Ann Yellowlees founder of Quantics Biostatistics explores the importance of accuracy and precision in bioassays and how it can help boost manufacturing efficiency.
An essential part of the pharmaceutical manufacture of many biologics is a bioassay to measure batch potency. The fundamental requirements for the assay are that it should meet the required accuracy and precision criteria.
Looking to lean
Whilst the pharmaceutical industry is, in general, investing in the ideas of lean manufacturing to reduce waste and improve efficiency - "to do more and more with less and less" - little or no attention has been given to the bioassay that is an essential part of batch release.
An assay may be in use for years and is often a hidden factor that significantly increases costs and resource use. So, what is the scope for "leaning" a bioassay in routine use to contribute to efficient manufacturing?
Leaning a bioassay
In bioassay development, the final method is normally locked when adequate precision and accuracy have been achieved, formally documented as part of the statistical validation step. Once validation is complete, the assay may remain in use in the same format for many years.
During development, assays often evolve in a complex way. In the hunt for acceptable accuracy and precision many different designs are tried out, and there are many choices to be made. This can include: the number of doses; dose spacing; replicates within and across plates; statistical models; system and sample suitability criteria, as well as many different biological design choices.
To quote E.B. White, “There’s no limit to how complicated things can get on account of one thing always leading to another” and in bioassay development the result is often an assay that fulfils the primary objective but is far from optimal.
Things to consider
It should go without saying that for efficient manufacturing, the assay should also only pass samples that should pass, and fail those that should fail. This is a problem because bioassays are statistically (and biologically complex) and the pass / fail criteria are usually dependent on probabilities, or "p values", typically set at 0.05.
Consider an assay that has one such test on the reference and one on the sample. These are independent and, on average, ~10% of samples that should pass will be recorded as failing. It is the way the math works; it is called Type 1 error. (It is a bit more complex if the tests are correlated.) How many suitability criteria do you have?
To avoid deviations in the pass / fail criteria, manufacturers should ensure that only strictly necessary system and sample suitability criteria are used. Aim for 99% pass rate of samples in usual ranges (USP 1032).
Be as simple as possible from a laboratory point of view – Reduce plate "real estate"
Unnecessary replicates and plates will lead to more laboratory errors and assay or sample failures. More than one plate will increase maths type 1 failures as the number of suitability criteria applied increases. Therefore, the lab should be ensuring that there are no unnecessary replicates / doses / plates in the design.
Maximise throughput and reduce laboratory requirements
Removing unnecessary replicates and plates will allow more samples per plate and higher throughput.
Action: Minimise the plate real estate required per sample.
Reduce quality assurance (QA) and quality control (QC).
A simple laboratory process is easier for QA/QC, but the resulting increased throughput increases QA/QC burden. However, automated QA/QC can speed production release and reduce QA costs. Look to choose appropriate software that is suitable for automation and consider fully automated continuous real time Validation systems if available.
Other Considerations:
- …Be simple to "tech transfer" - A simple assay is also much easier to transfer.
- … Be straightforward to monitor - Simple assays make it easier to determine whether a sample is failing the assay, or the assay is failing the sample.
How and when?
Once the biology is stable and well characterised, it is time to re-evaluate the design – think about leaning your assay before validation. Stop, step back and review whether all the elements of your design are actually required before continuing.
By this stage, there is usually lots of data available and a biomathematical approach can use techniques such as variance components analysis and simulation to optimise the design with respect to plates, doses, layouts and replicates. Often little or no further laboratory work is required.
Many suitability criteria are mathematically related to each other, and a competent biomathematician can reanalyse development data sets to look at the passes and fails with different criteria and in almost every case reduce the number of proposed tests significantly. NB: USP General Chapter 1032 states:" …it is advisable to accept large fractions of these control ranges (99% or more) and to assess system suitability using only a few uncorrelated standard response parameters."
Leaning in practice - What can be achieved?
Doubling throughput and cutting assay failure rate by 30-50% each would not be unusual. As an example, Public Health England asked Quantics to lean an assay for them. The results were presented at the BEBPA conference in 2018. There was a lot of historical data so no further laboratory work was required. Leaning the assay resulted in a reduction of the number of doses required from eight to four; the throughput was doubled. So, if you are about to validate your assay, now is the time to stop and ask… is this lean?