AOAC Gluten Quantitative Validation Guidance-Round 1(Nov 2023)

ELISA includes: aliquot variation, heterogeneity in the extract, reagent pipetting variance, differences in coating of the wells, well-to-well sensitivity variation, rinsing issues, pipetting volumes, different optical density of each well, reader issues, timing of color development, how fast you pipet from start to finish, different development times across the plate. 47 In order to make the software work, you need to give a name to each of the 4 variance components, 48 with the understanding that there will be several sources of variation within each variance component 49 category. I suppose you could call them Level 1, Level 2, Level 3, Level 4, but the usual way to do this is 50 to take what you think is the most important source and use it as the “name” of the variance 51 component – keeping in mind that the name is only a label and if you call the 3 rd level “Test Portion” 52 that doesn’t mean that all those other sources are gone – this is just the Label we are using for 53 convenience. (In these experiments, “Test Portion” will usually always include extraction sources as 54 well.) 55 It’s critical to do the categorization of variance sources into variance components for 2 reasons: First, it 56 is important to define terms, but more importantly, it will come in handy to determine if the factor is 57 nested. 58 Please note – if you only do the 3-factor experiments such as Design 1a or 1b, the variance components 59 above labeled as “TP” and “ELISA” will be combined into 1 component. So, you may call that combined 60 component “TP”, but it will contain all of the ELISA variance sources in addition to the other sources. 61 (Maybe “TP” is not a good name for that in the 3-factor design.) Researchers are free to use any label 62 for the name of the variance component, but this should always be understood that there are more 63 sources of variation within a variance component than the one that is used as the label. 64 What is a “Nested” experiment? When can we consider one factor to be “nested” within another 65 factor? 66 Nested experiments are ones where you may have two or more factors involved and you have a 67 hierarchical order of nesting of factors. This would be different from a factorial design where the factors 68 are varied independently, and the conditions for one factor can be adjusted to be the same at all the 69 other factor levels. In the case where we are doing a variance component analysis of a method take for 70 example the factor “Test Portion.” Because each test portion is destroyed in the extraction, we can’t 71 really have the exact same test portions for kit Lot 1 as kit Lot 2, so TP will always be a factor nested 72 within some other higher level factor. In the same way, we pipet each extract into 2 wells on the plate to 73 estimate well-to well ELISA variance, since the 2 wells that are used for extract #3 cannot be reused for 74 extract #4, again the factor ELISA is nested within the TP factor. Statisticians will say that for a factor to 75 be nested, there needs to be a significant “separation” in that factor across the different levels of the 76 factor one level higher in the hierarchy. Separation is achieved because the test portion is destroyed 77 and can’t be recovered. If a factor is not nested then we say (some authors use this terminology) that 78 the factors are “crossed”, meaning they need to be treated as a factorial design, such as a “2x2” 79 factorial. It does not mean to imply there are interactions fitted in the model. To avoid this confusion, 80 some authors refer to these 2 factors as “Main Effects.” The area where this will be difficult in these 81 validation designs is the level that includes Analyst/Day/Calibration. For each method and experimental 82 design, we will need to determine if the Analyst/Day factor can be considered nested within the Lot 83 factor, or if there is inadequate separation between Analyst/Days for one lot to another and so will have 84 to be considered as 2 main effects. To make this easy, Lot will always be a Main Effect at the top of the 85 hierarchy, and TP and ELISA (if replicated) will always be nested. The other easy thing is that the ANOVA 86 calculations in R are simple, and R can do the analysis either way, with a minor change to the code. 87 Proposed Decision Rules for Determining Nested Variables 44 45 46

88 89

Design

# Analysts

# Days

# Calibrations

Adequate Separation?

Factor Is: