The Commission on Risk Assessment and Risk Management retained Cambridge Environmental Inc. to conduct case studies of health risk assessment that conform with proposed regulatory reform legislation¹ and to comment, as risk assessors, on the required methods. The principal relevant mandate in these legislative proposals is that the conservative point estimates of risk currently generated and relied upon be augmented with estimates that are in some sense "best"--that are central tendency estimates, generated by taking better account of the uncertainties and variabilities in the underlying data and assumptions.

To illustrate the techniques required to satisfy such a mandate, we studied four cases. The objective of the first case study was to estimate incremental lifetime risk of cancer to an individual in a population whose water supply had been contaminated with part-per-billion levels of 1,1-dichloroethylene (1,1-DCE). The second case study differed from the first only in that 1,1-DCE was allowed, consistent with its dose-response data, to have either an anticarcinogenic or a carcinogenic potency, rather than being constrained to have only a carcinogenic potency, as is the current regulatory norm. The third case study differed from the first only in that it considered exposure similar levels of vinyl chloride, a potent and known human carcinogen, rather than exposure to the equivocally carcinogenic 1,1-DCE. The fourth case study estimated incremental lifetime risk of cancer associated with occupational exposures, rather than low-level environmental exposures, to 1,1-DCE.

For each case study, we first estimated the incremental lifetime risk of cancer to a "reasonably maximally exposed individual" using the methods currently recommended by U.S. EPA. We then prepared a distribution of risk estimates by choosing parameter values for each variable from the distribution defined for that variable and combining these choices in the risk equation. These latter tasks required (1) significant research in the scientific literature, and (2) not a small amount of statistical and computational expertise, Using computer software we created, we repeated the risk calculation about 20,000 times, gathering up each estimate of incremental lifetime risk of cancer to define its distribution. From the distribution, we could estimate the mean, median and 95^th percentile (and other statistics) of the distribution for the incremental lifetime risk of cancer. Each of these might be considered a "best" estimate of risk.

The results of the four case studies are summarized in the following table.

Several comparisons are noteworthy. In the first case study, U.S. EPA methods (specifically, those used for risk assessment of Superfund sites) yielded a point-estimate of risk of 1.3 x 10^-4. Although such an upper-bound point estimate is typically assumed by many to be at about the 95^th percentile of the risk estimate distribution, it corresponded here to the 99.8th percentile of such a distribution. The probabilistic method employed here found that the 95^th percentile of the distribution was about 80-fold lower -- 1.7 x 10^-6. These two different estimates -- both upperbound -- would likely indicate dramatically different intervention strategies. Risks as high as the former often require extensive remediation, whereas risks as low as the latter usually do not.

The second case study, in which exposures to 1,1-DCE were allowed to confer either beneficial or detrimental effects on cancer risk, yielded two central tendency estimates of risk that were negative -- so suggested that low levels of 1,1-DCE might confer no excess risk of cancer, and might even confer a small benefit. Nonetheless, the 95^th percentile of the distribution of risk estimates in the second case study was identical to that estimated in the first case study (1.7 x 10^-6). Thus, allowing the relevant portions of the bioassay data themselves to define the slope and bounds of the dose-response curve -- as opposed to imposing standard, regulatory restrictions on that curve -- yielded both dramatically different central tendency estimates and identical upper-bound estimates.

The third case study, in which exposures to vinyl chloride were substituted for dose-equivalent exposures to 1,1-dichloroethylene, yielded a point estimate of risk (4.1 x 10^-4) that was only three times larger than the point estimate generated in the first study for 1,1-DCE. Such a minor difference belies the substantial differences in the quality and quantity of data surrounding the carcinogenicity of these two chemicals. In contrast, the probabilistic methods yield a 95^th percentile estimate for the risks from vinyl chloride that is some 120-times larger than the estimate from 1,1-DCE.

Finally, the fourth case study suggested that (1) occupational exposures to 1,1-DCE were as expected, substantially riskier than low-level environmental exposures, and (2) that the point estimate of risk is only some three-fold larger than the 95^th percentile estimate. Under certain circumstances, such as relatively high exposures, the deterministic and probabilistic methods may thus yield reasonably similar upper-bound estimates of risk.

Working through these case studies, we have reached certain conclusions about the proposed risk assessment reforms. Among these opinions are: