Utilising identifier error variation in linkage of large administrative data sources

Area of Science:

• Health Informatics
• Data Science
• Biostatistics

Background:

• Probabilistic methods for linking administrative data rely on independent identifiers.
• Data quality variations can cause identifier errors, violating independence assumptions and introducing bias.
• This study addresses identifier error variation in Hospital Episode Statistics (HES).

Purpose of the Study:

• To measure identifier error rate variations in HES data.
• To develop and evaluate new methods for match weight calculation incorporating identifier error information.
• To reduce selection bias in linked administrative datasets.

Main Methods:

• Linked 30,000 HES records with Personal Demographic Service (PDS) data.
• Calculated error rates for sex, date of birth, and postcode.
• Used multi-level logistic regression to assess individual and organizational factors influencing errors.
• Derived attribute- and organization-specific match weights.

Main Results:

• Found significant variations in error rates by age, ethnicity, and sex (p < 0.0005).
• Postcode errors were prevalent (53%), while sex and DOB errors were low (0.11%).
• Simulation showed attribute- and organization-specific weights reduced bias compared to traditional methods.

Conclusions:

• Empirical evidence confirms significant identifier error variation in administrative data.
• A novel method for match weight derivation incorporating data variation is proposed.
• Accounting for individual-level characteristics in linkage weights can mitigate bias.