© 2006 American Thoracic Society
Comments on the Updated Harvard Six Cities StudyTo the Editor:In their recent article, Laden and colleagues have added 8 more years of follow-up to the Harvard Six Cities study of air pollution and mortality (1). They suggested that total mortality, as well as mortality for cardiovascular disease and lung cancer, were associated with ambient fine particulate air pollution (PM2.5), and that the results were consistent with other cohort studies. We suggest that this conclusion should be modified based on the actual data presented rather than a strict reliance on model results. Graphing the data indicates that the concentration–response functions are nonlinear for the complete follow-up and also for Period 2. Thus, the effect estimates from linear models overestimate risk and graphical displays suggest a threshold or no association. In the following, we consider total mortality, but similar findings for cardiovascular mortality can be confirmed by the reader. The overall effect estimate of 1.16 for the total follow-up is based on a linear model. The model is incorrect and overestimates risk since the relationship is not linear, as observed when the data are plotted (Figure 1). Inspection suggests a threshold below 20 µg/m3. The total follow-up is a combination of the concentration–response functions in Periods 1 and 2, but they are not comparable because the data are approximately linear in Period 1 and nonlinear in Period 2. This is seen by connecting the data points in Laden and coworkers' Figure 2 (1). The heterogeneity between Periods 1 and 2 is so great that it appears statistically improper to combine these results to develop an overall estimate. Similarly, the effect estimate of 1.13 in Period 2 is incorrect because it is also based on a linear model that overestimates risk for all cities except Harriman. There appears to be no association with PM in Period 2 as none of the cities have significantly elevated effect estimates, and three cities have estimates at or below 1.0; the most polluted cities show decreasing risk as concentration increases (Laden and coworkers' Table 2).
Laden and coworkers say that their conclusions are consistent with other cohort studies, but other cohort studies provide some support for nonlinear relationships. Nonlinear functions were considered likely in the reanalysis of the original American Cancer Society (ACS) study (2), and nonparametric analyses suggested thresholds of 20 µg/m3 or greater in the updated ACS study (3). Threshold models for all-cause and nonmalignant respiratory mortality were statistically significant while the linear models were not significant in the ASHMOG study (4). Laden and coworkers' new data are important and require thorough analysis considering these points.
Somerset, New Jersey
ExxonMobil Biomedical Sciences, Inc., Annandale, New Jersey FOOTNOTES Conflict of Interest Statement: J.F.G. is retired from full-time employment with ExxonMobil Biomedical Sciences, Inc. M.J.N. is currently employed full time by ExxonMobil Biomedical Sciences, Inc. REFERENCES
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
