Risk assessment for complex chemical exposure in aquatic systems: the problem of estimating interactive effects

The traditional, but little used, way of assessing effects of the interaction between known chemicals is to use factorial experimental designs. Such designs allow one to test for less than additive (antagonistic) and greater than additive (synergistic) effects. Whilst synergism can be demonstrated in such experiments the concentrations at which synergistic effects occur are extremely high and are unlikely to occur in nature. Recently developed techniques allow one to measure directly the effects of combined stressors in the field. These biological effect techniques range from tests on individual organisms to tests on communities. At the biochemical level the tests can indicate that the organism has been exposed to certain groups of chemicals (for example cytochrome P-450 enzymes responding to PAHs or metallothioneins responding to heavy metals). At the community level of organisation there are highly sensitive statistical techniques that indicate clearly the combined effect of stressors. The effects of oil exploration and production on benthic communities in the North Sea can be linked to concentrations of chemicals. However, such relationships are correlative and do not necessarily indicate cause and effect. Experiments are needed to test the hypotheses generated concerning the interactive effects of chemicals on the benthic species. The statistical analyses do, however, show which species have been affected and their relative sensitivity to chemical and physical disturbances. Such species are preferable to the traditional "laboratory weeds" usually utilised. A strategy for risk assessment is needed that combines an experimental protocol for making predictions, from laboratory experiments, of likely effects to be found in the field. This should be combined with field monitoring that allows one to detect changes that were not predicted. At present most monitoring designs cannot adequately detect trends. This is due to concentration on Type-I statistical errors rather than properly considering Type-II errors. By concentrating on Type-II errors one can design monitoring programmes that are able to detect trends with a given degree of precision. There are also strong ethical grounds for a change to giving more emphasis to Type-II errors.