Statistics from Altmetric.com
We consider that the strength of the evidence linking drinking water to endemic gastrointestinal illness in developed nations has been overstated in the recent paper in this journal by Schwartzet al 1 and the accompanying editorial.2
As noted in the editorial, there have been a number of documented waterborne outbreaks in countries with good water treatment practices. In such outbreaks the association between drinking water and disease has been supported by a substantial amount of evidence including very large effect size, evidence of water treatment failures, robust epidemiological study designs, or identification of the responsible microorganisms in both water and stool specimens. In contrast, the existing body of evidence linking drinking water to endemic gastrointestinal illness is not of an equivalent standard.
The paper by Schwartz et al presents a complex analysis of the relation between variations in drinking water turbidity and hospital admissions for gastrointestinal illness among elderly residents of Philadelphia. The paper is similar in nature to a previous publication by two of the authors,3 which attracted considerable controversy and was the subject of a detailed peer review by the US EPA that criticised many aspects of the analysis (EPA internal memorandum dated February 19, 1998 from W Diamond and J Wiltse to R Levin on EPA Peer Review of Article Linking Finished Water Turbidity with Gastrointestinal Disease in Philadelphia).
The authors responded to some of the criticisms raised in the EPA review,4 however they failed to resolve the fundamental problem with the quality of the turbidity dataset upon which their analyses are based. According to information supplied to the EPA by the Philadelphia Water Department, 75% of the turbidity readings were below the lower calibration limit of the turbidity meters (0.20 NTU). We believe it is not acceptable to include these data in the analysis unless suitable statistical techniques are used to account for the lack of reliability in readings below the quantitation limit.
Moreover, the inherent weaknesses of such ecological studies such as their inability to assess individual exposures or control for bias or confounding, prevent their use in hypothesis testing. To provide substantive evidence of a causal association a consistent body of evidence from epidemiological studies of much stronger design is required. Indeed, when the magnitude of the effect is as small as that proposed for endemic waterborne gastroenteritis (around 15%), many would argue that only blinded randomised controlled trials are sufficiently rigorous to answer the question.5
Even if the problems with data quality and study design of the study by Schwartz et al are ignored, the hypothesis that gastrointestinal disease rates show correlation with minor variations in water turbidity is not biologically plausible. This theory implicitly assumes that physical removal is the major barrier to pathogens in the water supply. However, the Philadelphia water supply is also chlorinated to levels that provide at least 35-log inactivation of viruses and bacteria, and also provide a degree of inactivation for some protozoal pathogens (for example,Giardia). Thus one needs to postulate that most gastrointestinal illness is caused by chlorine resistant organisms (for example, Cryptosporidium) and that these rise significantly in number when minor variations in turbidity occur. However, this is not supported by the clinical experience that such pathogens are relatively uncommon causes of community gastroenteritis.
The two intervention trials carried out in Canada6 7represented a significant improvement over ecological studies but were still subject to a number of methodological concerns. In both trials the participants were not blinded to the interventions and thus bias in the reporting of illness may have influenced the results. In addition, interpretation of the second study was hampered by a very high drop out rate (50%) in one of the intervention groups.
While a number of observations have suggested that drinking water meeting accepted water quality standards may be contributing to endemic gastrointestinal illness, the strength of the epidemiological evidence is not convincing. To overcome the design problems associated with previous studies, randomised controlled trials investigating waterborne disease are being undertaken in a number of countries.8 9The outcomes of these studies will provide the strongest possible quality of evidence on the existence of endemic waterborne disease and its magnitude in different water supply systems.
This issue is an important public health concern, but it must be scrutinised critically and judged on the best possible quality of evidence. Otherwise we risk committing large expenditures on changes in water treatment technology that may be unnecessary and produce no tangible benefit to public health.
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.