Discussion

We found no increased risk for cancers of the lung, bladder, kidney, liver, prostate, and colorectum, and for melanoma and non-melanoma skin cancers in association with low levels of exposure to arsenic through drinking-water. The risk for skin cancers decreased with increasing exposure. Results adjusted for enrollment area showed no significant risk associations except for with breast cancer, when the time-weighted average arsenic exposure was used and for melanoma skin cancer in the quartile-based analyses.
The median and mean arsenic exposure at enrollment were 0.7 and 1.2 μg/L, respectively, which are comparable to the concentrations found in Finland (median = 0.14 μg/L) (Kurttio et al. 1999), and the United States (mean = 2 μg/L) (Agency for Toxic Substances and Disease Registry 2005) but much lower than those found in some areas of Asia and Latin America.
Although previous studies provide evidence for an etiologic relationship between arsenic in drinking-water and cancer, they do not predict the cancer risk of low doses (Karagas et al. 2001). The arsenic levels in the Danish drinking-water are 100–1,000 times lower than those reported in studies from Asia and Latin America. It is possible that arsenic concentrations in the Danish drinking-water are below a low effect level; however, the results of the present study cannot rule out a weak adverse effect that is impossible to detect with the method used and the study size.
Conflicting results have been obtained in studies of arsenic and cancer conducted in areas of low arsenic concentrations in drinking-water. A Finnish case–cohort study reported increased risk for bladder cancer in association with exposure to arsenic (Kurttio et al. 1999) based on 61 cases and significant only for exposure 2–9 years before diagnosis for one of the three exposure measures used (Kurttio et al. 1999). Interpretation of the finding is therefore not straightforward. In contrast, our study, based on 214 cases, showed no increased bladder cancer risk. In line with the results of our study, the Finnish study did not find an association with kidney cancer (Kurttio et al. 1999). Studies carried out in the United States found no increased risk for bladder cancer with increasing arsenic exposure (Bates et al. 1995; Lamm et al. 2004; Steinmaus et al. 2003) in areas with arsenic concentrations in drinking-water of 0.5–160 μg/L. In one of these studies an insignificant tendency toward decreasing bladder cancer risk was seen with increasing exposure to arsenic ranging from 3 to 60 μg/L (Lamm et al. 2004). Another study in the United States showed an increased risk for prostate cancer in association with arsenic exposure (Lewis et al. 1999). We did not find such an association.
In the present study, higher exposure to arsenic was significantly associated with a lower risk for non-melanoma skin cancer in the overall analyses. Similar risk estimates were seen for melanoma skin cancer, although the results were not significant, possibly because of the small number of cases. These findings conflict with the results of some previous studies. In Taiwan, Wu et al. (1989) found a significant dose–response relation for non-melanoma skin cancer in association with exposure to arsenic, and a study in the United States showed a 1.9 times higher risk for skin cancer (type not specified) among persons exposed to drinking-water containing > 10 μg/L arsenic than those exposed to <>Knobeloch et al. 2006). Another study of exposure to low levels of arsenic showed no association with non-melanoma skin cancer (Karagas et al. 2001). In a study of non-melanoma skin cancers in which arsenic in toenail tissue was used as bio-marker of exposure, a nonlinear dose–response relation was seen with low exposures, with an inverse association at low levels and an increasing risk with concentrations > 0.09–0.11 μg/g toenail, corresponding to 1–2 μg/L in drinking-water (Karagas et al. 2002). This result is consistent with our findings, as only a small proportion of the cohort members were exposed to drinking-water containing arsenic at > 2 μg/L.
In an experiment in cells in vitro a low dose of arsenic had a protective effect against oxidative stress and DNA damage, supporting the hypothesis that low doses of arsenic could protect against cancer. In this study, the point, at which the protective effect was out weighted by the toxic effect was 1 μmol/L corresponding to 50–60 μg/L (Snow et al. 2005). The findings of inverse risk associations for skin cancer in the present study further support the hypothesis that low doses of arsenic might be inversely associated with risk for skin cancer.
Nevertheless, the negative association between arsenic and non-melanoma skin cancer virtually disappeared when adjusted for enrollment area and when separate risk estimates were made for the two enrollment areas. This might be interpreted as confounding by some regional factor for which we did not adjust. For example, exposure to the sun is a risk factor for both melanoma and non-melanoma skin cancer (Scotto et al. 1996), and this might have confounded the results of the overall analysis if such exposure was more pronounced in the Copenhagen area, as the arsenic concentrations in drinking-water were generally higher in the Aarhus area. This interpretation is, however, contradicted by the fact that the inverse risk association for melanoma skin cancer persisted when risk estimates were calculated separately for the two enrollment areas. Further, the lower risk for confounding obtained by adjustment for enrollment area might be counterbalanced because this adjustment would make it more difficult to detect any effect of arsenic exposure, as part of the variation in exposure relates to differences between the two enrollment areas. Altogether, our finding of negative associations between arsenic and non-melanoma and melanoma skin cancers should be interpreted with caution.
To our knowledge, no epidemiologic study of an association between arsenic and cancer has included breast cancer. The borderline significance of the finding of an increased risk for breast cancer in association with arsenic exposure among cohort members enrolled in the Aarhus area should therefore be interpreted with caution, and more studies are needed to determine if arsenic in drinking-water is a risk factor for breast cancer.
Cases were identified in the virtually complete, reliable nationwide Danish Cancer Registry (Storm et al. 1997), and the Danish Population Registry provided complete follow-up of the cohort members. Although the exposure of the cohort members was assessed independently of who developed cancer, some degree of nondifferential misclassification of arsenic exposure inevitably occurred. This would in most cases be expected to bias risk estimates toward the neutral value (Rothman 2002), and it may therefore have contributed to the null results of the present study. Factors contributing to such exposure misclassification include the following: a) Recent arsenic measurement were assumed to represent historical exposure, in line with the approach of other studies (Bates et al. 2004; Kurttio et al. 1999). b) For 14% of the addresses, we assumed that the nearest water utility provided drinking-water to the address. However, exclusion of persons, who had lived at one of these addresses changed the risk estimates only marginally. c) Some water utilities might have closed during the study period, and supply structures might have changed. It is likely though that drinking-water from past and present water utilities that are spatially close would have similar arsenic concentrations, as the geologic composition of aquifers is fairly homogeneous over small geographic areas. d) There is a lack of information about exposure to arsenic through foodstuffs; however, arsenic in food occurs mainly in the less harmful organic form, and the typical Danish diet does not include arsenic-rich foods such as seaweed, skate, or stingray (Mohri et al. 1990). e) There is uncertainty in the reported intake of tap water. However, the results for time-weighted average arsenic exposure would not be affected by such misclassification, and the results for these two exposure measures gave similar results. f) Use of domestic water supply as a predictor for source of drinking-water implies some uncertainty (Jones et al. 2006). Because most of the water supply areas in the study covered large areas, such misclassification would apply mainly to persons, who traveled far between home and work.
Lack of information on residential histories before 1970 could also have led to misclassification of the exposure. Different migration patterns for cases and noncases before 1970 would imply differential misclassification, but we consider this unlikely because of the long time span between the period of unknown migrations (before 1970) and time of diagnosis for the cancer cases (after inclusion between 1994 and 1997). The strengths of our study include the large study population, the reliable population-based Danish registers, and adjustment for many potential confounding factors. Also, the precise link between place of residence and water supply and the measurements of arsenic concentrations in the drinking-water that was piped to the consumers adds strength to the study.
The limitations of the study include the overall low arsenic concentration in Danish drinking-water and lack of information on other sources of arsenic. Further, the exposure of cohort members before 1970 could not be estimated, as the residential histories before that date were unknown. Therefore we were not able to assess early-life arsenic exposure, which is an important limitation of this study because early environmental exposures might be most significant for cancer risk. Finally, measurement of arsenic in nails or urine would provide more precise estimates of the personal exposure and should be included in future studies whenever possible.