Probabilistic linkage to enhance deterministic algorithms and reduce data linkage errors in hospital administrative...

Area of Science:

• Health Informatics
• Data Linkage
• Patient Record Systems

Background:

• The pseudonymisation algorithm for linking patient care episodes in England (HESID) lacked formal evaluation for data linkage errors.
• Existing deterministic algorithms may not accurately link all patient records.

Purpose of the Study:

• To evaluate improvements in data linkage accuracy by incorporating probabilistic linkage into the existing deterministic HESID algorithms.
• To quantify the reduction in data linkage errors and their impact on patient data.

Main Methods:

• Utilized 17 years of inpatient hospital admission data (1998-2015) from England's Hospital Episode Statistics (HES).
• Compared the existing deterministic HESID algorithm with a hybrid approach including an additional probabilistic step.
• Established a reference standard using enhanced probabilistic matching with additional clinical and demographic data.

Main Results:

• The HESID algorithm demonstrated a high missed match rate, decreasing from 8.6% in 1998 to 0.4% in 2015.
• Missed matches disproportionately affected ethnic minorities, deprived populations, foreign patients, and those without a fixed abode.
• Probabilistic linkage reduced bias in hospital readmission rate estimates for most patient groups.

Conclusions:

• Probabilistic linkage enhances HES data accuracy by reducing missed matches and correcting bias in readmission rate estimates.
• Modifying the current HESID algorithm is recommended to address data linkage errors.
• A retrospective update of existing data is necessary to rectify past linkage errors and their consequences.