives: To develop 4 claims-based algorithms to prospectively identify pregnancies and compare pregnancy outcomes of the identified populations with those identified by an adaptation of a published retrospective algorithm.

Methods: We developed 4 algorithms to identify pregnancies using a 1% random sample of the MerativeTM Marketscan® Research Database in 2000-2022. We identified pregnancies based on diagnosis/procedure codes and mifepristone/misoprostol prescription fills. Algorithms (A1-A4) differed by their handling of discordant pregnancy outcomes: A1 used a hierarchy of outcome types; A2 prioritized procedure codes > diagnosis codes > prescription fills, then outcomes; A3 and A4 refined the priority for code type (delivery procedures > non-delivery procedures > diagnosis codes > prescription fills). Unclear abortion types were grouped as unspecified abortions (A3) or spontaneous abortions (SAB) (A4). Prenatal encounters were linked to pregnancy outcomes, with unlinked encounters grouped and classified as unknown outcomes. We evaluated the distributions of pregnancy outcomes and gestational age at outcome for each algorithm. Agreement of pregnancies identified by A1-A4 with the retrospective algorithm for cases with a ≥1-day overlap was assessed.

Results: The total number of pregnancies identified by A1-A4 were similar (range: 118515-118567), of which 27% had unknown outcomes. After excluding unknown outcomes, 78% were live births, 15% were SAB, and 4% were induced abortions. Distributions of gestational ages at outcome were also similar. For pregnancies with unknown outcomes, 48% were lost to follow up between 9-12 weeks of gestation, 23% between weeks 13-20, and 15% between weeks 21-27.
The retrospective algorithm showed similar distributions of outcomes and gestational ages, with minor differences in the proportion assigned to < 8 weeks vs 9-12 weeks. 68% of A1-A4 pregnancies were matched to 99% of retrospective pregnancies. 85% of unmatched pregnancies from A1-A4 had unknown outcomes. Among A1–A4 pregnancies matched uniquely to a single retrospective pregnancy (1:1 match), agreement was high for live births (>99%) and SAB (93%) but lower for stillbirths (74%) and induced abortions (83%).

Conclusions: The novel algorithms demonstrated high agreement with the retrospective algorithm in identifying pregnancy in claims data, with some differential classification of non-live births that reflect differences in how algorithms use pregnancy outcome codes or handle conflicting data.