Sampling Statistics and the Sterility Test

Sampling in Quality Assurance

If you're in the realm of quality assurance and find yourself needing to assess the quality of material entering or products leaving your control, you might have encountered the term "acceptance sampling."

This method involves statistically evaluating the quality of a large batch of items using a smaller sample. Multiple standards exist for this and they follow the same basic mathematical approach

• ISO 2859 (ISO)

• ANSI/ASQC Z1.4, (American Society for Quality)

• BS 6001 (British Standards)

• DIN 40080 (German Standards)

Each of these is expensive to buy - but you can find the same tables to use as a learning tool here and here.

The concept behind acceptance sampling is that by examining or testing a sample from a product batch and extrapolating you can decide whether the entire lot is acceptable or requires rejection.

This approach is necessary in pharmaceutical quality assurance in particular. Imagine you work for an ATMP business receiving a shipment of 2000 single use reactors. Checking and testing each one would be time-consuming and costly and may not be feasible if it compromises their function or sterile integrity. Instead, examining a few units can determine if the whole batch meets standards or if it should be returned to the supplier.

But how many should you inspect? Acceptance sampling assists in determining the number of capacitors to examine and how many defects can be tolerated while still accepting the shipment.

Acceptance sampling by attributes focuses on the count of defects or defective items in a sample. This involves tallying total defects, even if a single item has multiple defects. Alternatively, you can count defective items, considering an item defective if it has any issues.

Attribute plans are generally straightforward: randomly select a sample of n units from a lot of N units. If there are x or fewer defects, accept the lot; otherwise, reject it.

For instance, using the tables shown here if you receive 2000 single use systems, inspect 125 of them (sample inspection level II, letter K). If we have an agreed AQL with the supplier of say 0.1%, If there are 0 defective systems we would accept the shipment. If more than 1 is defective, we would reject the entire lot.

It is extremely important to be aware that sampling only a portion of the lot introduces two risks:

Rejecting a good batch - Know as the producer's risk - as you potentially reject good material

Accepting a bad batch - Known as the consumer’s risk - as you potentially pass material that is below your acceptable quality standard.

The statistical inadequacy of sterility test sampling.

This allows us to understand and place into context just how poor the pharmacopeia sterility test is for quality assurance. You can reverse engineer the AQL tables to illustrate the difference between what a statistically based sampling would have you do and the parameters of the sterility test.

For a candidate sterile solution that is made in 2,000L lots and filled into a 50ml container a typical batch assuming 100% yield will provide 40,000 units - not unreasonable batch quantities.

To generate a sample size large enough to detect a sterility failure rate of 0.025% (a stupendously high limit given the stakes involved) the AQL tables suggest a sample size of 500 units. If one unit fails you would reject the batch.

This means a statistically sound basis to sample would have you destructively test 1.25% of the batch for this one test - and even in this circumstance you would have only assurance that a passing batch has a sterility defect rate below 0.25%. In theory (if you were quite unlucky) the batch could have 100 non sterile units that you would have no idea about based on the testing.

And 500units is 25 times the units specified in the pharmacopeia! It is must less sensitive than the approach described above.

This brings us to a couple of conclusions that must always be kept in mind when thinking about the pharmacopeia sterility test.

• The number of test units specified for the sterility test are partly driven by both the impracticality and the cost of testing - 500 units a batch would not be achievable. 20 is something that can be done readily.

• The container numbers tested are not scientifically set. Even if a prohibitively high number of units are tested the test has no meaningful ability to detect intermittent or sporadic sterility failure - which is a particular concern in aseptic filling.

• Retesting or additional testing when no assignable root cause is not just prohibited by the pharmacopeia but it is not logical or scientifically sound. You cannot test sterility assurance into a product full stop.

These concerns arise even before you consider that the sterility test has been shown to not recover all organisms that could contaminate a product as the growth conditions specified are not optimal for all isolates. There is also a concern as the number of units tested increase that a false positive growth will appear due to a test error.