Methodological Pitfalls Undermine Industrial Emission–related Lung Cancer Risk Assessment

To the Editor:

We read with interest the recent analysis of more than 440,000 National Institutes of Health/American Association of Retired Persons Diet and Health Study participants by Madrigal and colleagues, which provides compelling evidence linking residential proximity to industrial emissions with increased lung cancer risk (1). Nevertheless, certain pitfalls cast doubt on the specificity and magnitude of several agent-specific associations.

The exposure metric is built solely from Toxic Release Inventory (TRI) emissions reported between 1987 and 1995, yet cancers were ascertained through 2018. The authors implicitly assume that each facility’s releases remained stable for two further decades and every participant remained at the enrollment address. Both assumptions are improbable, as approximately 12–13% of Americans change residence each year; compounding those rates shows that nearly three-quarters relocate within a decade, and approximately one third experience a cross-country move (2). Such mobility introduces classical measurement error that can bias hazard ratios toward or away from the null in unpredictable ways. Early-life and postbaseline exposures, now considered critical windows for lung carcinogenesis (3), are therefore misclassified or missed altogether, undermining inferences about lifetime risk.

Inverse-distance-squared weighting, even when modified by prevailing wind, treats all stacks as ground-level point sources and neglects exit temperature, stack height, plume rise, and atmospheric stability. Metals such as cobalt and nickel condense quickly and deposit within 1 km, whereas volatile organic compounds behave far differently. State-of-the-art industrial exposure assessments couple TRI masses with Gaussian or Eulerian dispersion models. Such frameworks perform better, especially when pollution sources vary in height and thermal buoyancy (4). Without incorporating these parameters, the reported “distance-trend” (metals persisting to 10 km, organics not) may simply reflect systematic misclassification of exposure magnitude rather than true toxicokinetic behavior.

Facility emissions are mixtures: the same refinery that releases benzene also emits 1,3-butadiene, polycyclic aromatic hydrocarbons, and metals. Madrigal and colleagues mutually adjusted for “correlated carcinogens,” a procedure that can introduce bias when exposures share sources and measurement error. Simulation works have demonstrated that conventional mutual adjustment inflates confidence in the wrong pollutant and widens uncertainty in regard to the true mixture effect (5). Bayesian kernel machine regression, weighted quantile sum, or principal-component survival models are better suited to disentangling highly collinear emissions while preserving overall risk estimates. Until such techniques are applied, the singled-out role of cobalt or diethyl sulfate should be regarded as provisional.

Collectively, these limitations indicate that modest hazard ratios, particularly those hovering around 1.15, may lack biological relevance when time-varying residence, plume physics, and mixture structure have been rigorously addressed. Future analyses that use updated TRI inputs to cover the full follow-up, link participants to longitudinal address histories, and employ multiple-pollutant dispersion and mixture models would offer firmer guidance for regulation and public health.