I currently work for a medical device manufacturer that makes class I, II and III product. I have been asked to manage the plant's process validation program for manufacturing equipment; however, I am faced with the problem of choosing sample size for process validation runs. In the past we have used AQL sampling plans. We assigned an AQL level to each of the classes of product and then used the projected amount of product that would be produced in 3 shifts to give us a sample size for a validation run. This worked quite well for us but frequently required large numbers of samples and testing for those samples. Lately we have been trying to use the sample size formula n=Z^2*stdev^2/E^2 in order to reduce the number of samples taken and the testing for those samples; however, at no time do we go below 30 samples for a run even if the formula tells us we can do less. The problem is that this sample size formula is primarily for hypothesis testing where you are trying to detect a certain level of change (E) as a result of a treatment to a control group. The result is that there is confusion here as to what to select for E. I have seen a medical device white paper recommend using a percentage of the specification range for E but I was wondering if this is valid. I have had one Engineer tell me to forget the formulas and just state a constant sample size for each of the product classes and be done with it. The process validation runs typically include high and low optimization runs and 3 process qualification runs - so you can see that a large sample size for a run requires a substantial number of samples, testing, and time.

My question - What do other medical device facilities use to determine sample size for a process validation run for a piece of equipment?

Thanks in advance,
Andrew

Andrew !
I'm not in this field, maybe this site will help you.
http://www.ghtf.org/sg3/sg3-final.html
György

Andrew !
I'm not in this field, maybe this site will help you.
http://www.ghtf.org/sg3/sg3-final.html
GyörgyYes, especially Annex A.

I currently work for a medical device manufacturer that makes class I, II and III product. I have been asked to manage the plant's process validation program for manufacturing equipment; however, I am faced with the problem of choosing sample size for process validation runs. In the past we have used AQL sampling plans. We assigned an AQL level to each of the classes of product and then used the projected amount of product that would be produced in 3 shifts to give us a sample size for a validation run. This worked quite well for us but frequently required large numbers of samples and testing for those samples. Lately we have been trying to use the sample size formula n=Z^2*stdev^2/E^2 in order to reduce the number of samples taken and the testing for those samples; however, at no time do we go below 30 samples for a run even if the formula tells us we can do less. The problem is that this sample size formula is primarily for hypothesis testing where you are trying to detect a certain level of change (E) as a result of a treatment to a control group. The result is that there is confusion here as to what to select for E. I have seen a medical device white paper recommend using a percentage of the specification range for E but I was wondering if this is valid. I have had one Engineer tell me to forget the formulas and just state a constant sample size for each of the product classes and be done with it. The process validation runs typically include high and low optimization runs and 3 process qualification runs - so you can see that a large sample size for a run requires a substantial number of samples, testing, and time.

My question - What do other medical device facilities use to determine sample size for a process validation run for a piece of equipment?

Thanks in advance,
Andrew

Andrew, another way to look at E is to determine how big of a shift would give unacceptable product. Set E such that you have a 99% acceptance rate.

First of all - thanks to everyone for their replys. It would appear that the Global Harmonization Task Force is suggesting the use of acceptance sampling tables to determine sample sizes; although they give no guidance for acceptable AQL levels - any help here would be appreciated. I wouldn't mind exploring the setting of E to give a 99% acceptance rate but I am not sure how to go about it.

Andrew

First of all - thanks to everyone for their replys. It would appear that the Global Harmonization Task Force is suggesting the use of acceptance sampling tables to determine sample sizes; although they give no guidance for acceptable AQL levels - any help here would be appreciated. I wouldn't mind exploring the setting of E to give a 99% acceptance rate but I am not sure how to go about it.

Andrew
Just to make something clear, and I have not reviewed the GHTF documentation, you should not and the FDA frowns on using AQL based sample size for continuous data. That is why I recommended using tolerance intervals.

First of all - thanks to everyone for their replys. It would appear that the Global Harmonization Task Force is suggesting the use of acceptance sampling tables to determine sample sizes; although they give no guidance for acceptable AQL levels - any help here would be appreciated. I wouldn't mind exploring the setting of E to give a 99% acceptance rate but I am not sure how to go about it.

AndrewThey do suggest the use of acceptance sampling tables to determine sample sizes for performing the capability study, not as a means of validating the process. You can't run the process, take a sample and say that based on the sample, the process is validated.