Details
The U.S. Food and Drug Administration (FDA) initiated this study in the Sentinel System to improve methods for evaluating the safety of medical product use during pregnancy. Accurate identification of pregnancies and estimation of gestational age are critical for assessing prenatal drug exposure. However, data used to identify pregnancies in administrative claims databases generally lack precise indicators of pregnancy start dates, which might lead to wrong conclusions about prenatal drug exposure.
To address these limitations in administrative claims databases, FDA pursued a descriptive analysis to determine whether pregnancy identification could be enhanced by capturing pregnancies without using pregnancy outcome codes (e.g., a code for a live birth) as starting points. This study focused primarily on two analyses. First, certain pregnancy marker codes (including fertility codes, prenatal test codes, and gestational age codes) sometimes are seen in administrative claim records that don’t contain codes for pregnancy outcomes. We evaluated the frequency with which such isolated codes occurred. Second, certain of these pregnancy marker codes are not prioritized by the current Sentinel pregnancy algorithm for determining pregnancy duration and start date. We evaluated the frequency of occurrence of these "lower priority codeswithin pregnancies that were identified based on codes describing pregnancy outcomes.
Over 3.97 million females aged 10–54 years with at least one isolated pregnancy marker code in the Merative™ MarketScan® Research Databases were identified between October 2015 and March 2023, including 3.4 million females (86%) without any pregnancy identified. Among the 8.59 million isolated pregnancy marker codes, more than half (52%) occurred in females who remained in the database for over nine months after the code was recorded. The most frequent isolated code was Current Procedural Terminology, Fourth Edition (CPT-4) code 87081 (Group B streptococcus screening), which comprised 4.57 million codes and was non-specific for pregnancy. Additionally, among 2.08 million live birth pregnancy episodes with at least one pregnancy marker code recorded, there were over 25 million lower-priority pregnancy marker codes not used for pregnancy duration estimation, including gestational age codes in 1.57 million episodes (75%) and fertility codes in 44,312 episodes (2%).
The Sentinel study results enhanced FDA’s understanding of the complexities involved in identifying pregnancies and estimating pregnancy start date using administrative claims data. The findings informed discussions on potential methodological improvements to pregnancy identification approaches, but changes to the current algorithm based on these findings will not be implemented at this time. As the FDA pursued this study for the purpose of methods and development, no regulatory action was needed following this study.