Regulatory Requirements for the Validation of Analytical Methods – Part 2

What does "suitable" method validation entail?

In particular, the aim of validation, i.e. proof of the suitability of the test method in routine use, must always be kept in mind. As far as possible, the routine conditions should be reflected in the experimental validation studies, for example by using samples of representative composition for testing the accuracy and lower limit of the working range, or by use of authentic samples in precision studies. Only then does the sample preparation, which often makes a significant contribution to precision, correspond to that used in batch release testing. If the routine conditions cannot be realized for experimental reasons, at least the possible influence on the assessment should be considered. Consideration of these "regular test conditions" is explicitly referred to in revision 2 of the validation guideline.

The user must identify the critical performance parameters for the test method and its use, determine these parameters experimentally in a reasonable and effective way and define sufficient acceptance criteria to confirm suitability.

Due to the iterative nature of test method development and validation, a rational design of the test procedure is the first prerequisite for optimal validation. A comprehensive understanding of the test procedure is a key objective of the new ICH Q14 guideline "Analytical Procedure Development", with a systematic quality-by-design approach (valid from June 2024).

Definition of acceptance criteria

To be able to assess whether a test method is suitable for the intended application, acceptance criteria must be defined. The definition of acceptance criteria was supplemented in Revision 2 of the validation guideline, but there is often a lack of guidance for derivation of the criteria. The correct requirement in the draft guideline (phase 2) that the precision must be compatible with the specification limits was removed in the final version of the guideline.

The acceptance limits of the specification represent the requirements for the respective quality attribute. For example, ICH guideline Q3A defines a required reporting limit for unknown impurities of 0.05 % or 0.03 %, depending on the daily dose of the respective active substance (≤ or > 2 g). Consequently, the test method must be able to quantify unknown impurities at this concentration, i.e. the reporting limit represents the maximum limit of quantitation.

The maximum permissible analytical variability for determinations of content is defined by the specification limits. Other performance parameters are method-specific and depend, for example, on the type of calibration, or on the experimental validation approach, but the random variability always represents the minimum contribution that must be considered.

Statistical significance tests check only for statistical significance, such as when using the mean value t-test of the difference between two mean values, which directly represents an acceptance criterion. Whether such a statistically significant difference has any practical significance is not the subject of the test.

Two scenarios can be distinguished here:

• Producer risk: On the one hand, very low variability or a large amount of data can quickly lead to statistical significance. In this case, we speak of a producer risk, i.e. a test method that is acceptable in itself that does not meet the acceptance criteria. By simply increasing the amount of data, any difference – no matter how small – will eventually become significant.
• Consumer risk: On the other hand, excessive variability can also mask practically significant differences, i.e. these are not tested as significant. In this case, one speaks of a consumer risk, i.e. poor performance is not detected, which can lead to the release of batches that do not conform to specifications.

This problem with significance tests is also pointed out in the USP general chapter on the validation of compendial methods <1225>: "Setting an acceptance criterion based on the lack of statistical significance ... is not an acceptable approach."

Statistical equivalence tests are recommended in ICH Q2(R2) for precision and accuracy: "The observed ... 100(1-α) % confidence interval (or justified alternative statistical ... interval should be compatible with the corresponding accuracy acceptance criteria". Here the question is – correctly stated – not whether the observed difference is statistically significant, but whether this difference (including its confidence interval as a measure of uncertainty) lies within a previously defined acceptance range. The latter introduces a measure of practical relevance. As confidence intervals become smaller with increasing numbers of data points and decreasing levels of variability, the probability of a positive test result increases, in contrast to the significance tests. This procedure is also described in the general USP information chapter <1010> as well in specialist literature and allows a (numerically) validated risk control via the α-error which is used. The inclusion of uncertainty (compared to a simple comparison of the calculated parameters with the acceptance criteria) requires either wider acceptance limits or a larger number of values to reduce the confidence intervals.

Unfortunately, no context is discussed in ICH Q2(R2), i.e. whether equivalence testing is expected in all cases. In the author's opinion, the comparison approach should be chosen depending on the risk of the quality attribute and the test procedure (complexity). If the risk is low, a simple comparison is perfectly acceptable.

Acceptance limits for precision

As the final result is compared with the specification limits (USP <General Notices>), its precision is decisive in determining the suitability of the test method. The specified limits for precision must include at least the manufacturing and analytical variability. The manufacturing variability can be estimated for a manufacturing technology type based on experience or determined by determining a larger number of batches in a measurement series using analysis of variance. The minimum requirement for the precision of the final result can be derived from the remaining available range, for example from the probabilities of the normal distribution. An acceptable probability to be within or outside (OOS) the specification limits serves as a measure of suitability.

For liquid chromatographic methods of content determination for active substances, there are specifications for the precision of injection in Ph.Eur. and USP. These depend on the number of injections and the difference between the upper specification limit of the active substance monograph and the theoretical upper limit of the true content 100 %, i.e. the available range for analytical variability. To ensure the same uncertainty, the acceptable precision decreases with decreasing number of injections. The random spread of standard deviations depends on the number of values (chi-squared distribution), but a general factor of 2 can be assumed between the true standard deviation and the upper limit of the distribution. The acceptance limit of 0.85 % for the injection precision from 6 injections and a 102 % upper specification limit would therefore be realistic up to a true RSD of about 0.4 %, but with 3 injections the true RSD must not exceed 0.2 %.

When quantifying impurities or other trace analytes, the strong concentration dependence of the precision must be taken into account. The reporting limit can be used here as a uniform reference concentration to ensure that the requirements are met, as a repeatability or intermediate precision of 25 % or 30 % allows acceptable quantification.