Sir

I found your Special Report on air-pollution control in the United States (“The politics of breathing” Nature 444, 248–249; 2006) to be generally well balanced. I would like to point out, however, that there is by no means universal agreement among scientists that air pollution at contemporary US levels affects human health. I am one of the sceptics.

The report seems to take at face value the conclusion of “two large, well-respected epidemiological studies”, that every additional microgram of fine particles per cubic metre in the air causes tens of thousands of deaths a year in the United States. Yet joint pollutant analyses — with sulphur dioxide and either sulphates or fine particles both included in the statistical models — show that sulphur dioxide is associated with mortality; fine particles are not (D. Krewski et al. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality Health Effects Institute, 2002). The association of sulphur dioxide with mortality remains unexplained, as there is no plausible biological mechanism by which it could be causing death.

Further, the pollution studies mentioned used the proportional hazards model for analyses of the data. This model assumes that the relative risks of air pollution and potential confounders remain constant over time. It is clear, however, that the basic assumption of proportionality of hazards is satisfied neither for air pollution nor for a strong potential confounder, cigarette smoking (S. H. Moolgavkar Inhal. Toxicol. 18, 93–94; 2006). Use of this model when the assumption of proportionality of hazards is violated can have serious consequences for the inferences drawn from the data. It may, for example, explain the very different results of observational epidemiological studies of hormone replacement therapy in the 1990s and the recently concluded Women's Health Initiative randomized trial (R. L. Prentice et al. Am. J. Epidemiol. 162, 404–414; 2005). Departing from assumptions of proportionality of hazards for potential confounders may also bias the estimates of main effects in cohort studies, particularly when the confounder is a strong risk factor. In air-pollution studies, the use of a manifestly wrong model to adjust for confounding by smoking probably biases the estimates of small air-pollution effects on mortality, although the direction of the bias will depend upon the structure of the correlation between smoking and air pollution.

We do not currently have the methods to reliably estimate small environmental risks.