Uncertainty in risk estimation

SC 1-5 Uncertainty in Risk Estimates SC 1-6 Basis for the Linearity Assumption SC 1-7 Information Needed to Make Radiation Protection Recommendations for Travel Beyond Low-Earth Orbit SC 9 Structural Shielding Design and Evaluation for Medical Use of X Rays and Gamma Rays of Energies Up to 10 MeV SC 46 Operational Radiation Safety... 

Risk characterization is the most important and final part of a risk assessment. It summarizes and interprets the information from hazard identification, dose-response, and exposure steps, identifies the limitations and uncertainties in risk estimates, and communicates the actual likelihood of risk to exposed populations. The uncertainties identified in each step in the risk assessment process are analyzed and the overall impact on the risk estimate(s) is evaluated quantitatively and/or qualitatively. 

Objective consideration is needed of the quality of risk estimation. There are many sources of uncertainty in risk estimation. There are errors in the descriptions of the extent and level of site contamination, the transport of contamination to the receptor and the response of the receptor to the contamination. All of these errors require at least qualitative estimation. Then an overall statement of uncertainty of estimation is needed for each risk that is being evaluated. 

In most situations, uncertainties in frequency estimates are the greatest contributor to uncertainties in risk estimates. 

In using any risk measure, it should be remembered that risk measures, at best, are only estimates of possible event frequency and consequences All risk measurements have uncertainties. In some situations, the uncertainties can be highly significant. The fact that risk measurement is imprecise should be a consideration in any risk-based decision-making process. Chapter 5 of Reference 4 provides further discussion of uncertainty in risk decision making. 

In the final phase of risk analysis—risk characterization—one integrates outputs of effects and exposure assessments. Risk is expressed in qualitative or quantitative estimates by comparison with reference values (e.g., hazard quotient). The severity of potential or actual damage should be characterized with the degree of uncertainty of risk estimates. Assumptions, data uncertainties and limitations of analyses are to be described clearly and reflected in the conclusions. The final product is a report that communicates to the affected and interested parties the analysis findings (Byrd and Cothern, 2000). 

Bayesian statistics are applicable to analyzing uncertainty in all phases of a risk assessment. Bayesian or probabilistic induction provides a quantitative way to estimate the plausibility of a proposed causality model (Howson and Urbach 1989), including the causal (conceptual) models central to chemical risk assessment (Newman and Evans 2002). Bayesian inductive methods quantify the plausibility of a conceptual model based on existing data and can accommodate a process of data augmentation (or pooling) until sufficient belief (or disbelief) has been accumulated about the proposed cause-effect model. Once a plausible conceptual model is defined, Bayesian methods can quantify uncertainties in parameter estimation or model predictions (predictive inferences). Relevant methods can be found in numerous textbooks, e.g., Carlin and Louis (2000) and Gelman et al. (1997). 

Reproductive risk descriptors are intended to address variability of risk within the population and the overall adverse impact on the population. In particular, differences between high-end and central tendency estimates reflect variability in the population but not the scientific uncertainty inherent in the risk estimates. There is uncertainty in all estimates of risk, including reproductive risk. These uncertainties can result from measurement uncertainties, modelling uncertainties and assumptions made due to incomplete data. Risk assessments should address the impact of each of these uncertainties on confidence in the estimated reproductive risk values. 

One possible variation of Alternative 3 would be to set RMCLs as a range of finite risk levels. This alternative would recognize the lack of accuracy and precision of risk calculations and the inherent difficulties in selecting one finite level as the only appropriate health goal in view of the numerous scientific uncertainties of risk estimates. 

Deterministic models use a single value for input variables and provide a point estimate of exposure or dose. Probabilistic models take into account the fact that most input variables will have a distribution of values. These models use probability distributions to develop a range of plausible exposures for the population of concern. Understanding exposure distributions will allow understanding of the range of exposures as well as prediction of risk for the entire population. It will also allow prediction of risk for the most highly exposed individuals. Sophisticated models can be used to develop distributions for different pathways and populations. They can also be used to develop information on interindividual variability and uncertainty in the estimated distributions and to predict the variables that are most important for both exposure and dose. 

Constraints, uncertainties and assumptions having an impact on the risk assessment should be explicitly considered at each step in the risk assessment and documented in a transparent manner. Expression of uncertainty or variability in risk estimates may be qualitative or quantitative, but should be quantified to the extent that is scientifically achievable. 

The present monograph suggests a four-tier approach for characterizing the variability and/or uncertainty in the estimated exposure or risk results. These four tiers are described below. 

Finally, CA and RA both apply algorithms that combine the results of single substance evaluations to produce an estimate of the mixture risk. The uncertainty in the estimate is a combination of the uncertainties in the individual components. Calculation rules for evaluating the overall uncertainty based on the uncertainties in the individual components are provided in Appendix 1. 

Quantifying Uncertainty in Risk Assessment Practical Approaches and their Application to Estimating Ecological Risks of Pesticides , report on the findings of a workshop held in Pensacola, FL, USA., SET AC, February 24 to March 1,2002 (in preparation). 

A conventional way of addressing uncertainty in risk assessments is to estimate and assign reasonable worst-case conditions for our evaluations or models. Thus, in this case one would typically pick the worst case (highest G and lowest Q) to estimate a worst case for C. Next, this could then be combined with best case estimates (lowest G and highest Q) to provide a range for C. Finally, the impact or sensitivity of G or Q on either best or worst-case scenarios could be determined by calculating the results of varying these predictors from maximum to minimum individually in each. 

Hoffman and Hammonds 1994). In addition, standard data distributions have been proposed for a variety of exposure variables, such as age-specific distributions for soil ingestion rates, inhalation rates, body weights, skin surface area, tap water and fish consumption, residential occupancy and occupational tenure, and soil-on-skin adherence (Finley et al. 1994). It should also be pointed out that these techniques can be combined with other advanced risk assessment methods (i.e., PBPK modeling) to further reduce uncertainty in exposure estimates (Cronin et al. 1995 Simon 1997 Nestorov 1999, 2003). 

Once historical exposures have been estimated, these values can be used to calculate potential dose, which, in turn, can be nsed to estimate the theoretical cancer risk associated with the exposure of intoest. General dose equations have been presented, but it is important to note that these eqitations may incorporate many assumptions, which may lead to uncertainty in the estimations. Each of these assumptions should be carefully evaluated, and, depending on the anticipated nse of the estimated values, an in-depth uncertainty analysis should be considered (Paustenbach and Madl 2008). 

The US EPA has subsequently published a comprehensive toxicological review of bromate (US EPA, 2001). Studies with rats based on low-dose linear extrapolation, using the time-to-tumour analysis, and using the Monte Carlo analysis to sum the cancer potency estimates for kidney renal tubule tumoms, mesotheliomas, and thyroid follicular cell tumours, gave an upper-bound cancer potency estimate for bromate ion of 0.70 per mg/kg day. This potency estimate corresponds to a drinking water unit risk of 2 x 10 per pg/L, assuming a daily water consumption of 2 litres/day for a 70-kg adult. Lifetime cancer risks of 10 , 10 , and 10 are associated with bromate concentrations of 5, 0.5, and 0.05 pg/L, respectively. A major source of uncertainty in these estimates is from the interspecies extrapolation of risk from rats to humans. 

For both perspectives we conclude that risk cannot be adequately described and evaluated simply by reference to the summarising probabilities and expected values. In the classical case we have to take into account the uncertainties in the estimates, and in the Bayesian perspective we have to acknowledge that the computed probabilities are subjective probabilities conditional on a specific background information (Aven 2008a,b). 

Uncertainties in Estimating Frequency. The greatest influence on uncertainty in risk results can be attributed to uncertainties in frequency estimates arising from ... 

Firstly, the use of a BBN as an advanced method for risk calculations yields new possibilities for risk analysis. Platypus generates risk analyses that yield risk distributions rather than point estimations. These risk distributions automatically address aleatory uncertainties in risk analysis (whose origin lies in randomness) and identifies instances of high risk that are overlooked in setting risk standards. The distribution will also yield new leading risk indicators and new methods and instruments for risk management. When estimates of epistemic uncertainty ranges are available, these can be incorporated in the overall distributions as well. 

---