AOAC Guidance on FA Immunoassay Validation (August 2023)

Annex C

1325

Statistical Methods: Data Analysis Guidance and Example Datasets 1326 1327 1. Intermediate Precision and Repeatability Estimation from Nested Designs: Analysis of Nested SLV 1328 Designs in R (Courtesy of Paul Wehling, ChemStats Consulting, LLC) 1329 1.1. As is described in Section 5.2.4.2 of the guidance, intermediate precision and repeatability can 1330 both be estimated from one of several nested designs. 1331 1.2. Basic Principles of the Nested Designs 1332 1.2.1. Defining the Variance Components: 1333 1.2.1.1. When validating a method with a nested experiment, it is strongly 1334 recommended that researchers define terms used to describe the experimental 1335 factors. Because all methods are different, and researchers tend to use different 1336 words to convey the same meaning, it is important to define terms in order to avoid 1337 confusion. For example, in the largest design in the Guidance, Design 2b, there are 1338 potentially 4 levels of experimental factors that can be differentiated and estimated: 1339 Lot, Analyst/Day, TP, and ELISA. Now in all designs, 1a, 1b, 2a, 2b, there is an explicit 1340 understanding that Analyst and Day are confounded and will be included in the 1341 model as a single factor. In addition, each of these levels may have many more 1342 sources of variation than just those given by the 4 terms used. It is recommend to 1343 explicitly write out the sources of variation and how they contribute to the 4 1344 variance components that will be estimated experimentally. Nested experiments are 1345 unique in this aspect. Generally, with a factorial experiment, you can control the 1346 conditions so that only the interested factors are varied. 1347 1.2.1.2. Terminology: 1348 1.2.1.2.1. “Source of variance” refers to a specific source of variation in the 1349 method for example, weighing variation. This refers to all of the small sources 1350 of variation that add together to make the overall measurement uncertainty. 1351 1.2.1.2.2. “Variance component” is a statistical term for a collection of one or 1352 more sources of variation that will be estimated by the validation experiment. 1353 In this case, we will have 4 variance components. The purpose of this exercise 1354 is to take all of the known sources of variation and assign them to one of the 4 1355 variance components. The distribution of sources of variation depends on the 1356 experimental conditions and how the analyses were performed. 1357 1.2.1.3. Example of variance component description for a nested experiment of a typical 1358 ELISA method. 1359 1.2.1.3.1. Note: the following are for a hypothetical ELISA method - ALL METHODS 1360 ARE UNIQUE and will be different – this should be performed for each method 1361 and each validation. 1362 1.2.1.3.2. LOT Includes: manufacturing variance of the lot, potentially different 1363 response of antibodies. Certain reagents are unique to each lot, so there will be 1364 reagent variance. 1365