A statistical evaluation of the safety factor and species sensitivity distribution approaches to deriving environmental quality guidelines

Abstract

The species sensitivity distribution (SSD) distribution approach to estimating water quality guidelines (WQGs) is the preferred method in all jurisdictions reviewed (Australia, Canada, New Zealand, Organisation for Economic Co-operation and Development [OECD] members, South Africa, United States) and is one of the recommended methods for European Commission members for 33 priority and priority hazardous substances. In the event that jurisdiction-specific criteria for data quality, quantity, and taxonomic representation are not met, all of these jurisdictions endorse the use of additional safety factors (SFs) applied to either the SSD-based WQG or, the lowest suitable toxicity test endpoint. In Canada, the British Columbia Ministry of Environment endorses this latter approach as the preferred approach in the belief that so-derived WQGs are more protective than SSD-based WQGs. The level of protection afforded by the latter SF approach was evaluated by statistically sampling minima from random samples of the following distributions: normal, Gumbel, logistic, and Weibull, using a range of coefficients of variation (cVs) and applying the SFs of 2 or 10 used in British Columbia. The simulations indicate that the potentially affected fraction of species (PAF) can be as high as 20%, or, approach 0%. The PAF varies with sample size and CV. Because CVs can vary systematically with mode of toxic action, the PAF using SF-based WQGs can also vary systematically with analyte class. The varying levels of protection afforded by SF-based WQGs are generally inconsistent with the common water quality management goal that allows for a small degree of change under long-term exposure. The findings suggest that further efforts be made to develop high-quality WQGs that support informed decision making and are consistent with the environmental management goal instead of using SFs in the hope of achieving an acceptable but unknown, degree of environmental protection.

Recent Developments in Species Sensitivity Distribution Modeling

“…Following its introduction in the 1980s (Stephan et al 1985; Kooijman 1987; van Straalen and Denneman 1989), the SSD has remained the most widely used method for deriving water quality benchmarks (guidelines, criteria, or standards, depending on the jurisdiction) to characterize effects of chemical contaminants on water quality and/or for ecological risk assessment purposes. The SSD has proved to be a useful, practical, and intuitive tool (Belanger et al 2017; Belanger and Carr 2019), albeit not without numerous limitations (e.g., Organisation for Economic Co‐operation and Development 1992; Forbes and Forbes 1993; Smith and Cairns 1993; Warne 1998; Newman et al 2000; Forbes and Calow 2002; Wheeler et al 2002a, 2002b; Zajdlik 2006; Hickey and Craig 2012; European Centre for Ecotoxicology and Toxicology of Chemicals 2014), including the implausibility of the many assumptions underpinning SSDs and concerns arising from inconsistent statistical results. Despite a significant body of published research and numerous intensive reviews (e.g., Organisation for Economic Co‐operation and Development 1992; Posthuma et al 2002; European Centre for Ecotoxicology and Toxicology of Chemicals 2014; Fisher et al 2019) over the past 20 yr aimed at improving SSD methods, the fundamental SSD approach employed by jurisdictions around the world has remained similar.…”

Advancing Fifth Percentile Hazard Concentration Estimation Using Toxicity-Normalized Species Sensitivity Distributions

“…Traditionally, uncertainty in conventional HC5 estimates has not been reported or only the confidence region of the SSD is shown. Various statistical approaches to estimating HC5 uncertainty are available but have not been generally adopted in part because of their computational complexity. ,− We developed a simple method of computing a confidence interval around the HC5 using an adaptation of LOO (N-1) error estimation. , Across all four chemical-taxa groups, LOO estimates were highly accurate and confidence intervals were extremely precise compared to the conventional SSD approach. LOO variance estimation has practical advantages over conventional SSD error estimation.…”

Assessment of a New Approach Method for Grouped Chemical Hazard Estimation: The Toxicity-Normalized Species Sensitivity Distribution (SSDn)

“…Acute toxicity values for a given chemical and species often vary by magnitudes up to 5-fold for aquatic animals across different laboratories . Safety factors, often ranging from 2 to 10, are also commonly applied to HC5 values as a precautionary measure to account for uncertainty . Therefore, SSDn HC5 values falling within 1- to 5-fold of the single-chemical SSD HC5 were considered to be “good,” and those falling within 5- to 10-fold were considered to be “reasonable.” SSDn HC5 values that were more than 10-fold different from the single-chemical SSD HC5 were categorized as < or >50-fold different to provide a sense of scale regarding the magnitude of variance between the SSDn and SSD methods.…”