Probabilistic record linkage is a valid and transparent tool to combine databases without a patient identification...

Area of Science:

• Public Health
• Biostatistics
• Health Informatics

Background:

• Population-based perinatal registries are crucial for public health surveillance.
• Linking disparate health databases is challenging due to the absence of unique identifiers.
• The Dutch perinatal registries (LVR1, LVR2, LNR) lack a common unique identification number.

Purpose of the Study:

• To detail the technical approach for probabilistic linkage of three Dutch perinatal registries.
• To validate the accuracy and effectiveness of the developed linkage methodology.
• To assess the feasibility of linking anonymous health data without unique identifiers.

Main Methods:

• Employed a combination of probabilistic and deterministic record linkage techniques.
• Utilized maternal, delivery, and child information for record matching.
• Developed specific algorithms to handle challenges like twin births and data errors.

Main Results:

• The probabilistic linkage procedure successfully integrated the three Dutch perinatal registries.
• Validation confirmed the effectiveness of the linkage despite inherent data errors.
• A high-quality linked database was created from the raw registry data.

Conclusions:

• Probabilistic linkage is a robust method for creating comprehensive health databases from unlinked sources.
• The developed techniques are broadly applicable to linking health data with partial identifiers.
• This approach is valuable for studies involving overlapping cohorts, multiple births, and incomplete data.