You are here:
Assessing Uncertainty of Interspecies Correlation Estimation Models for Aromatic Compounds
Citation:
Bejarano, A. AND M. Barron. Assessing Uncertainty of Interspecies Correlation Estimation Models for Aromatic Compounds. SETAC North America 36th Annual Meeting, Salt Lake City, UT, November 01 - 05, 2015.
Impact/Purpose:
Reporting on development of ICE models for aromatics and assessment of uncertainty
Description:
We developed Interspecies Correlation Estimation (ICE) models for aromatic compounds containing 1 to 4 benzene rings to assess uncertainty in toxicity extrapolation in two data compilation approaches. ICE models are mathematical relationships between surrogate and predicted test species that can be used to extrapolate chemical toxicity from one surrogate test species to species with unknown toxicity. ICE models can also be used to augment taxa diversity in Species Sensitivity Distributions (SSDs), which are probabilistic distributions of toxicity data across species. One set of ICE models (Type1) were developed with data compiled across studies, which assumes that there is inherent inter-laboratory variability introduced across studies. The second set of ICE models (Type2) were developed with data compiled within studies, which assumes that variability is reduced when toxicity data come from the same study. Tissue residue-based ICE models were also developed using target lipid modeling (TLM) by calculating compound specific Kow-based target lipid concentrations using the Type2 model dataset (Type2-TLM). Most statistically significant models (90%) had Mean Square Errors (MSE) <0.27, and adjusted coefficients of determination (adj-R2) >0.59, with the lowest amount of variation in MSEs noted in Type2 and Type2-TLM models. Comparison of the fold difference between observed versus predicted values within matching surrogate-predicted species pairs showed that Type2 and Type2-TLM models had fold differences generally 3-fold smaller than those from Type1 models. Despite differences in model predictive ability, no statistically significant differences (p-value >0.05) were found between most ICE-based SSDs and empirical SSDs. In most cases and regardless of model choice the 5th percentile hazard concentrations of the SSDs were within or below the 95% confidence intervals of the empirical values. Despite lower uncertainty in Type2 and Type2-TLM models, improvements in model accuracy are lost when values from predicted species are used to develop SSDs likely due to larger differences across species than across models. This research demonstrates that TLM and ICE models can be coupled and ICE models developed for a range of narcotic aromatic compounds can be used to generate species sensitivity information for a broad range of taxa.