1. Sources of uncertainty and current practice for addressing them: analytical perspective Roy Macarthur roy.macarthur@fera.gsi.gov.uk

2. Analytical perspective Uncertainty about quantitative measurements Chemical, microbiological, biotechnological analytes. What analysts won’t tell you What analysts will tell you but you shouldn’t necessarily take on face value Uncertainty about qualitative detection

3. Quantitative analytical uncertainty A quantity associated with an analytical result. Expresses the range of values for true concentration in a sample that are consistent with the measurement result. Estimation of uncertainty is required by ISO 17025. Samples taken for official control ‘fail’ if result - uncertainty is over limit (Codex, EU). Upper limits for expected measurement uncertainty in European Legislation are used to qualify analytical methods for official control.

4. Options for estimating uncertainty Produce a model of the method, combine components of uncertainty Use real measurement results Use a general estimate based on usual performance

5. Options for estimating uncertainty

6. Using real measurement results Single laboratory validation and QA Measurements undertaken within a laboratory to estimate or check variation, bias etc. Collaborative trial. Analysis of replicate samples in different labs using the same method. Proficiency tests Analysis of replicate samples by different labs using any method they like.

7. Using real measurement results Single laboratory validation IUPAC 2002 “validation is...the task of estimating uncertainty” Collaborative trial. The expected variation of results across labs using that method Proficiency test results produced by a lab using the method Estimate of uncertainty for results produced by that lab using that method Proficiency test results produced by all labs using all methods Estimate of variation between all labs and methods

8. The size of analytical uncertainty depends on what we know

9. General estimates Sector Size: of 95% confidence interval Basis Chemical > 0.12 mg/kg Variable, e.g. ± 32% at 1 mg/kg Collaborative trials, Horwitz (1980), proficiency test results Chemical < 0.12 mg/kg ± 44%Collaborative trials, Thompson (2000), proficiency test results

10. General estimates Sector Size: of 95% confidence interval Basis MicrobiologicalFactor of x3 Common practice, proficiency test results Biotechnological Factor of x2 Collaborative trials of whole methods, Macarthur, Feinberg (2010) Factor of x3Proficiency test results

11. General estimates

12. What analysts won’t tell you - sampling

13. What is the effect of sampling uncertainty? Depends on the ‘size’ of variation –Expressed as a variance, RSD, or hotspot size Depends on the ‘shape’ of the variation –Expressed as a particular statistical distribution. We rarely know the size. We never know the shape. Size is more important than shape.

14. Effect of hotspots on measurement: additional uncertainty

15. Effect of sampling on detection: greatly increased limit of detection n: number of samples R: sampling RSD V: total volume sampled r: analytical false negative rate

16. What analysts won’t tell you - blunders Results produced when an analysis is not done properly Includes testing the wrong sample, transcribing errors etc Generally ignored in collaborative trials, ISO5725 allows 2/9 results to be discarded. Considered not to be part of analytical uncertainty.

17. Proficiency test results reveal blunders About 5% of proficiency test results appear to be blunders Royal Society of Chemistry Analytical Methods Committee, Technical Brief 49 Cadmium in food stuffs (mg/kg)

18. What analysts will tell you....LOD Different definitions and methods for calculating limit of detection can lead to estimates varying by a factor of 1000 (Lloyd Currie, 1968) At least 8 current definitions for limit of detection (MoniQA* project, last week). *MoniQA, “Monitoring and Quality Assurance in the Food Supply Chain”, European Commission funded network of excellence. www.moniqa.org Check definition used.

19. Typical performance for trace analysis

20. Critical level (often incorrectly called limit of detection)

21. Limit of detection (sometimes incorrectly used as a critical level)

22. Qualitative methods of detection Increase in low cost, rapid methods of detection. No internationally accepted standard (correct, simple, broadly applicable) for validation and uncertainty. IUPAC / MoniQA and AOAC standards under development. Based on estimating probability of detection across a range of concentrations.

23. Validation of a qualitative method of detection Concentration (%MBM)

24. Analytical perspective Use, proficiency test results, collaborative trials and single lab validation or “general estimates” for analytical uncertainty. Variation of proficiency test results gives a good estimate when little information is available Chemical <±44%, bio ×3, biotech ×2 to ×3 Be aware of sampling uncertainty and potential blunders Get basis for claimed limits of detection.