APPARENT BENEFIT OF WATER FILTERS MAY BE AN ARTIFACT OF STUDY DESIGN

Colford et al. should be congratulated on an excellently conducted and presented study of the potential impact of filtering drinking water on gastrointestinal illness in older adults.¹ There is, however, one issue that deserves further consideration. The study basically compared self-reported illness rates among people who were given an active and a sham water filter. The study was a crossover design in that people were randomized to have an active or sham filter for 6 months and then were swapped to receive the alternate filter. Although crossover designed studies are frequently used in randomized controlled studies and have a number of advantages, they have serious problems when intervention in phase 1 influences the outcome in phase 2.²

Relative risk (RR) of illness associated with active filter use was very different in phase 1 compared with phase 2. From the mean episodes of highly credible gastrointestinal illness (HCGI) presented and person-years at risk from HCGI given in Colford et al.'s table 2, it is possible to calculate the crude RR for phase 1 and 2 independently. In phase 1 the crude RR of illness in people with the active filter was 1.030 (95% confidence interval [CI] = 0.905, 1.172) whereas in phase 2 this was 0.740 (95% CI = 0.622, 0.879). In other words all of the excess risk associated with the sham filter was seen among people who had previously used the active filter and then reverted to drinking unfiltered water. In people who had not previously used the active filter there was no excess risk and indeed the illness rate was slightly higher in the active group.

Although the authors included cycle number as a possible confounder in their models, this would be inadequate to identify any interaction with the order of filter use (active–sham versus sham–active). Simply including cycle phase in the model will be confounded by the decline in reporting of self reported symptoms with time since recruitment that is usually seen in prospective studies of self-reported symptoms.³

This issue will have important implications for the conclusions. As currently presented the conclusion is that if you are an older person drinking water from such a supply, then installing a filter will reduce your illness by around 12%. When taking into account the interaction with order of filter use, the conclusion may be that, if you are an elderly person drinking from the supply, installing a filter will not affect your risk of gastrointestinal illness. However, if you do install a filter and then stop using it after 6 months your risk of illness over the following 6 months will then increase by around 35%. Such an observation would be consistent with the theory that repeat exposure to pathogens in drinking water can influence immunity and that immunity to many enteric pathogens is relatively short lived, lasting for only a few months.^4,5

This observation should not be taken as implying that the drinking water at the study area was not a risk to public health. Indeed, those people who have not built up their immunity through drinking unfiltered Sonoma water for a long time, especially visitors and young children, could be at a substantially increased risk, much greater than the 12% excess risk suggested by the initial analyses.