The ICH-Q2 has stated that the objective of validation of an analytical procedure is to demonstrate that it is suitable for its intended purpose, yet it – as well as the ICH-Q14 – fails to clearly define the actual aim of the analytical procedure, leading to misunderstanding and confusion. Bruno Boulanger, Ph.D., Global Head Statistics and Data Science at PharmaLex, discusses how Bayesian statistics and interpretation bridge the gaps in the guidelines.
In early 2022, ICH released two important guidelines for comments: ICH-Q2 “Validation of Analytical Procedures” and ICH-Q14 “Analytical procedure development”, both of which are closely interconnected through the concept of the Analytical Target Profile (ATP), which is used to assess the quality of results generated by analytical methods. The ATP and ICH-Q14 are intentionally more aligned with the Quality-by-Design (QbD) ICH-Q8 document on process development, qualification and control. This introduces the concept of QbD applied to analytical procedures (AQbD).
But, as opposed to ICH-Q8 which starts with the target product profile (TPP) or the properties that a product and its related process should achieve, both ICH-Q2 and Q14 miss the central point that applies to all analytical procedures: defining what is a good fit for-purpose measurement or reportable value. The fact that the actual aim of the analytical procedure is not clearly defined causes misunderstanding and confusion about various concepts such as accuracy, linearity and range.
Bayesian statistics and interpretation address these misunderstandings, helping to bridge the gaps that exist in these two guidelines.
Uncertainty of Measurement
The ICH-Q2 (R1) describes the linearity of an analytical procedure as “its ability (within a given range) to obtain test results which are directly proportional to the concentration (amount) of analyte in the sample”.1 The idea is that by increasing the quantity or the potency, the result will be proportional to the increase. Trust in the result during routine use of a validated method is the concept of the uncertainty of measurement – a concept routed in Bayesian statistical theory. Applying Bayesian statistics allows the user to take results obtained during the validation of the analytical procedure and predict the uncertainty around any future result.
In pharmaceutical manufacturing, it is this concept of uncertainty, or Target Measurement Uncertainty, that allows the user to determine with a high probability (for example 95%) that the true value of the batch or sample is within a pre-specified quality range. So, if an analytical procedure is used to release a batch of a drug product and the specification limits are that it must fall within +-2mg, then the uncertainty associated with a result should be much smaller than the +-2mg in order to keep the risk acceptable. (See figure 1). In keeping with Six Sigma thinking, the rule of thumb is that TMU should not be greater than one-sixth of the specifications of the product to reduce the risk of making a wrong decision.
Going back to the concept of AQbD, elements of which are employed in the Q14, the point is to be sure that for any future test the product will be within specifications. However, what is missing from the Q14 and the Q2 guidelines is they don’t define what the specifications on the uncertainty should be in relation to the quality of the product, which means that the fit-for-purpose concept is not explicitly defined.
The importance of this has been underscored by the International Organization for Standardization, which states in its standard ISO 21748: “Without quantitative assessments of uncertainty, it is impossible to decide whether observed differences between results reflect more than experimental variability, whether test items comply with specifications, or whether laws based on limits have been broken. Without information on uncertainty, there is a risk of misinterpretation of results.”2
The objective of an analytical procedure is to be able to provide any reportable value close enough to any future unknown quantity (within a predefined range), with a high probability. Since ICH-Q2 is about the validation of analytical procedures, it’s important to work with a known sample to try to determine the accuracy of a result (a combination of bias and precision). By applying Bayesian theory, the reviewer uses predictive distribution – the distribution of possible future values given the results observed in validation – to evaluate future uncertainty. As long as the uncertainty of future measurement is small in validation, assuming there is a definition of “small” in the TPP of the analytical target performance, the analytical procedure can be accepted. But this is missing in the ICH-Q2.
