Strategies to deal with information of different reliability exemplified by the use of QSARs to fill the algae data gaps in LCIAs of plastic additives
Konferensbidrag (offentliggjort, men ej förlagsutgivet), 2015
Data gaps are problematic when screening for
dangerous substances or in impact assessments where several chemicals are considered for evaluation. Lacking testing information can be replaced by non-testing information such as Quantitative Structure Activity Relationships (QSARs), but even though this latter information comes with lower reliability, this is seldom taken into account in the
forthcoming assessments. The difficulty to meet standards for best information calls for strategies to handle data gaps which take varying reliability in information into account. Using safety factors when reliability is low can be problematic since this result in more conservative evaluations of substances for which information is of low
reliability and an unknown level of risk aversion in the assessment. An alternative is to reflect lower reliability using probability distributions representing the expected error in the information and propagate this uncertainty in the forthcoming assessments using Monte Carlo analysis.
It is even possible to let the error to expect from QSARs depend to what extent a substance falls inside the models domain of applicability.QSARs cannot fill all gaps in data. Default values can be used instead of leaving substances out of assessments, but if so, these should reflect low
reliability as well. We demonstrate the practical implications of four strategies to handle varying reliability in information on algal toxicity in a Life Cycle Impact Assessment on 159 plastic additives of concern
using emissions from societal plastic materials in Sweden. A review concluded that a small amount of these substances had toxicity data for algae Pseudokirchneriella subcapitata. A QSAR was constructed which provided non-testing algal information of substances inside and on the
border of the models domain of applicability evaluated by PmodXPS.Substances with neither testing nor non-testing information were assigned default values. Screening based on characterization factors resulted in different rankings of substances when changing the level of cautiousness. The different strategies to handle varying reliability in
information do more or less open up for quantifying uncertainty in Life Cycle Impact Assessments.