Which percentile?

[Edel Duffy (Creme)]

If probabilistic modelling is to be used to estimate exposure to a chemical so that the result can be compared with a safety level what percentile value should be chosen to compare with the safety level and should this value always be the mean of the percentile or an upper value of the percentile.

If probabilistic exposure assessments are to be used should there be European concensus for which percentile to use?

[Enda Cummins (UCD)]

I guess this is where risk management comes in, it is up to the risk managers to set the acceptable level, and a associated acceptable level of uncertainty. It would be nice to have a european concensus on which percentile to use, I guess the 95 percentile is the accepted norm. I dont like when the mean is used, as in these situations, it is always in the tails that the problem arrises (the devil is in the (de-)tail)

[Cronan McNamara (Creme)]

Yes, I agree that this is a job for the risk manager.

A nice feature of probabilistic modelling is that a range of the percentiles can be given for any demographic group in the population. These results are also reported with confidence intervals. The risk manager is then free to use that information as she sees fit.

I am always concerned when I see only an average population intake result produced, also when the confidence intervals in the result are not given. How do others feel about this?

[Christina Cowan-Ellsberry (P&G)]

I have been extensively involved in better understanding how probabilistic and probabilistic-aggregate exposure approaches can be applied to consumer products. As I have explored the question of which percentile of the exposure to use, I have had to go back to the data that is used to understand exposure to consumer products. Typically these data come from diary studies where products are given to a person and they are asked to record how often the product is used. The difference in the weight of the product at the beginning and end of the study is also collected. It has been found that in these situations, there may be errors in the record of how often the product is used, and very often the user will use more than they do for a product they purchased themselves. Also if they like the product, they will take some out before returning it. This plus the errors in recall that happen when consumers are asked to recall past use are well recognized in the consumer knowledge community. Given these potential sources of error and biases that are often introduced, typically the higher percentiles, for sure above 90th are treated with some suspicion. We have more work on-going to quantify the level of error in these type of studies in hopes of making a more definitive recommendation. Please let me know if you have any additional ideas or thoughts.