Characterization Testing - NO acceptance criteria, no minimum performance requirement

I have an established product. We are trying to get a new 510k for this medical device (a filter), and in so doing, have been advised by the Agency to perform a certain test. However, I have NO acceptance criteria, no minimum performance requirement for this test. We are just trying to establish the performance level for our product for this given solute. Our medical team has the opinion that the results from this test are low risk. In other words, the primary function of the filter, and the high risk solutes are taken care of by other tests. I'm not looking for lot acceptance. I'm trying to figure out how many to test to establish a mean value of performance that we can publish. Also, we have the ability to make 'verification' samples at the lower and upper end of our manufacturing process. So, I have typical manufactured samples, as well as upper and lower limit samples that I can test. The occurrence of these upper/lower limit samples in actual manufacturing is in the PPM range. I should also note that the test is expensive and time consuming to run.
Thanks in advance.

Ronen E

Problem Solver
Staff member
Super Moderator
You need statistical validity. As a rule of thumb you need at least N=6 for mean validity and N=10 for STD validity and starting to talk about characterising the distribution.
Also look at Bootsrap to minimise actual testing.
Caveat: I'm not a specialist statistician (there are some around here that will hopefully chime in).

Top Bottom