Colon and rectal cancer incidence and water trihalomethane concentrations in New South Wales, Australia

Epidemiological studies have reported increased risks of bladder, colorectal and renal cancer, and adverse reproductive and developmental outcomes in people exposed to chlorinated drinking water or chemical by-products of chlorination (disinfection by-products or DBPs), although not consistently [1–7]. These studies have usually used measurements of trihalomethanes (THMs, e.g., total THM, chloroform, bromoform etc.), a major sub-group of DBPs, as a measure of exposure.

A recent meta-analysis of case–control and cohort studies of the association of DBPs with colorectal cancer (CRC) found summary odds ratios (ORs) of 1.30 (95% CI 1.06, 1.59) for rectal cancer and 1.27 (95% CI: 1.08, 1.50) for colon cancer, comparing the highest exposure category with the lowest [8]. Many of the studies in the meta-analysis did not report the actual concentration of THM in the water. Weaknesses in exposure measurement and in the control of potential confounders in many of the studies make these findings uncertain. In addition, all the studies included in the meta-analysis were conducted in Europe or in North America. It has been revealed in a recent meta-analysis and pooled analysis that the method of disinfection varies importantly between North America and Europe which may contribute to the differences in the risk of cancer [2]. Therefore, the findings from studies conducted in other countries may not be applicable to Australia.

Due to limited generalizability of the previous studies, we have conducted an ecological study as a starting point to explore the association in Australia. In this study, we investigated whether the incidence of CRC is associated with THM concentrations in two water supply regions in New South Wales, Australia. All public water supply systems in these regions are served by chlorinated or chloraminated water.

We undertook a spatial ecological study with local government area (LGA) by calendar year as the unit of analysis. The study covered two separate geographical areas: the Sydney Water Corporation (SWC) region, which supplies water to 47 LGAs and the Hunter Water Corporation (HWC) region which supplies water to 5 LGAs. For each of the LGAs we sought data on the incidence of colorectal cancer from 2001 to 2006 and estimates of THMs in water supplies from 1995 to 2001, to allow a five-year lag. Among the DBP species, only THMs were routinely measured in these supplies; they had been consistently measured since 1995 in the SWC region and since 1997 in the HWC region.

In the HWC water supply system, water is piped into reservoirs in six water supply zones and then distributed to homes located in nine water distribution zones that contain the sampling sites. The SWC water supply system has 14 delivery systems encompassing 33 distribution systems, which, in turn, encompass 180 reservoir zones that contain the sampling sites. A detailed description of the study areas, the water supply system, disinfection practices etc. is given elsewhere [9].

The indirectly age-standardized incidence rate of CRC for the period 2000 to 2006 for ages 35 years and over in the SWC region was 69.8 (95% CI 67.6, 72.0) per 100,000 in men and 55.9 (95% CI 54.1, 57.7) per 100,000 in women. The corresponding rates in the HWC region were 81.3 (95% CI 75.6, 87.0) in men and 60.4 (95% CI 55.3, 65.6) in women.

The adjusted IRR for year of diagnosis of CRC was below one for both sexes and for all species of THMs, indicating that CRC incidence rates were falling over the period of the study (not shown in the analysis). The adjusted IRR for the relative index of inequality were below unity for all exposure categories, more so in women than men. Overall, men in the lowest socioeconomic quintile had 7% less risk of CRC than men in the highest quintile; women in the lowest socioeconomic quintile had 12% less risk than women in the highest quintile (not shown in the analysis). Men whose drinking water came from sources other than tap water had 24% less risk and for men who engaged in high risk drinking there was a 29% higher risk of CRC but none of these associations was statistically significant.

We found a positive association between CRC incidence in men lagged five years, with bromoform concentrations in water. This association was largely confined to colon cancer. There was no appreciable association of colorectal cancer with any other species of THMs or with total THMs for either sex. The results showed little sensitivity to removal of SES (IRSD), the spatial random effect term, or addition of the within-area THM variance to the model. When the exposure was averaged across the calendar years, in place of the five-year lag, the positive associations became slightly stronger with reduced precision, possibly because of less attenuation due to non-differential measurement error.

The ecological study design is a primary limitation of our study because we could not adjust adequately for individual level covariates such as consumption of meat and vegetables, which might confound the association of CRC with THM concentrations. The ecological study design also did not allow us to incorporate estimated exposures relevant to the etiologic period that might be gleaned from long-term residential and drinking water histories. That we used only five years of lag between exposure and outcome is another limitation; other investigators have argued that even a 34 year period between the beginning of exposure and the end of follow-up might be too short to detect an effect of environmental exposure on mortality from cancer [27]. However, available evidence on THMs suggests that they may cause cancer by gene mutation or by cell proliferation followed by cytotoxicity, with the former more prominent for the brominated species [28–30]. Thus it is possible that effects could be seen after both short and long periods. In our study the THM exposure data required for a long lag period were not available (available only from 1995). Another major limitation to our study was that we made no assumption about contribution of THMs from different exposure routes (inhalation and dermal absorption through showering, bathing or swimming or other water use activities) which may be significant contributors to the overall exposure to THMs [31]. The associations observed may not be directly related to THMs but with unmeasured DBPs correlated with their formation [32].

A further weakness of the ecological design is that people often change their residence and hence their exposure status, if DBP concentration varies spatially (which was true in our case and necessary to the ecological design). Australian Census data for 2007 show that 80% of people in NSW did not change their residential address in the previous year, 6% moved to another address in the same Statistical Local Area (SLA) (SLAs are the same as or, for a small number of cases, subsets of LGAs), and 14% moved a greater distance [33]. For the previous five years, 55% of people did not change their residential address, 13% moved to another address within the same SLA and 32% moved a greater distance [34]. We did not take into account the exposure misclassification which might have resulted from this population mobility.

The main strengths of our study are the use of estimated annual mean concentrations of THM species for small areas, albeit often with missing data; a lag period between exposure and outcome; and more advanced statistical methods. No previous study of this type has used exposure data at this level of detail or specificity (monthly measurement of THMs), or used a lag period in the analysis [35–40]. Annual exposure estimates and disease rates, and a lag period may be quite important because THM concentrations in water vary over time due to factors such as changes in disinfection practice (e.g., change from chlorination to chloramination and change in the chlorine to ammonia ratio), installation of filters in water treatment plants, rechlorination practices throughout the distribution systems and change in environmental conditions (drought breaking rains triggered the increase in brominated trihalomethanes and chloroform concentration around 1998) [9].

Another strength are the analytical methods used which included Poisson variability, spatial random effects and within-area variability of exposure that permitted control for pure specification bias. Previous ecological analyses of associations of DBPs with cancer risk have used only correlation coefficients or simple linear regression. Correlation coefficients mix the strength (size) and precision of an association, whereas the regression coefficient reports them separately. Simple linear regression using rate as the outcome is prone to several problems. First, the assumption of homoscedasticity (that is, the error term of the regression model has a constant standard deviation that does not depend on the value of the outcome) is a particular issue for studies that have used unweighted linear regression. Generally, small areas have unstable rates and including them in a regression model without taking account of this factor can cause unpredictable bias. Second, simple linear regression assumes independence of spatial units, such that there would be no clustering of adjacent areas with similar rates, which is unlikely. Third, least squares estimators assume that the rates or log of the rates are normally distributed, whereas we would generally expect rates in small areas to have a Poisson distribution. The analytical methods we have used address all of these issues; thus our results are more likely to faithfully reflect the reality than results based on linear regression. In addition, our methods have allowed us to largely rule out within area variability and spatial autocorrelation as explanations for the positive associations we observed between bromoform concentrations and colon cancer.

To our knowledge, this is the first study in which the associations between colon and rectal cancers and THM species concentrations in drinking water have been examined. The finding of a relationship between bromoform and colorectal cancer might be dismissed as being a chance finding, given the multiple testing. However, some recent results suggest otherwise. In a case–control study of rectal cancer in men, Bove and co-workers [1] found a positive association of rectal cancer with bromoform (OR 1.20, 95% CI 1.05, 1.35). Like us, Bove and co-workers [1] found no association of rectal cancer with chloroform or total THMs. While it is possible that the lack of an association between rectal cancer incidence and bromoform in our study could be due to the short lag period, since ‘low-level’ environmental exposures may require several decades to cause a detectable effect on cancer risk [27, 41], it might also be due to lack of statistical power, since we did find an association with colon cancer. The only other study of the association of colon or rectal cancers with a specific THM found a positive association of chloroform with colon cancer (RR 1.68, 95% CI 1.11, 2.53), but not rectal cancer (RR 1.07, 95% CI 0.60, 1.93), in postmenopausal women; other species of THMs were not examined [42]. We found no evidence of such an association in either men or women.

Our recent meta-analysis of cohort and case–control studies of total THM exposure and colon and rectal cancers [8] found that risks of both were increased in the highest category of exposure relative to the lowest: colon cancer OR 1.33 (95% CI 1.12, 1.57) and rectal cancer OR 1.30 (95% CI 1.06, 1.59). These results suggest that THMs may increase the risk of both colon and rectal cancer and that if the effect is specific to a particular species of THM that species is a large component of the total or it has a strong effect. The former, at least, is not true for bromoform [28, 43]. There were too few sex-specific results to see if there were risk differences by sex of colon or rectal cancers with THM exposure in the meta-analysis.

Expert evaluation of the evidence has concluded there is limited evidence for the carcinogenicity of bromoform in experimental animals. This evaluation, however, was made in 1999 and was based on only one relevant experimental study [29]. With respect to the other THM species, there was considered to be sufficient evidence for the carcinogenicity of chloroform [44] and bromodichloromethane [45] in experimental animals, when evaluated in 2005, but only limited evidence for dibromochloromethane [46] (evaluated in 1999). Some individual studies, however, have found that brominated trihalomethanes (bromoform, bromodichloromethane and dibromochloromethane) are more carcinogenic than chloroform [47, 48]. In our study, both bromodichloromethane and dibromochlormethane were more strongly positively associated with colorectal cancer, or just colon cancer in men, than were chloroform or total THMs. The IRRs for all brominated trihalomethanes in men were 1.072 (95% CI 1.033, 1.113) for colon cancer and 1.080 (95% CI 1.031, 1.131) for rectal cancer (not shown in the analysis). Considered together the experimental and epidemiological results suggest that further investigation of the associations of brominated trihalomethanes with colon and rectal cancers is warranted.

While causality between THM exposure and colorectal cancer has not been established, most people in developed countries, drink water containing THMs. Therefore, a small association projected across the globe could translate into a large number of cases. Thus the reason the associations between THMs, and particularly bromoform, and colon and rectal cancers should be investigated further in higher quality epidemiological studies.

Since the existing epidemiological studies provide some, although weak evidence, that the effects of different THM species could differ by sex and cancer site, future studies should measure individual species of THMs and study the effects on men and women with regard to colon and rectal cancers separately. The studies should be sufficiently powered to detect small associations, as that small association could translate into an important number of cases.

We found weak evidence of a positive association between incidence of colorectal cancer in men and total THM concentration in water supplies and a somewhat stronger association for bromoform concentration. This was primarily because of an association of colon cancer incidence in men and bromoform concentration. Although an association of rectal cancer in men with bromoform concentration has been observed previously, the association we observed with colon cancer has not been previously reported. The inconsistency of these observations by cancer type and the ecological study design prevent inference of causation. However, the potential population impact of such an association justifies further research into the effects of THMs in drinking water, particularly the brominated species, and colorectal cancer.

Institution where the work was done: School of Public Health, The University of Sydney, NSW 2006, Australia.