USP 1033: Top 3 Highlights from New Draft Guidances

The United States Pharmacopeia (USP) guidance on biological assay validation, General Chapter <1033>, will soon be updated. In the leadup to the publication of the new guidance, Quantics co-founder Dr Ann Yellowlees has presented on the changes at the recent BEBPA conferences in both the US and Europe. Here, we summarise some of the most important changes in the new version of USP <1033>, highlighted by Ann, and discuss what they mean for those looking to undergo an assay validation in future.

A crucial part of validating an assay is determining acceptance criteria for its precision and accuracy. Accuracy is defined as relative bias (%RB), while the intermediate precision (%IP) is measured using the geometric coefficient of variation. The 90% confidence intervals on the measurements of %RB and %IP must fall within the acceptance limits in order for the assay to be considered valid.

In the previous version of the guidance, acceptance criteria on the relative bias and intermediate precision were evaluated using the Process Capability Index (C_pm), which is a function of %RB, %IP and the assay format.  C_pm provides an approximation to the probability that the assay result will fall out of specification by chance (Prob(OOS)), so pairs of values of %RB and %IP can be checked for compatibility with a required Prob(OOS).

C_pm typically overestimates the probability of out-of-spec.  The new guidance provides a method for directly calculating Prob(OOS) as a function of %RB, %IP and the assay format. This direct calculation of Prob(OOS) can relax the acceptance limits for accuracy and precision, in turn reducing the sample size required in a validation study.

In the previous guidance, it is suggested that the accuracy and precision of the assay are validated separately. However, the utility of an assay depends on both accuracy and precision: it’s difficult to decouple the two when determining whether an assay is able to provide good results. The new guidance suggests an alternative approach: using Total Analytical Error (TAE). This statistically combines measures of accuracy and precision into a single acceptance criterion, related to Prob(OOS), which gives an overall check of the validity of an assay.

Relative bias considers the accuracy of the assay at a single potency level – for example, testing a reference against itself to check that 100% potency is reported at 100%. It is, however, also important to determine how the bias of the assay changes with the potency which is being measured. That is, is the assay’s bias greater when, say, a sample of 50% potency is being measured compared to a 100% sample?

This property is known as linearity, and it’s typically measured by performing a linear regression between the measured potency and expected potency on the log scale. If the slope of the fitted line is one, then the bias is consistent across the range of potencies tested. Meanwhile, the intercept of the line gives the relative bias when the expected log potency is zero – that is, the expected potency is 1. That means that a “perfect” assay – one with zero bias for all potencies – would show a fitted line with a slope of one and an intercept of zero.

Examining the linearity of an assay is important for determining the range of dilutions over which the assay demonstrates a suitable level of bias and variability. Linearity is, therefore, given increased prominence in the new edition of the USP <1033> guidance compared to the previous version. Linearity is also included in the worked validation example to demonstrate how a linearity study might be performed using some sample data.

The USP guidance documents are just that: guidance.  They do, however, provide a representation of best practice, including how it has evolved over the years between guidance editions.

Want to find out more about assay validation? We cover the important statistical concepts behind validation in our biossay statistics training course.