Probabilistic data linkage: generating a reproductive histories dataset from states’ vital records data

Datasets that follow pregnancy histories over time are lacking. In this pilot study, we used Pennsylvania’s fetal death and birth records to generate a longitudinal maternal dataset by linking records for the same mother to each other. We explored how to best achieve this linkage when lacking a unique record identifier. We demonstrated how Stata’s existing probabilistic matching tools can use nonunique identifiers to facilitate record linkage. To validate the effectiveness of this probabilistic linkage, we compared its results to the linkage results generated from deterministically linking the records using social security numbers. Compared to the deterministic linkage, the probabilistic linkage had a sensitivity of 94.3% and a positive predictive value of 96.7%. Our pilot study can serve as a guide for researchers in other states to generate longitudinal maternal datasets from their states’ vital records. Such longitudinal datasets can be a valuable resource for conducting epidemiologic analyses in the field of maternal and child health and answering research questions that relate to the period between pregnancies. Results from these studies can be used to improve health outcomes of mothers and children