The CPSC claims that it has dealt with sources of uncertainty and variability, but that is not quite accurate. They have considered a range of values for some of their parameters (such as concentration of arsenic on children’s hands, hand-to-mouth transfer, exposure frequency, and bioavailability), which is commendable, but they have not considered the uncertainty in their methodology or the risk estimates of cancer from arsenic in drinking water taken from the National Research Council (NRC) and U. S. Environmental Protection Agency (EPA) reports. “Uncertainty refers to lack of knowledge in the underlying science” (NRC1, p.109), and uncertainties require assumptions, either implicitly or explicitly, to derive numerical values of risk. The credibility of the assumptions affects the extent to which the CPSC claim of excess lifetime risk of lung and bladder cancer of two to 100 per million for children who play on CCA-treated playground structures is warranted (or to what is referred to as “an approximation of reasonable ‘worst’ and ‘best’ cases”, ranging from 0.2 per million to 5,000 per million across their range of parameters, is warranted). <?xml:namespace prefix = o ns = “urn:schemas-microsoft-com:office:office” />
Two critical assumptions are discussed below, the first related to CPSC’s technical approach and the second related to the Taiwan data used in the NRC and EPA risk assessments for arsenic in drinking water. In the first case, the assumption is due to insufficient information about the mode-of-action of arsenic carcinogenicity, a biological consideration. In the second case, the assumption is because the exposure data in the principal study used for risk assessment of arsenic in drinking water is highly aggregated, instead of specific to individuals, a statistical consideration (technically refered to as ecological exposure data, and thus making statistical inference subject to what epidemiologists refer to as the “ecological fallacy”). Both assumptions are the result of genuine limitations of the science or the available data, and both are important to warranting CPSC’s estimates of the cancer risk.
The premise in the discussion that follows is that conclusions from risk analysis for health effects should be as strong as warranted, but not stronger. Just how strong is warranted depends on statistical characteristics, such as the quality and level of detail of the available data and the appropriateness of the statistical methodology, and on consistency of the data analysis with biological expectations. In short, claims based on data analysis for health effects are warranted by both statistical and biological considerations.