Article Text

Download PDFPDF

Asthma and swimming pools: statistical issues
  1. B Armstrong1,
  2. D Strachan2
  1. 1Public and Environmental Health Unit, London School of Hygiene and Tropical Medicine, Keppel St, London WC1E 7HT, UK;
  2. 2Department of Public Health Sciences, St George’s Hospital Medical School, Cranmer Terrace, London SW17 0RE, UK

    Statistics from

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Bernard and colleagues1 presented results from several studies investigating childhood asthma in relation to swimming pool use. Though the studies were generally well conducted, there are some respects in which the statistical analysis and interpretation are misleading.

    The study of asthma prevalence in relation to swimming pool use was essentially an ecologic design—the unit of analysis was the school. Though the study of correlations between asthma prevalence and indexes of pool use respects this (the p values are appropriate), the logistic regression analysis does not. The analysis is carried out as if there were 1881 independent observations of asthma and swimming pool use. In fact, observations are not independent—there is “clustering” of asthma by school—even after allowing for effects of swimming pool use and other covariates. The extremely low p values in fig 6 therefore cannot be relied on. The analysis applied to the study of chronic effects on lung epithelium is also limited in not allowing for possible clustering by school. Again, the p values presented overestimate the strength of evidence for an association.

    Caution is also required in interpreting the correlations and p values in figs 5C and 5D. These do not test the association of asthma with pool use, but with a composite index of pool use, pets, and passive smoking. It is not possible from the results presented to distinguish the contribution of each. The authors acknowledge this, but reader confusion may arise because the term “adjusted” is more usually used in epidemiology to describe adjustment for confounding of one effect by another—this is not the case here.

    Finally, the correlations shown in figs 5A and 5B are selected from a wider range of measures of pool attendance, as shown in fig 4. Among these non-independent indices of exposure, the authors have chosen the one showing the strongest correlation with asthma prevalence. For this reason, the “significance” of the p value in fig 5B should be interpreted as suggestive rather than definitive.

    We conclude that the epidemiological evidence for an association of asthma with swimming pool use is not as strong as claimed by the authors.