Transparency requires that results are both accessible and understandable. Consequently, consideration should be given to the form of the presentation of hazard characterization results for both technical and non-technical audiences. A recommended schema for presentation of hazard characterization results is given in Appendix A
A quantitative interpretation of available information is preferable, and methods available for quantitative evaluation of subjective information should be assessed (Cooke, 1991). Even if there is no means for evaluating qualitative or subjective data, the results of the quantitative development of a dose-response relationship must be discussed and reconciled with additional epidemiological data in order to put it in context with the larger body of information. This is important in order to obtain acceptance of the results in the wider public health community.
When a quantitative approach is used, these guidelines recommend a biased neutral approach to hazard characterization, i.e. the best estimates of dose-response should be presented, together with attendant uncertainty. This approach requires distinction between variability and uncertainty. Variability is the observed differences attributable to true heterogeneity of diversity in a population or exposure parameter. In hazard characterization and MRA, variability cannot be reduced, only more precisely characterized. Uncertainty is ambiguity arising from lack of knowledge about specific factors, parameters or models. In the case of quantitative hazard characterizations, the most likely parameter values and their uncertainty should also be clearly communicated.
A sample of parameter values is usually necessary for use in risk assessment studies, and should, if possible, be made available in digital form, possibly through the Internet. Tables summarizing the results should include confidence intervals as well as characteristic or most-likely values. Graphical presentation of results can be useful in presenting uncertainty. A commonly used presentation graphs the response as a function of the dose (often in logarithmic form), with data points representing observed data and lines showing the fitted model (e.g. best fitting curve and the 5th and 95th percentile limits). However, for the small populations used in most experiments, the fraction of responses can only assume a limited number of discrete values (e.g. 0, 50 or 100% if two subjects were studied), which may unjustly suggest a poor fit of the model. Data clouds of parameter values or line graphs with error bars may be also be helpful. The information that should be presented includes any assumptions made, summary statistics, and references for the data and methods.
A formalized means of presenting and disseminating results of hazard characterization (and MRAs) (e.g. a clearinghouse for dose-response functions and documentation) could speed the application of risk assessment.
Risk assessors should consider the effects of uncertainty from a number of sources - including model uncertainty, measurement uncertainty and extrapolation uncertainty - on the results of their model. Also, the sensitivity of the analysis to various assumptions and decisions made should be carefully evaluated and fully documented. The clear presentation of results could improve the guidance provided for the design of effective public health interventions for the pathogen of interest.
Hazard characterization and MRA results should be shared to the maximum extent possible in order to facilitate work in MRA. Dose-response relationships developed in one context will not be directly adaptable to another and will require careful consideration of differences in populations. Differences in age and immune status of the population, pathogen virulence and other variables will alter the dose-response relationship and these variables must be carefully considered. Adaptation of dose-response relationships outside of the population for which they were developed will require consideration of differences affecting the relationship and will require validation with data from the new population of interest.