EPA’s Integrated Risk Information System (IRIS) “is an electronic data base containing information on human health effects that may result from exposure to various chemicals in the environment.” IRIS values are hazard – not risk – values that are often cited or used in other chemical safety evaluations, by EPA, federal agencies, states, and a diverse array of stakeholders.
Simon et al. 2016 explores and refines the 2014 National Academy of Sciences (NAS) recommendation that EPA develop and expand its use of formal quantitative methods, including the Bayesian method, when developing values used in IRIS assessments.
The Simon et al. co-authors (including Dr. Yiliang Zhu, a biostatistician who was on the NAS panel) took this example further and applied it to 24 IRIS non-cancer derivations. Through example calculations, the research demonstrates that using a distribution of values instead of a single-point estimate for each uncertainty factor can have a large impact on the final reference value.
Why is this paper important to chemical risk assessments?
Chemical risk assessment is precautionary, that is, it is geared to overstating the likely risk from a chemical exposure so that we are sure sensitive individuals are protected. This paper gives us a sense of how protective we really are.
The most basic idea of Bayesian methods is that the state of knowledge should be responsive to new information. Even in the 1990s, statisticians familiar with the IRIS process for derivation of toxicity values for non-carcinogens suggested that Bayesian methods could provide a means of updating the toxicity values as new information became available. Anyone familiar with IRIS knows that many of the toxicity values have become outdated. Some of these were developed in the 1980s and early 1990s and have never been updated to include any new data. Hence, the advent of these methods is long overdue.
What does it change?
Risk managers would prefer to make judgments on data that are more certain. The approach described in this paper shows some of the uncertainty in risk assessment, thereby helping managers make better decisions. Instead of relying on a single point estimate of safety, using statistical methods, the Bayesian approach allows risk assessors to rely on a representative distribution.
In deriving hazard values for non-carcinogens, the point of departure dose, often obtained from animal testing, is divided by a composite factor capped at a value of 3000, to represent five distinct areas of uncertainty. The numerical uncertainty in each of the five areas can be represented numerically by a bell curve, not just a point estimate. Creating a bell curve allows the use of statistical techniques to obtain a toxicity value that is a more precise accurate estimate of safe exposure for humans than the process that is currently being used by IRIS.
Why did the National Academies suggest this research?
To increase the understanding of how risk assessment is done by showing how uncertain it sometimes is.
Both the National Academy and the paper recommend that the uncertainty be represented numerically by a probability distribution, e.g., a bell curve. The National Academy foresaw that in the future, toxicity values would likely be derived from different data sources, including animal testing, observational epidemiology and high throughput in vitro assays. Bayesian methods would be ideal for updating existing toxicity estimates based on these different types of data.
How hard will it be for agencies and programs, like IRIS, to make changes?
Forward leaning scientists are moving in this direction and EPA says they are interested in using Bayesian methods for dose-response analysis. While incorporating the methods described by the National Academy and in the paper should be quite easy, it will still take time for these methods to gain acceptance. For example, the benchmark dose method took EPA 12 years between initial development and first use. However, if the will exists to use more robust methods, implementation should be fairly easy.
Any other important comments?
It is important to remember that risk assessment is imprecise. Ranges in risk assessment values are being encouraged. This method allows the development of a range.
One of the real advantages of the method is the ability to use sampling techniques to combine probability distributions. Doing so will provide risk managers information that will help them to communicate how confident they are in their risk estimate of a chemical.
This paper shows that the simple point estimate method is overestimating hazards compared to more robust approaches that could be used in hazard assessments developed by federal agencies, like EPA.
ACC encourages all stakeholders to discuss this new research and to encourage the IRIS program, and others, to modernize their approach to hazard assessment.