Development and validation of data quality rules in administrative health data using association rule mining

Area of Science:

• Health Informatics
• Data Quality Management
• Clinical Coding

Background:

• Assessing data quality in administrative health data is crucial for research integrity.
• Coded diagnostic data presents unique challenges for quality assessment.

Purpose of the Study:

• To develop and validate a set of coding association rules for International Classification of Disease, 10th revision (ICD-10) coded diagnostic data.
• To establish a tool for monitoring and comparing ICD-10 data quality.

Main Methods:

• Association rule mining was performed on re-abstracted Canadian hospital discharge data (ICD-10).
• Rules were extracted at three-digit and four-digit levels across four age groups.
• A modified Delphi process involving physicians and classification specialists validated the rules.

Main Results:

• 388 three-digit and 275 four-digit ICD-10 coding association rules were developed.
• Rules demonstrated meaningful age-specific clinical associations, particularly in the ≥65 age group.
• Variance and bias metrics were defined to identify data quality issues.

Conclusions:

• A validated set of ICD-10 data quality rules was created.
• These rules serve as a valuable tool for assessing and monitoring the quality of ICD-coded health data.
• The rules facilitate data quality comparisons across different settings and jurisdictions.