© 2004 American Thoracic Society
Acute Effects of Air Pollution on AdmissionsReanalysis of APHEA 2To the Editor:The validity of statistical methods used in the analysis of temporal associations between outdoor air pollution and health has been questioned recently (1, 2). Default convergence criteria used in the estimation of the model parameters together with programming within S-PLUS software to calculate standard errors in GAM may have resulted in bias in the effect estimates and underestimation of the standard errors. The APHEA 2 (Air Pollution and Health: A European Approach) project recently published results of their analysis of the short-term health effects of particulate air pollution on numbers of hospital admissions for respiratory diseases (3). These data have been reanalyzed in light of these new concerns. This reanalysis included a reduction in the criterion for model convergence from 10–3 to 10–14 and an increase in the number of iterations allowed for this convergence criterion to be met from 10 to 1,000. Possible underestimation of the standard errors was addressed by using parametric smoothing terms (natural cubic splines) based on the number of degrees of freedom used in the original models. The original analyses studied the effect of particles with an aerodynamic diameter of less than 10 µm (PM10) and Black Smoke on hospital admissions for asthma (ages 0–14 and 15–64 years), chronic obstructive pulmonary disease and asthma (ages 65+), and all respiratory diseases (ages 65+) in eight European cities. In general, the summary effect estimates changed little. Table 1 illustrates the original results for PM10 together with those obtained when the revised convergence criteria are applied. Black Smoke estimates also were largely unaffected by this change. Changing from nonparametric smoothing to parametric smoothing (using natural cubic splines) revealed that the effect estimates were largely insensitive to the smoothing function (data not shown). Further details are given in the Health Effects Institute special report (4).
The above analyses show that, in the case of hospital admissions for respiratory diseases in European cities studied by the APHEA 2 project, the results are robust to both the convergence criteria used by the S-PLUS software and the choice of smoothing function applied. Therefore, the overall conclusions from the original study remain unaltered (3). This robustness may be a result of the modeling techniques used in APHEA 2 where the seasonal and nonlinear relationships with the environmental confounders have been correctly accounted for or the lack of concurvity (nonparametric equivalent of correlation) between the environmental data.
St. George's Hospital Medical School London, United Kingdom FOOTNOTES Conflict of Interest Statement: R.W.A. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. REFERENCES
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
