气候变化领域集成服务门户

Interspecies correlation estimation (ICE) models were developed for 30 nonpolar aromatic compounds to allow comparison of prediction accuracy between 2 data compilation approaches. Type 1 models used data combined across studies, and type 2 models used data combined only within studies. Target lipid (TLM) ICE models were also developed using target lipid concentrations of the type 2 model dataset (type 2-TLM). Analyses were performed to assess model prediction uncertainty introduced by each approach. Most statistically significant models (90%; 266 models total) had mean square errors < 0.27 and adjusted coefficients of determination (adj R-2) > 0.59, with the lowest amount of variation in mean square errors noted for type 2-TLM followed by type 2 models. Cross-validation success (> 0.62) across most models (86% of all models) confirmed the agreement between ICE predicted and observed values. Despite differences in model predictive ability, most predicted values across all 3 ICE model types were within a 2-fold difference of the observed values. As a result, no statistically significant differences (p > 0.05) were found between most ICE-based and empirical species sensitivity distributions (SSDs). In most cases hazard concentrations were within or below the 95% confidence intervals of the direct-empirical SSD-based values, regardless of model choice. Interspecies correlation estimation-based 5th percentile (HC5) values showed a 200- to 900-fold increase as the log K-OW increased from 2 to 5.3. Results indicate that ICE models for aromatic compounds provide a statistically based approach for deriving conservative hazard estimates for protecting aquatic life. (C) 2015 SETAC