Sampling Plan for Destructive Testing

[v9991]

background :-
in pharmaceutical drug products, most of the testing involves destructive testing,
some tests have quantitative results;
using single-units to report single values.
others using multiple units to report single value.

further
some tests have qualitative(attribute) results, mostly as pass/fail with only one criteria.
again some tests use single units and others use multiple units

Question :-
A) when a sample size is determined, lets say n=100, (with c=0)

• does it mean that each test characteristic to be analysed on 100 samples. (i.e., if there are 10 tests, then sample size =1000)

OR

• can 100 be distributed among 10 tests with respective way (viz., one test might require 1 sample, another 20, another 10 etc)

request clarifications on interpretation of data.


references :-
reference on US FDA citing for a statistical basis for qualitative tests1

"Furthermore, revise your sampling plans to ensure they are statistically appropriate and non-uniform contamination can be detected. The revisions should encompass finished API testing, in-process testing, and raw material acceptance testing."

another reference...

For example, your firm's finished product sampling plan for (b)(4) and (b)(4) in Propofol Injectable Emulsion is not representative of the batch produced. A total of thirteen units are sampled per lot, with three tested for bacterial endotoxin and ten tested for bioburden. This sampling of thirteen units is irrespective of lot size, which may vary from (b)(4) to (b)(4) units (vials) per lot.


Your response, dated August 10, 2009, is inadequate. You do not reference a sampling plan that utilizes basic elements of statistical analysis or provide a scientific rationale for sampling that would vary the amount of samples taken according to the lot size. Your response does not define a confidence limit to ensure an accurate and representative sampling of the product.

there are similar expectations are outlined for the quantitative tests as well.

 

[Bev D]

The sample size that is statistically valid applies to individual characteristics. So if a unit can only be tested for 1 characteristic then yes you would need 1000 samples to get 100 units for each of the 10 characteristics. OR have a scientific rationale for a different size.

The key here is use of the term non-uniform. The downside of all acceptance sampling is the assumption of a random sample and a mostly homogenous - or uniform - distribution of the defect. Contamination is often NOT uniformly distributed and small samples even if taken randomly are inadequate to catch the contamination. Your sample plan would have to directed at time points when contamination might occur (this is the scientific part of your sampling plan).

 

[v9991]

Thank you Bev D; just want to reconfirm and make sure that i represented my question completely.

So if a unit can only be tested for 1 characteristic then yes you would need 1000 samples to get 100 units for each of the 10 characteristics. OR have a scientific rationale for a different size.

1) if a test requires viz., 10 samples for single determination (destructive testing), then should we take 1000 samples for that test alone?

The key here is use of the term non-uniform. The downside of all acceptance sampling is the assumption of a random sample and a mostly homogenous - or uniform - distribution of the defect. Contamination is often NOT uniformly distributed and small samples even if taken randomly are inadequate to catch the contamination. Your sample plan would have to directed at time points when contamination might occur (this is the scientific part of your sampling plan).

i agree; this is just a thought i would like to share with you, FDA seems to be thinking of "statistical relevance" (ref. above citations quoted in first post)