Overview on the statistically-based methods for establishing MRLs based on supervised field trial data

Maximum residue limits (MRLs) are legal limits for residue concentrations of pesticides on/in foods and feeds.MRLs are intended primarily as a check that the GAP is being followed and to assist international trade in importing and exporting produce treated with pesticides.Therefore the methods used for setting these MRLs were also of great concern.Several statistically-based calculation methods that had been developed and available to the international regulatory community were reviewed.The technical details of each of these methods are provided followed by a discussion of their merits and shortcomings.For example,the EU methodⅠwas in use and accepted internationally in the EU and the method could result in consistent estimates when the same or similar data was used.However,this method assumes that pesticide residues data are distributed normally and the EC provides no recommendation for testing the validity of this assumption even though the calculation produces biased estimates if the normality assumption is not met.As for EU Method Ⅱ,one advantage of the calculation used in this method is that it is not based on any distributional assumptions.And the calculation involved is generally not influenced by the presences of high values in the data set.However,when using Method Ⅱ,the estimated MRLs is unaffected by the upper quartile of the data set.Additionally,there is no statistical support provided for the rationale of this method.NAFTA Lognormal Method,is the modified version of EU Method Ⅰ ,it is also referred to as the "95/99 Rule".The method could produce unbiased estimates of the 95% UCL of the 95th percentile for the data that are distributed lognormally.However,the method requires that the data set approximate a lognormal distribution;that the data set be of sufficient size such that a reasonable UCL on the 95th percentile or point estimate of the 99th percentile can be estimated;and that there be a sufficient number or proportion of detectable ( >LOD) residue values such that the method can be used.The OECD working group has recently dropped the distributional approaches from the calculator.The precision of the OECD calculator method is increasing as the dataset size increases and the method can provide reasonably good estimates of high percentiles like the 95th for datasets that have at least between 15 and 20 points.The specific method now being explored is the "Mean+4×Standard Deviation" method,where the number of standard deviations taken is increased to four to get a better estimate of the 95th percentile of a lognormal distribution.