A mixture model for the analysis of data derived from record linkage.

Area of Science:

• Statistics
• Biostatistics
• Data Linkage

Background:

• Combining data from multiple sources is crucial for comprehensive analysis.
• Accurate record linkage is essential, especially when unique identifiers are unavailable.
• Traditional methods risk introducing bias due to incorrect record matching.

Purpose of the Study:

• To develop a robust statistical model for record linkage without unique identifiers.
• To address the challenge of potential misclassification in partially identified records.
• To improve the accuracy of regression analyses following data linkage.

Main Methods:

• Proposed a mixture model treating record match status as missing data.
• Utilized a pairwise pseudo-likelihood approach maximized via an expectation-maximization algorithm.
• Developed a procedure to efficiently handle large numbers of potential record matches.

Main Results:

• Simulations demonstrated accurate estimation across various linkage scenarios.
• The proposed method outperformed existing approaches, avoiding bias in challenging cases.
• Successfully applied to estimate associations between pregnancy durations using linked maternal data.

Conclusions:

• The novel mixture model offers a statistically sound and accurate solution for record linkage.
• This approach mitigates bias and enhances the reliability of analyses using linked datasets.
• The method is effective even in complex scenarios and real-world applications.