Mitigating Omitted Variable Bias in Air Pollution Health Risk Assessment: A Unified Multi-Pollutant Framework

Background: Isolating the independent health effects of atmospheric constituents remains a challenge due to their complex physicochemical coupling. Traditional single-pollutant models frequently neglect these correlations, leading to systematic Omitted Variable Bias (OVB) and distorted disease burden estimates.

Methods: I introduces a novel "post-hoc adjustment meta-regression" framework to quantify and correct OVB. By integrating extensive epidemiological data with high-resolution global atmospheric reanalysis products, the approach utilizes location-specific pollutant correlations to retrieve unbiased causal estimates.

Results: Applying this framework across varying temporal scales and chemical components revealed that single-pollutant models consistently overestimate health risks. Specifically, correcting for OVB in short-term PM_2.5 and ozone co-exposures reduced the estimated global mortality burden by approximately 16%. In long-term assessments, unadjusted models were found to inflate ozone risk estimates due to confounding by PM_2.5. Furthermore, for specific chemical constituents, neglecting non-Black Carbon (BC) components exaggerated BC's mortality risk by a median of 147%, obscuring its true, albeit higher, intrinsic toxicity relative to other particulate matter. Some of the case studies have been published after peer review.

Conclusions: OVB introduces significant, pervasive errors in current epidemiological syntheses. This unified multi-pollutant correction framework provides a robust solution for refining health impact assessments, underscoring the necessity of accounting for co-pollutant confounding in future air quality policy-making.