Comparison of Long-Term Air Pollution Exposure from Mobile and Routine Monitoring, Low-Cost Sensors, and Dispersion...

Area of Science:

• Environmental Epidemiology
• Air Pollution Exposure Assessment
• Geospatial Health Analysis

Background:

• Accurate assessment of long-term outdoor air pollution exposure, particularly traffic-related pollutants like ultrafine particles (UFPs), black carbon (BC), and nitrogen dioxide (NO2), is crucial for epidemiological studies.
• Existing exposure assessment methods, including land use regression (LUR) models, low-cost sensor networks, mobile monitoring, and dispersion models, vary widely, leading to potential heterogeneity in health effect estimates.

Purpose of the Study:

• To develop and compare long-term ambient air pollution exposure estimates using diverse methods: low-cost sensors, mobile and fixed-site monitoring, and dispersion modeling.
• To evaluate the predictive performance of different exposure assessment methods for spatial variation using external validation data.
• To compare the resulting air pollution health effect estimates from various exposure models within epidemiological studies.

Main Methods:

• Evaluated annual average concentrations of UFPs, NO2, BC, and PM2.5 across the Netherlands using empirical LUR (SLR, Random Forest, LASSO) and deterministic dispersion models.
• Compared model predictions at 20,000 addresses and validated performance using external data from 2021-2023 and historical datasets.
• Conducted epidemiological analyses in three cohort studies to assess associations between different exposure model estimates and mortality, stroke, coronary events, lung function, and asthma incidence.

Main Results:

• Exposure predictions for BC, NO2, and PM2.5 were highly correlated across different years (2010-2019), indicating stable spatial patterns.
• Models generally showed moderate to high correlations in predicting historical exposure patterns (>10 years), especially for BC and NO2.
• While different models yielded similar conclusions on health associations, the magnitude of effect estimates varied substantially, contributing to heterogeneity in systematic reviews.

Conclusions:

• Differences in exposure assessment methods can significantly contribute to the heterogeneity of health effect estimates observed in epidemiological research.
• Model performance validation and predicted exposure contrast are key factors influencing the heterogeneity of effect estimates.
• No consistent performance differences were found between empirical LUR algorithms (SLR, Random Forest, LASSO) or between mobile, dispersion, and fixed-site LUR models.