Categorical Regression Analysis of Acute Inhalation Toxicity Data for Hydrogen Sulfide
Categorical regression is one of the tools offered by the U.S. EPA for derivation of acute reference exposures (AREs), which are dose-response assessments for acute exposures to inhaled chemicals. Categorical regression is used as a meta-analytical technique to calculate probability-response functions for health effect data that have been classified into ordered severity categories. The software developed by U.S. EPA, CatReg (http://www.epa.gov/ncea/catreg.htm), uses both concentration and duration to predict the probability of achieving a specified severity of effect. The application of categorical regression to the analysis of acute inhalation toxicity data for hydrogen sulfide is presented. A technical literature search identified 22 studies with adequate exposure and effect information for acute inhalation exposures to hydrogen sulfide. The database included rats, mice, and limited human data. The most frequently noted toxic endpoints were respiratory and ranged from reports of biochemical ro pathological changes. A few systemic endpoints were also noted., Effect data were categorized into four severity categories: no adverse effect, mild adverse effect, severe adverse effect, and lethal effect then analyzed using CatReg software. Various model options were tested and evaluated for fit, parallelism of the curves for various effect severities, and for species differences in slopes and intercepts. The analyses showed that humans were more sensitive than rats or mice; rodents were approximately equally sensitive. The model of choice is an analysis with species-specific intercepts and common (i.e., same for all species) concentration and duration parameters. The model most relevant to human health risk assessment is the human model with a human-specific intercept parameter and common concentration and duration parameters. This analysis shows how categorical regression analysis can be used to (1) avoid the common default approaches for duration extrapolation and (2) supplement limited human data with animal data to extrapolate to concentrations and durations beyond the human data.