Utilizing a Hybrid Apprach to Link Maternal and Neonatal Records

Linkage of independent datasets allows comprehensive and robust analysis. This study aims to utilize a hybrid strategy to link maternal records with neonatal data with an overarching goal of investigating correlates of adverse birth outcomes. To link 126,757 records from Nevada Medicaid with 249,181 maternal records from Birth Registry, a hybrid linkage approach was utilized. Data normalization was first performed for the standardization of linkage keys. First, a deterministic approach was used to link these records using a unique identifier followed by a fuzzy or probabilistic algorithm using a set of block variables. These block variables included date of birth, gender, zip code of birth place, and first/last name of the neonates. For this matching, a minimum threshold of 40 out of 50 was deemed appropriate (10 points for each block variable). After the linkage, data were de-identified to adhere with the ethical guidelines. All linkage procedures were performed via PROC SQL functionality (left join) in SAS 9.4 software. In the first deterministic phase, 63% of the records (n= 80,107 of 126,757 records) were linked (true positives). From remaining 46,650 records, 26,095 were matched through the probabilistic linkage which sum up to 106,202 records in total. Further, 538 records were identified as false positives, which were removed from the dataset. This results in an 83.4% linkage efficiency with a total number of 105,664 linked records. The hybrid linkage offers a viable strategy to improve the linkage efficiency and numerous benefits to population-based research.