Evaluation of a commercial database to estimate residence histories in the Los Angeles Ultrafines Study

Abstract

Background. Completeness of residential histories in epidemiologic studies influence the accuracy of environmental exposures estimated at residential locations. Commercial databases can be useful in locating participants over time; however, the quality of such information is uncertain. Methods. We conducted a commercial database address search (using name and social security numbers) for participants of the NIH-AARP Diet and Health Study cohort (aged 50-71 years) living in Los Angeles (Los Angeles Ultrafines Study; N=50,260). We compared participants with and without commercial addresses(s) at two times (enrollment: 1995-1997 and follow-up: 2004-2005) across strata of participant characteristics. A commercial match to self-reported address was determined by two criteria: ≤250m geocoded distance between addresses and ≥80% string agreement for attributes: street number, street name, city, ZIP code, and state. For both self-reported and commercial addresses at enrollment, we estimated two important predictors of ultrafine particles in a land use regression model, nitrogen dioxide (NO2;ppm) and high intensity developed land (HIDL;%) within 5km. We compared estimates of NO2 and HIDL (Spearman’s rho) and changes to exposure classifications (quintiles) between commercial and self-reported addresses. Results. The commercial database identified 69% of participants at enrollment and 95% of those still in California who participated at follow-up (N=23,669). Percentages were similar across sex, race, and education categories and were ≥5% higher for individuals aged 65+ years and never smokers. Most participants had a match to their self-reported addresses, with similar match rates for enrollment (86%) and follow-up (82%). When using commercial versus self-reported addresses, correlations were high for estimated NO2 (rho=0.90) and HIDL (rho=0.92). Only 8% of participants were classified into different quintiles for both predictors. Conclusions. Our results demonstrate the potential of commercial databases to provide missing residence information. Future studies should consider how time period, geographic location, and population characteristics may influence the accuracy of information obtained.