Linking Biomedical Data Warehouse Records With the National Mortality Database in France: Large-scale Matching...

Area of Science:

• Biomedical Informatics
• Data Science
• Public Health

Background:

• Vital status is crucial for biomedical data warehouse (BDW) value in research but often missing or uncertain.
• The French National Mortality Database (FNMD) provides open-source death records, but matching it with BDWs presents challenges like lack of unique identifiers and data inconsistencies.

Purpose of the Study:

• To develop and evaluate a novel algorithm for matching BDW records with the FNMD.
• To improve the accuracy and efficiency of linking large-scale health databases.

Main Methods:

• Developed a deterministic algorithm utilizing advanced data cleaning and the Damerau-Levenshtein distance (DLD).
• Assessed algorithm performance (sensitivity and specificity) using data from three university hospitals (Lille, Nantes, Rennes).
• Compared the DLD-based algorithm against a direct matching algorithm with minimal data cleaning.

Main Results:

• The DLD-based algorithm achieved 11% higher sensitivity (93.3%) compared to the direct algorithm (82.7%).
• Higher sensitivity was observed for men and patients born in France, with significant variations across centers.
• Specificity exceeded 98% across all subgroups; the algorithm efficiently processed millions of records using parallel computing and low RAM.

Conclusions:

• The DLD-based algorithm significantly enhances the sensitivity of matching BDW records with mortality databases.
• Advanced data cleaning and name-matching techniques like DLD are vital for improving data linkage accuracy.
• The developed algorithm and associated Inseehop R package offer a scalable, secure solution for linking large datasets, improving BDW utility.