System Suitability. The test system must be properly controlled to ensure reliable release-testing results. The system suitability criteria should be established during the AMD and optimization phase. This is usually accomplished by running a set of control points. For each test, system suitability will be satisfied (valid test results generated) if all control points are in established limits. A test system must be able to reproduce measurable results of a homogeneous sample (control) to allow examination of differences between batches. Small differences in batches are normal and acceptable, but the sources of variation should be identified. Ultimately, we will have more certainty when we can separate differences in production batches from assay variability.²⁰

Sample Suitability. Technically, sample suitability is part of system suitability so these parameters can be evaluated together. Sample suitability should be established during AMD and should ideally ensure that samples, controls, and standards are prepared identically and run simultaneously. In addition, sample suitability should include a statistical analysis of the number of replicates needed to generate significant release results. Single measurements may be acceptable if the production-process sampling can deliver truly batch-representative samples and the precision of assay repeatability is high compared to the product specifications—and therefore high compared to the batch-to-batch variation on which these specifications are based. Often, however, assay precision is relatively low, and multiple measurements will substantially increase the level of certainty in the corresponding test results. Several detailed standard practices are published by ASTM.^15–19

Also, we should evaluate batch sampling to ensure capturing and accounting for apparent variability. For example, product potency in final containers may vary among samples taken from the beginning, middle, and end of fill due to protein adsorption or protein aggregation during fill. In this case, samples should be taken from each of the three fill stages to ensure a more accurate representation of the batch's average potency.²⁰

Sometimes, data transformation (for example, logarithmic conversion) may lead to improved linearity. However, most biological assay response curves are not linear, even after mathematical transformation.²⁰ These are particularly difficult to deal with and should only be handled by experienced statisticians. Nonlinear models, such as four or five-parameter logistic functions with weighted factors and tests for parallelism, may be the best approach.²⁰ Just as different test methodologies have different biases, changing statistical models may significantly change the final results. Some models may simply be inappropriate or may not provide acceptable results over the entire assay range.²⁰

Robustness. Robustness, defined as the lack of a significant effect when small changes are deliberately introduced into the test system, should ideally be addressed during the method optimization phase and not as part of AMV. We should know the degree of robustness of a method before starting the formal AMV phase. Critical test system parameters (for example, the acceptable range of diluting the test sample) must be identified and controlled with appropriate operational limits. These limits should be described in the AMD report and documented in the method SOP. The SOP will then contain operational limits which are in the context of the overall system suitability criteria and which are adhered to during the validation phase. In addition, robustness should be tested in the AMD phase during or after method optimization because significant differences in the AMV results (from challenging the critical operational limits) may be difficult to explain in the AMV report.²⁰

We must remember that AMV is the formal evidence that this method is suitable to be used under strictly controlled QC testing conditions. The AMV protocol should be set up to deliver this evidence through appropriate acceptance criteria by varying sample batches and concentrations, operators, instruments, days, and other factors that are expected to vary during routine testing—in established sample and system suitability conditions and operational limits.²⁰