Extending the Fellegi-Sunter probabilistic record linkage method for approximate field comparators

Area of Science:

• Data Science
• Biostatistics
• Health Informatics

Background:

• Probabilistic record linkage is crucial for identifying duplicate demographic records.
• The Fellegi-Sunter method, using log likelihood ratios, is a standard approach.
• Existing methods can be limited by exact data matching.

Purpose of the Study:

• To introduce an extension of the Fellegi-Sunter method incorporating approximate field comparators.
• To evaluate the performance of this extension in identifying duplicate records.

Main Methods:

• The extended Fellegi-Sunter method was developed using approximate field comparators.
• A case study was conducted using a large academic medical center's data warehouse.
• Performance was compared against the standard Fellegi-Sunter method using various fields and cutoffs.

Main Results:

• The approximate comparator extension misclassified 25% fewer pairs compared to the standard method.
• A larger Welch's T statistic was observed for the extension across all field sets and cutoffs.
• Accuracy gains increased with less information and higher matching cutoffs.

Conclusions:

• The approximate comparator extension offers a significant improvement in probabilistic record linkage accuracy.
• This enhancement has substantial potential impact in clinical and research data linkage.
• The method is particularly beneficial when data is incomplete or matching criteria are stringent.