Estimating the accuracy of geographical imputation.

Area of Science:

• Epidemiology
• Geographic Information Systems (GIS)
• Biostatistics

Background:

• Researchers often geocode cases to postal code centroids to supplement precisely geocoded street addresses.
• Assigning finer-level geographic data, like census tracts, to these cases is typically done using postal centroids or geographical imputation (geo-imputation).
• Geo-imputation uses case demographics and postal area population characteristics for location assignment, but its accuracy has not been systematically evaluated.

Purpose of the Study:

• To evaluate the accuracy of census tract assignment for cancer cases using geo-imputation methods.
• To compare the performance of geo-imputation against traditional postal code centroid assignment.

Main Methods:

• Utilized a large dataset of breast, prostate, and colorectal cancer cases from the New Jersey Cancer Registry.
• Compared census tract assignment accuracy using demographic-based geo-imputation versus postal code centroid assignment.
• Measured accuracy by comparing the assigned census tract to the tract originally identified from the full street address.

Main Results:

• Demographic-based geo-imputation, using race/ethnicity, resulted in higher correct census tract assignment rates than postal code centroids.
• Incorporating age characteristics further improved match rates.
• Accuracy was significantly influenced by the geographic distribution of race/ethnicity and population density.

Conclusions:

• Geo-imputation offers advantages over postal code centroid assignment for census tract allocation without significant drawbacks.
• Researchers should consider the impact of geocoding quality and potential geographical bias in their analyses.
• Geo-imputation presents a viable alternative for improving geographic data accuracy in epidemiological studies.