Comparison of different exposure settings in a case--crossover study on air pollution and daily mortality: counterintuitive results

Abstract

Because of practical problems associated with measurement of personal exposures to air pollutants in larger populations, almost all epidemiological studies assign exposures based on fixed-site ambient air monitoring stations. In the presence of multiple monitoring stations at different locations, the selection of them may affect the observed epidemiological concentration--response (C-R) relationships. In this paper, we quantify these impacts in an observational ecologic case--crossover study of air pollution and mortality. The associations of daily concentrations of PM(10), O(3), and NO(2) with daily all-cause non-violent mortality were investigated using conditional logistic regression to estimate percent increase in the risk of dying for an increase of 10 μg/m(3) in the previous day air pollutant concentrations (lag 1). The study area covers the six main cities in the central-western part of Emilia-Romagna region (population of 1.1 million). We used four approaches to assign exposure to air pollutants for each individual considered in the study: nearest background station; city average of all stations available; average of all stations in a macro-area covering three cities and average of all six cities in the study area (50 × 150 km(2)). Odds ratios generally increased enlarging the spatial dimension of the exposure definition and were highest for six city-average exposure definition. The effect is especially evident for PM(10), and similar for NO(2), whereas for ozone, we did not find any change in the C-R estimates. Within a geographically homogeneous region, the spatial aggregation of monitoring station data leads to higher and more robust risk estimates for PM(10) and NO(2), even if monitor-to-monitor correlations showed a light decrease with distance. We suggest that the larger aggregation improves the representativity of the exposure estimates by decreasing exposure misclassification, which is more profound when using individual stations vs regional averages.