Bayes Success run Theorem for sample size during OQ&PQ

[Florent73]

Dear all,

Currently working for a medical devices manufacturing company, I was wondering if Bayes success run Theorem is considered by FDA or other EU notified body as a valid statistical methods to determine the sample size during process qualification stage (i.e. OQ and PQ). The below screenshot details the formula used. In other words, no matter the lot sample size, if one wants to prove with a 95% confidence level that the process is for instance 90% reliable a minimum of 29 sample with no defective (i.e c=number of failure =0). From my standpoint the interesting part with this approach is that
- it is pretty straightforward
- it does not depend on the lot size
- it is based on a consumer (=RQL=LTPD approach) risk approach.
- as it is a c=0 sampling plan the sampling size is much smaller than AQL ISO2859-1 approach (which by the way is a producer risk approach)



Many thanks in advance for your comments,
Best Regards,

Florent

 

[chris1price]

Hi, I have not seen it called Bayes Success run Theorem, but it looks like the Binomial calculations. I have used this method and FDA have not had an issue. However you do need to define it somewhere and explain how the confidence/reliability values correspond to your risk profile.

 

[Bev D]

This is a straightforward binomial calculation with random sampling for the lot. One of the benefits you list are spot on: it is always better to use RQL than AQL for these types of inspections.

It may be quite acceptable for PQ (although our rules are 3 successive passing lots based on sampling inspection)

I would not advise using it for OQ. OQ is intended to validate the maximum input settings and to look for margin erosion - aka capability.

 

[Florent73]

Hi Chris,

Thanks for the answer. It is indeed based on Binomial distribution. Hereinafter a link where you can find the term "success run Theorem". Fully agree with the fact that the sampling plan must be based on reliability/confidence level fixed in kind of risk management SOP matching each risk level to a reliability/confidence level.


Thanks again for the prompt reply.

 

[Aaron M]

Hi Bev,
Are you saying that for 95/90 (with no defects) that the sample size would be 29 units for 3 different builds/lots (87 units total) or a total of 29 units over 3 different builds?

It may be quite acceptable for PQ (although our rules are 3 successive passing lots based on sampling inspection)

Thanks,
Aaron