Omitted variable bias in single-pollutant epidemiological models for estimating long-term health effects of ambient...

Long-term exposure to PM_2.5 and O₃ is linked to various adverse health outcomes. However, many epidemiological studies on their health effects use single-pollutant models, leading to omitted variable bias (OVB). The percent bias, based on the classical OVB formula, depends on the PM_2.5-O₃ correlation and unobservable true effects. Data were sourced from two recent meta-analyses of the log-linear link between all-cause mortality and per-unit exposure to PM_2.5 or O₃. We then developed a new meta-regression method to correct biases, and its performance was verified through simulation. The OVB for PM_2.5 or O₃ can vary greatly from positive to negative across different spatial scales (like country, sub-national region, or city). By applying this method to 24 individual estimates, we found that a 10 μg/m³ increase in PM_2.5 was linked to a 7.4 % increase in all-cause mortality risk, while O₃'s association with all-cause mortality was not significant, which implies that PM_2.5 must be considered in epidemiological analyses to obtain reliable effect estimates for O₃. Our findings offer novel methodologies for the systematic assessment of the health effects of multiple pollutants. This contribution holds significant potential in fortifying health intervention strategies and minimizing the health risks posed by air pollution to the public.