Estimation of the Acute Dietary Exposure to Pesticides Using the Probabilistic Approach and the Point Estimate Methodology.

ABSTRACT

Aim: This report deals with the generation of work examples using food consumption data from the Netherlands and Sweden to facilitate the understanding of probabilistic modeling of dietary exposure to pesticides by risk managers. Imaginary field trial residue data were invented to be as useful as possible to show the potential of probabilistic modeling. Via seven work examples different aspects of acute dietary exposure assessment to pesticides and the use of the probabilistic approach were addressed. Approach and Results: In work example 1 the point estimate approach was compared with the probabilistic approach using the same input data. This work example showed that the point estimate resulted in higher estimations of exposure compared to the probabilistic approach, resulting from the conservative assumptions underlying the point estimate (one high level for both consumption and residue and a high default variability factor). It was demonstrated that with the probabilistic approach all consumption levels and field trial residue levels available could be used in one analysis, resulting in a better use of the available data. Also all foods contributing to the exposure could be addressed simultaneously in one analysis, as opposed to only one food at a time in the point estimate. In this way a more holistic approach to risk is possible. Work example 2 addressed the concept of ‘consumers only’. The example showed that, when exposure could occur via the consumption of more than one food, this concept could result in risk estimates that are difficult to interpret in probabilistic modeling, because the underlying database contains both consumers and non-consumers of the specific food items. This may hamper a clear risk management decision. It was also shown, when addressing only the consumers, that the exposure was influenced by the percentage of the population that consumes such a product. When that percentage was very low the exposure increased more compared to the situation where the whole population (consumers and non-consumers) was considered (e.g. kiwi fruit) than when the food was consumed by a large majority of the population (e.g. apple in The Netherlands or apple/pear in Sweden). The effect of processing on the acute exposure assessment was demonstrated in work example 3, showing clearly that processing influenced the level of exposure. It was demonstrated that with the probabilistic approach the risk assessor can address different types of food processing simultaneously with each food item being linked to the correct variability factor (e.g. variability applied when addressing apples and pears eaten whole, but not when addressing apples and pears mixed in juices and nectars). This is not possible with the point estimate. In work example 4 the effect of introducing variability in different ways into the probabilistic exposure assessment was studied. This resulted in different outcomes. The beta assumption on distribution reflects best what happens in real life (residue levels of individual units can be lower, equal or higher than the corresponding composite sample level) and the maximum residue level to be sampled is bounded at an upper level. The meaning of different endpoints in the acute exposure distribution created by probabilistic assessment was addressed in work example 5. It was argued that when a certain critical percentile of exposure exceeds the acute reference dose (ARfD) a critical examination of the exposures contributing to this, by studying for example the highest ten exposure levels with their corresponding consumption and residue levels, would be very useful, giving the risk manager an insight in the reliability of the upper percentiles. Related to this it was argued that the uncertainty factors used to derive an ARfD should be considered in relation to the occurrence probability of the exposure exceeding the ARfD. When this probability is very low (e.g. lower than 10-4 or 10-5) one can argue whether such a risk is acceptable or not. In the last two work examples the stability of the tail of the distribution was examined. In work example 6 this was done by examining the effect of the number of iterations on the upper percentiles of the exposure distribution. The results showed that the number of iterations should be sufficient for making a confident estimation of a certain percentile. The number of iterations depends on the amount of consumption data and residue levels available, and on the percentile of interest. The more data available the more combinations will be possible of food consumption and residue levels, resulting in a need for a higher number of iterations to describe the exposure within a population adequately. Also more iterations will be necessary to reliably estimate higher percentiles of the exposure distribution (≥ P99.9). In work example 7 the stability of the tail was examined by studying the effect of the presence in the food consumption database of consumers with an extreme food consumption pattern (e.g. an infant consuming 2 kg of apples) or the presence of an outlier in the residue database. It was evident that an outlier (high consumption or residue level) affected the result of an exposure assessment depending on the magnitude of the outlier compared to the other data present in the data set and on the largeness of the data set available. The effect was more evident on the P99.99 and maximum exposure level simulated than on the P99.9. It was argued that when outliers are present in the data, it is always important to visualise them (quality check on the data, for example related to reporting mistakes in the food consumption database), and to discuss the meaning of these outliers on a possible decision about the compound addressed. For example, it can be argued to what extent individuals with extreme dietary habits should be protected or that general advices concerning healthy eating habits should suffice. Conclusion: This report demonstrates clearly the potential of the probabilistic approach when dealing with acute dietary exposure assessment of pesticide residues compared to the current methodology used. The probabilistic approach to assess dietary exposure is also applicable to data from other countries, and, more importantly, others can be trained to perform risk assessments with the same model using their own data. Different aspects of an exposure assessment were addressed to help risk managers to understand better how exposures are calculated and to interpret the results of a probabilistic exposure assessment. The complete report can be downloaded for free from http//edepot.wur.nl/28647. Full report is also available as ‘Supplementary File’.