(328) Using Claims and Electronic Health Records for Pregnancy Safety Studies: A Targeted Review of Validated Algorithms.

Background: Healthcare administrative claims and electronic health record (EHR) data are increasingly used to evaluate post authorization real-world safety in pregnancy. Pregnancy episodes and outcomes identification can be challenging, and limited information exists regarding the performance of validated algorithms.

Objectives: To describe the performance of validated algorithms for pregnancy episode and outcomes identification in EHR and claims data in the United States (US).

Methods: We searched MEDLINE (search date: January 2024) to identify publications from January 2015 to December 2023, reporting validated algorithms, based on International Statistical Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) diagnosis and procedure codes, developed for identifying pregnancy episodes or outcomes. Search terms were: ("algorithm*” or “ICD-10 code*”) AND (“claim*” or “administrative*” or “electronic health record* data”) AND (“pregnancy*” or “gestational age” or “stillbirth*” or “abortion*”) AND (“United States” or “US*”).
Search output was manually screened by a single reviewer for eligibility and inclusion.
Algorithm development, composition and outcomes were identified and the algorithm performance (validation) was extracted.

Results: Eight studies were included. Algorithms to identify pregnancy episodes or cohorts (n=8), estimate gestational age (GA) (n=7) and identify outcomes (n=8) were included for analysis. Outcomes included live birth (LB) (n=6), full-term birth (FTB) (n=5), preterm birth (PTB) (n=4), stillbirth (SB) (n=8) and spontaneous abortion (SA) (n=7).
Validation approaches varied across studies, including comparisons to national data (n=2), medical charts (n=4) or both (n=2). Validation results were reported with positive predictive values (PPVs) (n=3), agreement percentages (n=4), and/or comparisons of estimates to national data (n=3).
The algorithms consistently demonstrated high agreement for LB and FTB with PPVs ≥90%, while SB and SA showed varying levels of agreement and PPVs between 70.8% to 100%. Relying on ICD-10 codes for PTB identification (n=1) resulted in a percentage agreement of 62.4%, while studies using GA for PTB identification (n=3) reported a potential underestimation of PTB prevalence when compared to national estimates.

Conclusions: The results highlight heterogeneity in the validation performance of algorithms for pregnancy safety evaluation in secondary data. Further efforts are warranted to develop and refine algorithms based on validation performance towards a more accurate identification of GA allowing for a robust pregnancy trimesters treatment exposure and safety outcomes assessment.