(317) The Impact of the Mother-Infant Linkage Process in US Health Insurance Claims Data on the Representativeness of the Infant Sample

Background: Health insurance claims databases are often used to study drug safety in pregnancy in relation to infant outcomes. Proper linkage between maternal and infant records is essential for such investigations. Few studies have explored whether there are systematic differences between linked and unlinked infant sub-populations potentially resulting in generalizability concerns.

Objectives: To compare the demographic and clinical characteristics, healthcare resource utilization, and health outcomes of linked and unlinked infant cohorts.

Methods: We used Merative MarketScan® Commercial Claims (CCAE) and Optum’s de-Identified Clinformatics® Data Mart databases (Clinformatics®) (2000-2023), transformed into OMOP CDM v5.4, to identify pregnancy episodes and link maternal and infant records (i.e., linked infants) via published algorithms. To identify an unlinked infant cohort with appropriate time zero, we curated a set of diagnosis and procedure codes as ‘birth markers’ to overcome the lack of exact birth date data in our databases. The birth markers were required within ± 60 days of the insurance enrollment start date for infants with age < 1. The cohort entry date was set as the earliest birth marker in both linked and unlinked cohorts. We compared the linked and unlinked cohorts using demographics, clinical, and pharmacy data at day 0, 1-30-, and 31-365-day time windows with standardized mean differences. Specific health outcomes of interest included prematurity, small for gestational age, serious infection, postnatal growth deficiency, infant developmental delay, and major congenital malformations (MCM).

Results: We identified 4.3 million unlinked infants (CCAE:2.6, Clinformatics®:1.7) and 4.6 million linked infants (2.7, 1.9). Among unlinked infants, 64% and 62% had birth markers in CCAE and Clinformatics®, respectively. The unlinked cohort and linked cohort had a similar proportion of females in both databases (48% and 49%) and similar racial/ethnic distribution (Clinformatics®); although less insurance enrollment post-birth (days as mean; CCAE: 950 vs. 1111 and Clinformatics®: 751 vs. 867). The unlinked cohort and linked cohort appeared similar across medical code-level characteristics and specific outcomes of interest (e.g., prematurity (%): 8.7 vs. 8.5 in CCAE and 9.6 vs 9.7 in Clinformatics®; MCM (%): 5.0 vs. 5.2 and 4.8 vs 5.0). The unlinked infants without birth markers were significantly different from the unlinked cohort with birth markers.

Conclusions: Our findings showed linkage process in US commercial health insurance claims data does not appear to significantly affect the representativeness of the infant sample. Further research is needed to understand if differential follow-up time may impact long-term pediatric studies.