The Associations Between Sensitivity and Specificity With Prevalence in Data Matching.

Area of Science:

• Health Informatics
• Biostatistics
• Data Science

Background:

• Accurate data matching is crucial for health research and public health surveillance.
• Understanding how population prevalence influences the performance of data matching algorithms is essential for reliable record linkage.

Purpose of the Study:

• To investigate the relationship between the prevalence of a health outcome in a source population and a study population with the sensitivity and specificity of data matching.
• To assess the impact of varying prevalence on the accuracy of probabilistic record linkage.

Main Methods:

• A synthetic dataset of 8 million records was created with simulated health outcomes, sex, and birth dates.
• Probabilistic record linkage using Link Plus 3.0 was performed on multiple synthetic "disease registry" and "study population" file pairs with varying prevalence.
• Sensitivity and specificity were calculated based on the linkage results across different prevalence scenarios.

Main Results:

• Sensitivity in data matching increased from 80.0% to 94.6% as source population prevalence rose from 0.1% to 10%.
• Specificity slightly decreased with increasing source population prevalence.
• Sensitivity and specificity remained largely stable (around 95.0% and 100.0% respectively) as study population prevalence increased from 10% to 99%.

Conclusions:

• Sensitivity is positively associated with source population prevalence, while specificity is negatively associated in data matching.
• Prevalence within the study population itself does not appear to significantly influence sensitivity or specificity.
• These findings highlight the importance of considering source population characteristics when evaluating data matching performance.