The Data-Adaptive Fellegi-Sunter Model for Probabilistic Record Linkage: Algorithm Development and Validation for...

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Comprehensive patient care relies on integrated health data.
• Missing data and field selection are key challenges in patient record linkage.
• Accurate patient matching is crucial for data quality and care.

Purpose of the Study:

• To evaluate the impact of the missing at random (MAR) assumption on the Fellegi-Sunter model for patient-record linkage.
• To assess the benefits of data-driven field selection in enhancing patient-matching accuracy.
• To improve the performance of patient-record linkage in real-world healthcare applications.

Main Methods:

• Adapted the Fellegi-Sunter model to incorporate the MAR assumption for handling missing data.
• Compared MAR-based adaptation against traditional methods (treating missing as disagreement).
• Evaluated performance across four diverse use cases including HIE deduplication and public health linkages using standard metrics (F1-score, sensitivity, etc.).

Main Results:

• Incorporating the MAR assumption maintained or improved F1-scores, irrespective of field selection method (expert vs. data-driven).
• The combination of the MAR assumption and data-driven field selection led to optimized F1-scores across all four use cases.
• This approach demonstrated improved patient-matching accuracy in complex, real-world data scenarios.

Conclusions:

• The MAR assumption is a valid and effective strategy for real-world patient record linkage.
• Integrating MAR with data-driven field selection offers the best performance for patient matching.
• This optimized approach is particularly beneficial for privacy-preserving record linkage applications.