Reply to Zhou et al.: Methodological Pitfalls Undermine Industrial Emission-related Lung Cancer Risk Assessment

From the Authors:

We thank Zhou and colleagues for their interest in our work (1). Several of their criticisms were discussed in a recent commentary (2), which concluded that on balance, our findings raise concern that air emissions of known and suspected lung carcinogens near the home are contributing to lung cancer risk.

We did not “implicitly assume” stability in emissions prospectively. We assessed historical long-term average exposures before study enrollment, the presumed etiologically relevant period given the long latency of lung cancer (3, 4). Air emissions of some chemicals and metals have declined but for several, such as benzene and styrene, remain in the millions of pounds. Although exposures during follow-up could contribute, they are not necessary for lung cancer development, and emissions trends during this period do not necessarily limit the interpretation of our findings. Evaluation of early-life exposures was not a study objective.

The estimates of population mobility cited by Zhou and colleagues are higher than those for the U.S. general population relevant to our study. Overall mobility is 6% among individuals older than 55 years, two-thirds of the moves in this age group occur within county, and the proportion of movers is lower among those with higher socioeconomic status and White individuals versus their respective counterparts (5). The expected low mobility rate in our population and our confirmation of residential stability in a participant subset (6) should assuage concerns about related impacts. Exposures were assessed before diagnosis, making it unlikely that related measurement error would be away from the null, as Zhou and colleagues assert.

Our inverse distance weighting approach has weaknesses for exposure characterization but was efficient for evaluating multiple agents and appropriate given the limited emissions data. Dispersion modeling that combines meteorology, topography, plume rise, and emission stack characteristics is often not feasible in large epidemiologic studies such as ours that include thousands of industrial facilities. We grounded our interpretations in biological plausibility, fate and transport of specific emissions in air, and prior epidemiologic, toxicological, and mechanistic literature. We found patterns of association consistent with our expectation; relationships with specific agents were stronger within shorter distances from the home. The magnitudes were also in line with those for ambient air pollutants and lung cancer (7, 8) and environmental risk factors more broadly, which are often modest relative to predominant drivers like cigarette smoking. We expect exposure misclassification to have operated independently of lung cancer (i.e., nondifferential), and although we agree that the magnitudes of some associations are potentially inexact, they most likely were underestimated.

The concern raised about correlated emissions lacks context. We mutually adjusted only for a small number of chemicals at specific distances. Some participants had multiple facilities near their homes, but a single facility infrequently reported emissions of multiple known or probable human carcinogens. It is unclear that formal mixtures analysis would be beneficial given few cooccurring emissions.

In conclusion, we thoughtfully caveated our interpretations and respectfully disagree with the assertion by Zhou and colleagues that the findings lack biological relevance. The persistence of lung cancer as the leading cause of cancer death in the United States despite decades of declines in cigarette smoking (9) motivates continued investigation of potential environmental causes of this malignancy.