Comparing methodologies for imputing ethnicity in an urban ophthalmology clinic

Area of Science:

• Public Health
• Biostatistics
• Ophthalmology

Background:

• Accurate patient ethnicity data is crucial for understanding health disparities and tailoring clinical care.
• Traditional methods for ethnicity imputation may lack precision, impacting research validity and health equity initiatives.

Purpose of the Study:

• To compare the accuracy of three distinct methodologies for imputing patient ethnicity within an urban ophthalmology clinic setting.
• To evaluate surname-based, geocoding-based, and a combined surname-geocoding approach for ethnicity imputation.

Main Methods:

• Utilized data from 19,165 patients, including self-reported ethnicity, surnames, and home addresses.
• Compared a surname method (2000 US Census data), a geocoding method (2010 US Census tract data), and a combined method employing Bayes' theorem.

Main Results:

• The combined surname-geocoding method demonstrated the highest accuracy across all ethnic groups, with strong overall agreement (κ = 0.76).
• Sensitivity and positive predictive values (PPV) for the combined method ranged from 71% (Hispanic) to 94% (Black) for PPV and 77% (Hispanic) to 92% (White) for sensitivity.
• Individual methods showed lower agreement: surname (κ = 0.23) and geocoding (κ = 0.58).

Conclusions:

• A combined methodology integrating surname analysis and Census tract data via Bayes' theorem significantly outperforms other tested imputation methods.
• This superior imputation technique is highly suitable for research applications utilizing clinical and administrative patient data.