You are here

Optimizing exposure assessment for inference about air pollution effects with application to the aging brain

Principal Investigator: 
,

University of Washington

This study will compare and contrast scientific and logistical benefits of different approaches to air pollution exposure assessment. The investigators will leverage large air pollution datasets obtained from low-cost sensors, mobile monitoring, and passive samplers. They will apply the exposure assessment approaches to determine associations with cognitive decline and dementia incidence in an ongoing cohort study, Adult Changes in Thought Air Pollution (ACT-AP).

Funded under
Status: 
Ongoing
Abstract

Poster abstract for HEI Annual Conference 2023

Impact of Mobile Monitoring Network Design on Inference About the Association Between Ultrafine Particulate Matter and Cognitive Function

Magali N. Blanco,1 Adam A. Szpiro,2 Lianne Sheppard1,2

1Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, WA USA; 2Department of Biostatistics, University of Washington, Seattle, WA USA

Background: Short-term mobile monitoring campaigns are increasingly used to assess exposures to less traditionally monitored traffic-related air pollutants (TRAP) including ultrafine particulates (UFP). Monitoring designs, however, vary substantially, and it is unclear how the selected designs impact subsequent epidemiologic inferences. The objective of this study was to investigate the degree to which different mobile monitoring designs impact inference about the association between UFP and cognitive function.

Methods: We leveraged an extensive mobile monitoring campaign that was conducted in the greater Seattle to assess unbiased annual average TRAP exposures for the Adult Changes in Thought (ACT) cohort, a study investigating the aging brain. Monitoring was conducted over the course of a year during all days of the week and most hours of the day. Data for this study consisted of 7,806 total stops (278 sites × ~28 visits each). We developed universal kriging – partial least squares (UK-PLS) UFP exposure prediction models using all the data as well as subsamples following common sampling designs (30 campaigns each) including: sampling fewer total stops (3,336 and 1,800) with no additional temporal restrictions; sampling fewer stops (3,336) during business or rush hours only; and sampling fewer stops (3,336) during 1-3 seasons only. We used the resulting exposure predictions to estimate the association between five-year UFP exposure and baseline cognitive function (as determined by the Cognitive Abilities Screening Instrument – Item Response Theory [CASI-IRT]) in the ACT cohort (N=2,184), adjusted for age, calendar year, sex, education, and apolipoprotein E (APOE) status.

Results: While predictions from the all-data campaign performed well (R2=0.8), most restricted designs performed slightly lower (median R2~0.7), with the business hour design performing particularly poorly (R2=0.4). Using the all-data campaign exposure model, the adjusted mean baseline CASI-IRT score decreased by -0.014 (95% CI: -0.027, -0.001; SE: 0.006) for every additional 1,000 pt/cm3 UFP. Exposure models with fewer stops or sampling seasons produced median (IQR) health effect estimates that were 0.002 (0.00-0.005) units higher, or with 17% (1-33%) error. Business and rush hour designs produced higher errors of 66% (60-73%) and consistently underestimated the magnitude of the all-data effect estimate. Unlike the biased business and rush hour designs, the fewer stop and season designs produced effect estimates that were broadly consistent with the all-data campaign and resulted in statistically significant associations between PNC and CASI-IRT 47% of the time.  

Conclusions: While the all-data campaign detected a significantly negative association between UFP exposure and CASI-IRT, reduced sampling design results were variable and only detected this association some of the time. Common business and rush hour designs never detected this association, potentially due to bias from systematic measurement error. TRAP exposure monitoring design is critical for accurate and well-calibrated epidemiologic inference.