Extending the SSD concept to explore some foundational model limitations: a Bayesian hierarchical approach



Hickey, Graeme ORCID: 0000-0002-4989-0054, Craig, Peter S, Marshall, Stuart, Price, Oliver, Mathijs, Smit, Hart, Andy, Luttik, Robert, Chapman, Peter and De Zwart, Dick
(2011) Extending the SSD concept to explore some foundational model limitations: a Bayesian hierarchical approach. 2011, Milan, Italy.

[img] Text
Platform.pdf - Unspecified

Download (3MB)

Abstract

Species sensitivity distributions are statistical constructs which model interspecies variation of sensitivity to a particular toxic stressor. The current REACH technical guidance document permits the application of SSDs in risk assessment, subject to a number of criteria. Notwithstanding noteworthy criticism received, the SSD is considered by regulators to be a pragmatic model for extrapolating to environmental toxicant concentrations of concern. The manner in which SSDs are currently applied is implicitly dependent on a number of (overlapping) statistical and ecological assumptions. These include (but are not limited to): (1) the measured species toxicity values being precisely known; (2) independence of SSDs for each separate chemical risk assessment; (3) a priori exchangeability of species toxicity values; (4) no correlation between species. In this research we propose a model which generalizes the SSD concept to include chemical effects and shared species effects. It offers flexibility to address or refine each assumption by hierarchically adding layers into the model. Models are fitted to RIVM and US EPA acute-effect toxicity databases under a Bayesian statistical framework to allow for transparent quantification of and flexible propagation of uncertainty. Important insight is gained from the inclusion of ‘species effects’ modelling which, expectedly, indicates increasing differences as taxonomic distances in SSDs increase. The magnitude of measurement error estimated, based on within taxa homogeneity, which also properly accounts for censored measurements, is likely to be of significance to risk assessors and warrant further consideration in either modelling framework. The current quasi-meta-analysis approach towards aggregating multiple chemical-species data points is untenable from an uncertainty viewpoint. Initial results indicate deficiencies in the current SSD concept, thus reducing the credibility and meaningfulness of any subsequently derived hazardous concentrations. Other recent model proposals which act as precursory tools to SSD modelling may not sensibly propagate uncertainty and/or succumb to modelling contradictions. A hierarchical model may overcome this, however will require a more radical approach to defining protection goals and environmental concentrations of concern.

Item Type: Presentation Material
Depositing User: Symplectic Admin
Date Deposited: 27 May 2015 08:12
Last Modified: 02 Apr 2021 07:17
URI: https://livrepository.liverpool.ac.uk/id/eprint/2012058