Local approach to attributable disease burden: case study for air pollution and mortality in Belgium

Abstract Comparative risk assessment is commonly applied to derive the share of the burden of disease attributable to specific risk factors, including air pollution. This method is usually conducted at national level, meaning areas and population groups within the country cannot be compared. We propose a novel approach, where the burden attributable to air pollution is derived locally. Mortality in Belgium due to long-term exposure to particulate matter with diameter <2.5 µm (PM2.5) and nitrogen dioxide (NO2) is derived for the year 2019. In the local method, the attributable burden is calculated at the statistical sector, the smallest territorial unit. Results for individual sectors with small populations are potentially biased when using a concentration-response function (CRF) for the general population. Therefore, the local method is validated by comparing the results, summed to the Belgian total, to estimates derived with a ‘global’ national-scale approach. The discrepancy between the two methods is compared with the uncertainty related to exposure (5th and 95th concentration percentiles) and to the CRF (95% confidence interval of the relative risk), where the central global estimate acts as a baseline. The difference due to methods is limited to under 2% for both pollutants: 8665 deaths derived globally vs. 8588 (-77) locally for PM2.5, and 3688 globally vs. 3633 (-55) locally for NO2. For PM2.5, the method discrepancy and exposure uncertainty (-128, +236) are comparable in magnitude, while both are severely outweighed by CRF uncertainty (-2039, +990). For NO2, the exposure uncertainty (-300, +735) is substantially higher, which could be expected given its greater spatial variability, although CRF uncertainty (-1819, +3495) is again greatest. The local attributable burden method shows potential for comparing areas and population groups at subnational level, on the condition that the results are aggregated to a sufficiently large scale to compensate for possible bias. Key messages • A local burden of disease method offers possibilities for comparing areas and population groups at subnational level. • Validation by comparison with global results shows that potential local bias is mitigated after sufficient aggregation.