Evaluating the importance of historical address histories in an air pollution and dementia study

Background Evidence linking long‐term air pollutant exposures with dementia‐related risks is increasing. Changes in the brain may occur decades before diagnosis, and studies assessing the effects of air pollution on dementia should account for participant residential history to reduce exposure misclassification. Using the Ginkgo Evaluation of Memory Study (GEMS), we evaluated cumulative PM 2.5 and NO 2 exposure on dementia risk both with and without historical residential addresses. Method GEMS was a randomized trial of 3,069 participants in Hagerstown MD, Pittsburgh PA, Sacramento CA, and Winston‐Salem NC from 2000‐2008. Residential historical addresses were obtained through LexisNexis, a company constructing address histories from public databases. PM 2.5 (1980‐1999) and NO 2 (1990‐1999) cumulative exposures were calculated for each participant from a spatio‐temporal model in two ways: 1) using residential address histories reconstructed from LexisNexis addresses from 1980‐1999 (Figure 1), and 2) assuming participants lived from 1980‐1999 at the address documented during the GEMS baseline visit (2000‐2002). Hazard ratios from Cox models of all‐cause dementia, Alzheimer’s disease (AD), and mixed vascular dementia (VaD/AD) were compared across both sets of cumulative PM 2.5 and NO 2 estimates. Models were adjusted for age, study site, randomization year, gender, treatment group, neighborhood deprivation index (NDI), education, smoking status, smoking pack years, passive smoking exposure and apolipoprotein E 4 (APOE4) status . Result Hazard ratios based on the LexisNexis historical addresses were generally larger than those from analyses without residential histories (Figure 2). The largest attenuation in effect after eliminating address history was for participants with vascular dementia either with or without AD (VaD/AD). An approximate interquartile range (IQR; 3 µg/m3) increase in 10 year mean NO 2 exposure was associated with a 39% (95% CI: 7%, 81%) greater risk of mixed VaD/AD, but this decreased to 25% (95% CI: ‐7%, 66%) without histories. Similarly, an IQR (2 µg/m3) increase in 20 year PM 2.5 was associated with a 19% (95% CI: 4%, 36%) greater risk of all cause dementia compared to 14% (95% CI:‐1%, 31%) without the historical addresses. Conclusion For epidemiological studies aiming to quantify long term place‐based exposures related to dementia risk, utilizing residential address databases may reduce misclassification in exposure assignments.