Bias in the estimated association between all-cause mortality and long-term exposure to a specific chemical component of fine particulate matter: The example of black carbon

Introduction: Long-term exposure to fine particulate matter (PM_2.5) has been linked to many adverse health outcomes, which can vary significantly depending on the chemical profile of the PM_2.5. However, many meta-analyses of the health effects of a specific component of PM_2.5 have ignored the effects of other components, leading to omitted variable bias (OVB). This study developed a new method to address this problem and conducted a simulation using black carbon (BC) as an example. Method: We used data from two published meta-analyses as input for our model, with supplementary information obtained from a reanalysis product of PM_2.5 components. Based on the classical OVB formula, we developed a post hoc adjusted model and verified its performance via a simulation study. We obtained pooled estimates of the effect of BC on all-cause mortality, with adjustment for the effect of non-black carbon (NBC) components. Finally, based on the estimated effects of BC and NBC, we investigated global patterns in PM_2.5 toxicity (i.e., the per-unit effect of PM_2.5) and the degree of OVB associated with ignoring the differential effects of BC and NBC. Results: The post hoc adjusted model included 46 individual estimates of the effects of BC or NBC on all-cause mortality. Results from the model indicate that a 10 μg/m³ increase in BC and NBC was associated with a 49 % (95 % confidence interval [CI]: 26 – 76 %) and 6 % (95 % CI: 3 – 10 %) increase in mortality risk, respectively. Based on global average total PM_2.5 mass composition values (6.1 % and 93.9 % for BC and NBC, respectively), we estimated that the relative risk of all-cause mortality increased by 1.09 (95 % CI: 1.06 – 1.12) per 10 μg/m³ increment in long-term PM_2.5 exposure. Estimation of the effects of BC on mortality based on observations obtained within one city yielded a median OVB of 147 % (95 % CI: −151 – 700) when using a single-pollutant model. Conclusion: In meta-analyses on the health impacts of PM_2.5 components, ignoring the differential effects of BC and NBC causes significant biases in estimating associations with all-cause mortality. Our study presents a novel method to adjust for OVB in meta-analyses, and we find that BC more harmful than NBC components of PM_2.5 by using the novel method.