A systematic review of validated methods for identifying atrial fibrillation using administrative data

Area of Science:

• Health Informatics
• Clinical Data Analysis
• Medical Algorithm Validation

Background:

• Electronic health records (EHR) offer vast potential for identifying patients with atrial fibrillation (AF).
• Validating algorithms used for AF detection in EHR is crucial for clinical accuracy.
• Previous research has focused on specific data sources and timeframes.

Purpose of the Study:

• To systematically review and characterize the validity of algorithms identifying AF from electronic health data.
• To identify research gaps in the current literature regarding AF detection algorithms.

Main Methods:

• Systematic literature review of publications from 1997-2008.
• Two reviewers assessed studies providing validation data for AF identification algorithms.
• Extracted data included algorithm sensitivity, specificity, and positive predictive value (PPV).

Main Results:

• 16 unique studies provided validation information from 544 reviewed abstracts.
• ICD-9 code 427.31 demonstrated a high PPV (median 89%) for prevalent AF.
• Algorithm sensitivity ranged from 57% to 95% (median 79%), with limited data on incident AF.

Conclusions:

• The ICD-9 code 427.31 is a relatively effective tool for identifying prevalent AF.
• Limitations include dated data, non-representative populations, and a focus on inpatient data.
• Future algorithms should integrate inpatient/outpatient codes and electronic ECG data for improved accuracy in contemporary populations.