Using categorical regression instead of a NOAEL to characterize a toxicologist's judgment in noncancer risk assessment

Abstract

Noncancer health risk assessment involves the evaluation of multiple types of toxic effects. For regulatory recommendations, such as the Reference Dose (RjD), the U.S. EPA relies heavily on expert judgment. This toxicologic judgment mixes toxic impact with likelihood: what effects are adverse, which of these is "critical," and which dose is the highest reliable NOAEL (No-Observed-Adverse-Effect Level). Uncertainty is indicated by qualitative statements of confidence. Statistical regression using ordered categories of overall toxicity is proposed as a superior alternative: uncertainty and variability are represented by statistical models, all relevant data are used, not just the NOAEL for the critical effect, and health risk can be estimated at exposure levels above the RfD. 1: BackgroundMost health risk assessments by regulatory agencies have one of two goals: to estimate a regulatory exposure level where health risk is minimal or zero, or to estimate the health risk at existing or proposed exposure levels.For the first goal, the oral Reference Dose (RfD) has been the mainstay of noncancer risk assessment in the U.S. Environmental Protection Agency (EPA) for several years.The RfD is defined as [ll:An estimate (with uncertainty spanning perhaps an order of magnitude) of a daily exposure to the human population (including sensitive subgroups) that is likely to be without appreciable risk of deleterious effect during a lifetime.The RfD has an inhalation counterpart, the Reference Concentration or RfC. The RfDs and RfCs are published in an electronic data base called IRIS [2]. For clarity without loss of generality, this paper will only discuss the RfD. The Reference Dose has official status in EPA as a scientifically derived exposure level that feeds into the standard-setting process that regulates oral exposures. The RfD has been applied most often to risk assessments for chronic or lifetime exposures.For the second goal, no procedures for noncancer risk have been universally adopted by the U.S. EPA.Statistically based approaches have been generally limited to characterizations of the experimental data set: extrapolation to probabilistic human risk estimates, as is performed for cancer risk, has not been advocated because of high uncertainty and lack of empirical verification. When human exposure-response data are available, three procedures have been used or recommended for obtaining a human risk estimate: expert panel judgment of risk, Bayesian modeling of the critical effect, and regression of adversity categories for all effects combined. Recently, the EPA's Risk Assessment Council issued a memorandum requiring adequate discussion of uncertainties in all risk characterizations. No standard guidance yet exists for conducting and reporting quantitative uncertainties in noncancer risk assessment.This paper presents some ideas on uncertainty that apply to noncancer risk assessments based on categorical regression of toxic adversity on exposure. The basic regression approach is contrasted with the Reference Dose app...

Benchmark Dose Calculation for Ordered Categorical Responses

“…The fitted model can be used to estimate the probability of a specific category for any given concentration level. CATREG, with an appropriate statistical model, has been applied to analyzing various types of toxicity data, including differential severity levels, multiple endpoints, exposure duration, and effects on various species . The statistical software CatReg is well suited to ordered categorical data analysis and is downloadable from the U.S. EPA website…”

“…The CATREG approach has advantages over the NOAEL approach that are similar to the advantages of the BMD approach, and it is also quantitative and potentially valuable for risk assessment using multiple response data and for combining information across studies. However, despite the availability of data analysis software, applications of CATREG have been limited to comparisons with the BMD and NOAEL approaches in deriving RfDs .…”

“…CATREG, with an appropriate statistical model, has been applied to analyzing various types of toxicity data, including differential severity levels, multiple endpoints, exposure duration, and effects on various species. (1,2,(10)(11)(12)(13)(14)(15)(16) The statistical software CatReg is well suited to ordered categorical data analysis and is downloadable from the U.S. EPA website. (17) The CATREG approach has advantages over the NOAEL approach that are similar to the advantages of the BMD approach, (10) and it is also quantitative and potentially valuable for risk assessment using multiple response data and for combining information across studies.…”

Characterization of U-Shaped Exposure-Response Relationships with Application to a Copper Database

“…Traditionally, toxicologists use an uncertainty factor of 10. The RfD is obtained by dividing the NOAEL by one or more of the uncertainty factors (Hertzberg and Dourson, 1993).…”

“…The RfD has been used as the traditional approach for human risk estimation because it is based on unanimous expert group judgment and it considers all available data corresponding to the selected critical toxic effect in the estimation process (Hertzberg and Dourson, 1993). Despite the expert judgment built into the RfD, the procedure for calculating the RfD does not include any measure of the estimation of error or the variability in the RfD.…”