[Assessment of a method for automatic match classification in probabilistic data linkage].

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Probabilistic data linkage is crucial for integrating health databases but often requires manual review.
• Manual inspection is time-consuming and resource-intensive, hindering large-scale data analysis.
• Automating true match identification in data linkage can enhance efficiency and accuracy.

Purpose of the Study:

• To assess the accuracy of a scoring method for probabilistic data linkage.
• To enable automatic identification of true matches, reducing the need for manual inspection.
• To evaluate the method's performance using real-world breast cancer data.

Main Methods:

• Utilized data from the Breast Cancer Information System (SISMAMA) in Minas Gerais, Brazil (2009-2010).
• Performed a 16-step probabilistic linkage after data cleaning and standardization.
• Manually inspected steps to establish a gold standard, then assessed sample accuracy.

Main Results:

• High sensitivity (>0.97), precision (>0.95), accuracy (>0.97), and F-measure (>0.96) were observed for samples with 200-300 matches.
• A reduced set of three linkage steps identified 99.24% of true matches in the total database.
• The method demonstrated high performance metrics, including area under the curve precision-recall (>0.98).

Conclusions:

• The proposed scoring method enables accurate, automatic identification of true matches in probabilistic data linkage.
• This automation significantly streamlines health data management and surveillance processes.
• The validated method facilitates the broader application of probabilistic linkage in healthcare settings.