Issue link: https://www.e-digitaleditions.com/i/1171020
Pharmaceutical Technology REGULATORY SOURCEBOOK SEPTEMBER 2019 15 of IP that may be performed for low-risk methods involves conducting analyses on two different days by different analysts, with different equipment, etc. The ECA Academy (22) refers to this approach and evaluating the two sets of numbers obtained, for example comparing the means from the two sets of conditions and assessing the percent relative stan- dard deviation from the combined sets of numbers. It is acknowledged that the latter is the smallest re- source possible estimate of intermediate precision as it only includes two conditions. Simpler studies like this may be appropriate for low-risk methods if an appropriate risk assessment concludes that fac- tors related to intermediate precision are well un- derstood, controlled in the method, and likely to have a minimal effect or the expected variability is much less than any applicable criteria (e.g., precision criteria detailed in an analytical target profile [23]). However, an improved quantitative estimate of IP can be obtained from more than two independent analytical runs; Kojima (4) recommends at least six, giving five degrees of freedom at this precision level. Where this is applied, the following design principles are recommended: • At least six independent runs are required thus giving at least five degrees of freedom for IP. • Every IP factor (i.e., a factor that presents a medium or high risk for causing variability at this level) should be included in the experi- ment with at least two levels. A higher number of levels is better if possible/practical (often this is most efficiently achieved by performing a reproducibility study) as this allows more op- portunity for any special cause variation to be observed as well as a better estimate of the variation for the factor contributing most vari- ability to the intermediate precision variation. The runs should be balanced as much as possi- ble across the levels. • Each independent run should have the level of at least one IP factor changed and incur changes likely to happen from day to day; changing inter- mediate precision factors in combination as per factorial statistical design is preferred. • A balanced design that does not confound the main effect of the factors is preferred if assess- ment of individual factors is likely to be useful to try to understand—and perhaps reduce through further controls—the larger sources of variation. • The order of the runs should be randomized where possible/appropriate. Borman et al. (15) provides additional informa- tion on the structure of designs when used to assess method ruggedness. Multiple measurements may be made within a run, improving the estimate of the IP and provid- ing an estimate of average repeatability across the conditions; Kojima notes this could be sufficient for studying the repeatability validation characteristic (4). For the NIHS design there are six degrees of freedom for repeatability (4). Kojima discusses that increasing the measurements within a run only adds degrees of freedom for repeatability and states that this does not lead to accurate evaluation of intermediate precision (4). Though the degrees of freedom for intermedi- ate precision are not increased, replication within a run can give a better estimate for the condition being explored, especially if the repeatability variability is large. It can also be beneficial to evaluate several batches of material covering the range and different sample matrices for the method. Ideally the precision evaluation will be per- formed on homogeneous material. Kojima de-