Associate Director Janssen Research & Development, United States
Background: Studies that use health care data to assess infant outcomes following maternal medication use in pregnancy require linkage of maternal and infant records. While such linkage is increasing among established and emerging data sources, linkage methods are heterogeneous and have not been thoroughly reviewed or evaluated to inform fitness for use.
Objectives: To describe methods used to link maternal and infant records in routinely collected health care databases.
Methods: Multiple PubMed searches were conducted using structured terms and known mother-infant linked data sources. Reference sections of identified key papers were also reviewed. From each publication, we abstracted information on the type(s) of data sources, data fields used for the linkage, linkage methodology, and any results of evaluations of the linkage.
Results: We reviewed 105 publications describing mother-infant linkages published from 1990 to 2023 that used data from North and South America, Europe, Asia, and Oceania. Linkage methods using administrative claims data only primarily relied on the following fields: a family ID and estimated dates of maternal delivery and infant birth; the proportions of mothers linked with infants ranged from 56-90%. Use of vital records, such as birth certificates, alongside claims data resulted in higher linkage proportions (>86% of deliveries); this has been done in Taiwan, Europe, and several US states. Utilization of birth records together with hospital discharge data resulted in linkages of over 90% of deliveries. Linkages involving additional types of data, such as those using electronic health records, were identified in Europe and the US. Mother-infant linkages were straight-forward in regions with national birth registers that natively include identifiers for both mother and infant, such as the Nordic countries, as well as within integrated delivery networks in the US that provide both health insurance and care. Although most mother-infant linkages were conducted using deterministic methods, probabilistic methods typically achieved greater linkage proportions while requiring more data fields. Validations of data linkage algorithms were uncommon; two studies demonstrated high positive predictive value (>95%) but lower sensitivity (20-88%).
Conclusions: Successful mother-infant data linkage occurs throughout the world and is enabled by aligned health care and technology infrastructure and policy. To increase trust in evidence generated from linked data, we encourage researchers to include in their publications a description of linking methods, a quantification of linkage success, and, when possible, an assessment of linkage validity.
