Keith R. Hayes
CSIRO Division of Marine Research, 2004
- Full report - Best practice and current practice in ecological risk assessment for genetically modified organisms (ecological-risk.pdf - 542 KB)
- Summary of best practice - Best practice ecological risk assessment for genetically modified organisms (bestpractice-summary.pdf - 542 KB)
This report compares current practice in ecological risk assessment for genetically modified (GM) plants and microorganisms, as evidenced by eight transnational and national frameworks, with what might reasonably be considered best practice. Best practice is defined for the scientific principles, hazard identification, risk calculation, social appraisal and monitoring stages of an ideal ecological risk assessment, and summarised in the following ten points:
- Carefully define measurement and assessment endpoints for environmental values for each stage of a genetically modified organism (GMO) release;
- Construct good qualitative models of all hazard scenarios using structured deductive and inductive hazard assessment techniques;
- Consider the influence of cognitive bias, framing effects, anchoring and sample size on qualitative decisions;
- Consider the full spectrum of ecological models from simple (screening) to detailed ecosystem models;
- Recognise that even simple models can incorporate uncertainty and be useful in ecological risk assessment;
- It is essential to include a transparent analysis of uncertainty;
- When information is sparse use probability bounding analysis to express uncertainty;
- Examine opportunities to promote appropriate and ongoing stake-holder participation in the risk assessment;
- Adopt a precautionary approach to high consequence, but highly uncertain, hazards; and,
- Consider statistical power, effect size and model based sensitivity analysis and other remedies to hidden conventional pitfalls in monitoring.
Most of the frameworks reviewed here provide evidence of best practice in the scientific principles and frameworks of ecological risk assessment. All of them, however, rely on simple checklists in the hazard identification stage, and only one discusses inductive techniques. Hazard identification as currently practiced is therefore largely restricted to prescriptive deductive techniques. Analysts will identify a larger range of potential hazards, and gain a better understanding of the event chains associated with these hazards, if they used inductive hazard identifications techniques.
Well-corroborated quantitative techniques exist for some of the potential hazards associated with GMO field release. However, despite the rich scientific literature on quantitative techniques and models, only one framework bridges the divide between science and regulation by identifying specific experimental techniques and models in the regulatory process. Some of the regulatory frameworks recognise that quantitative approaches are possible in certain circumstances, but neither the circumstances (i.e. which hazards) nor available techniques are identified. For the main part it is not clear when and how quantitative techniques are expected of the applicant.
Regulators can assist quantitative risk assessment by helping proponents identify models and analysis techniques relative to specific GMO hazards. Regulators should insist that proponents obtain the necessary data and information in order to achieve best practice and to reduce areas of significant uncertainty. Current field trials only appear to gather information on crop performance. These trials are an ideal opportunity to gather the types of data needed to improve the science of GMO risk assessment.
Quantitative techniques are not currently available for all of the potential hazards associated with GMOs. There are important gaps in the following areas: food-wed and trophic interactions, the transfer of viral genetic material to other viruses, increases in the host range of viruses, fungi and other pathogens, altered farm practice and physical habitat changes. National regulatory authorities should encourage data collection and research in these areas. High consequence, high uncertainty impacts (such as the creation of new viruses) are unlikely to be satisfactorily addressed by quantitative techniques in the near future. More rigorous qualitative techniques, however, including a wider social discourse and directed research, are achievable in the near term.
The degree of practicality, reliability and acceptance of quantitative techniques for less uncertain hazard scenarios varies from model to model. In general terms simple models are the most widely accepted and, when used in conjunction with a rigorous analysis of uncertainty, can provide meaningful answers for risk assessment purposes. Qualitative assessments are often recommended as an initial screen to eliminate low risk events from a potentially lengthy assessment process. This review, however, suggests quite the opposite: simple quantitative techniques should be used wherever possible to screen high and low risk scenarios-qualitative assessments become most important for highly uncertain but potentially high impact scenarios. To be successful these qualitative assessment should include a strong element of social appraisal including, for example, the use of systematic hazard identification techniques to capture the imagination and intuition of non-scientific 'experts'.
Uncertainty analysis is the very rationale of risk assessment, and yet this is by far the weakest component of current practice. None of the frameworks reviewed here, bar two, require a formal analysis of uncertainty as part of the risk assessment process. This is arguably the biggest gap between current practice and best practice in ecological risk assessment for GM plants and microorganisms. Well-established statistical techniques exist to describe random measurement error and environmental variability. Model error can be approached by ground-truthing risk assessment predictions and testing alternative model formulations. Techniques also exist that bridge the divide between qualitative and quantitative approaches to risk assessment, and thereby facilitate a progression from one to the other.
All of the frameworks reviewed here discuss or at least mention monitoring but none point to best practice in this area. All of the frameworks could be improved by drawing the analyst's attention to power calculations for typical monitoring strategies. Monitoring strategies will need to continue well beyond the usual period needed to assess the efficacy of the phenotype in order to detect potential ecological impacts. It is important that these strategies test the predictions of prior risk assessments and provide information that will inform future risk assessments, thereby "closing the regulatory loop". These strategies must explicitly include an appropriate power analysis to avoid blindness to Type II error.