Abstract: Study objective: The objective of this study is to demonstrate the methodological shortcomings of currently available analytical methods for single-city time series data, one of the most commonly used ecological study designs in air pollution and respiratory disease research.
Design and Methods: We analyse single city epidemiological time series of daily Chronic Obstructive Pulmonary Disease (COPD) (ICD codes 490-492, 494, 496) and daily asthma (ICD codes 493) hospital admissions in Melbourne, Australia from July 1989 to December 1992. Air pollution data comprise nitrogen dioxide, ozone and sulphur dioxide (all measured in parts-per-hundred-million) and air particles index consistent with particulates between 0.1 and 1mm in aerodynamic diameter. Statistical analyses were performed using generalized linear models, generalized additive models, Poisson autoregressive models and transitional regression models. Results are presented as relative risks for an increase from the 10th to 90th percentile of the respective pollutant.
Main Results: The estimated effect of nitrogen dioxide on COPD hospital admissions was similar across the different statistical models, RR = 1.06 (95% CI 1.01 to 1.11). Similarly the estimated effect nitrogen dioxide on asthma hospital admissions was also consistent, RR = 1.05 (95% CI 1.01 to 1.09). However, the effects of ozone, air particles index and sulphur dioxide were highly sensitive to model specification for both COPD and asthma hospital admissions.
Conclusion: In single-city studies of air pollution and respiratory disease, very different conclusions can be drawn from competing models. Furthermore, real time series data have greater complexity than any of the commonly-used existing models allow. None of the statistical models considered here were completely adequate in modelling strong correlation structure. Consequently, single city studies should use several statistical models to demonstrate the stability of estimated effects.
Keywords: air pollution, statistical methods, asthma, COPD, hospital admissions.