Arsenic is elevated in some soils and mine waste in the historic goldfields region of Victoria, Australia. This geographical correlation study investigated the exposure potential of arsenic in soil in relation to cancers with an a priori association with arsenic in drinking water, between 1984 and 2003.
The study area included 61 Statistical Local Areas (SLAs). We used incident cancers recorded by the Victorian Cancer Registry, georeferenced to SLA of residence at time of diagnosis. Geographic data manipulation was undertaken to convert cancer and demographic data to a common SLA boundary definition, enabling estimation of age-sex standardised incidence ratios (SIRs). Spatial empirical Bayes (SEB) smoothing was applied to reduce variance instability inherent in raw SIRs. Two soil arsenic exposure metrics were derived from available geochemical datasets: 1) predominantly geometric mean arsenic concentrations, with kriging predictions assigned to unsampled SLAs (range 1.4 to 1857 mg/kg); 2) kriging predictions only (‘smoothed' metric; range 12 to 299 mg/kg). Links between SIRs and natural logarithmic transforms of the exposure metrics were investigated using spatial autoregressive modelling, adjusting for spatial trends detected. SLAs were also aggregated by quintiles of metric 1, and SIRs estimated.
Raw and SEB smoothed SIRs of all cancers among males were significantly associated with interactions between both soil arsenic exposure metrics and a dichotomised score reflecting socio-economic disadvantage. For all female cancers, the significance of these interactions was slightly less consistent. The magnitude of effect per 2.7-fold increase in metric 1 was similar for raw and SEB smoothed SIRs of all cancers, although less than effects estimated using the smoothed exposure metric. SEB smoothed SIRs of individual cancers gave more conservative effect estimates, which are reported here. In more disadvantaged SLAs, risk of cancer overall increased for males and females (β=0.049 [95% confidence interval [CI] 0.016–0.082] and β=0.037 [0.009–0.065], respectively). Prostate cancer (β=0.043 [0.013–0.073]) and female melanoma (β=0.052 [0.017–0.087]) also increased in more disadvantaged SLAs, while male melanoma (β=0.047 [0.012–0.081]) increased when adjusted for disadvantage. Although linear trends were not evident, significant excess risks of all cancers (SIR 1.09 [95% CI 1.07–1.12] and 1.06 [1.04–1.09], and melanoma (SIR 1.13 [1.04–1.23] and 1.22 [1.12–1.31]) for males and females, respectively, and male prostate cancer (SIR 1.08 [1.03–1.13]), were found in the uppermost quintile of metric 1.
Discussion and Conclusions:
We found consistent evidence of small but significantly increased risks of all cancers, melanoma and prostate cancer among males, and some evidence of increased risk among females of all cancers and melanoma, in association with increasing soil arsenic levels in spatial autoregressive models. The robustness of this geospatial analysis is corroborated by excess risks observed among aggregated SLAs in the uppermost quintile of soil arsenic level.