Dual Acceptance Criteria: Cpk and Individual Values

[ilacatd]

When we test design changes or process changes at our medical device manufacturing company, we typically have a Cpk or %Confidence/%Reliability requirement we have to pass. However, it's possible to pass this requirement but have one or more individual values that fall outside the spec limits. Is this considered acceptable or should we have a second pass/fail criterion where every individual value must also pass? I have seen it both ways. I'm curious how FDA or ISO auditors would view this.

 

[debyang]

Re: Dual acceptance criteria: Cpk and individual values?

Cpk value is accountable only when the product distribution is close to normal distribution.
For quality concern(not for ISO nor FDA concern), having both criteria set is much more sensable.

 

[Bev D]

Dependign on your statistical reviewer at the FDA they might not even accept Cpk as a valid acceptance test because of all of the statistical issues with this index.

on the other hand, having values 'out-of-spec' is not necessarily a reason for rejecting the validation. The FDA (and ISO if your auditor is competant) realize that not all products are perfect. They do understand risk vs cost. (it is wrong to withold a product that will help 99% of the people and only fail to help, or submit them to 'acceptable' harm, even up to and including death).

almost all diagnostic products, drugs and medical devices have some level of failure. the point of validation is to prove that the failure rate is equivalent to your claim. it is the individual points that matter.

 

[TWA - not the airline]

You definitely can have OOS results and not fail your validation.
In my experience for medical devices however I have not seen such a case ever. Taking into account typical lot sizes for PQs any OOS would either fail with %Confidence/%Reliability and at least kill your assumption of normal distribution for calculating cpk (if not fail the cpk criterion itself). Of course if your acceptance criteria are not very demanding then you might find OOS and still be fine, but then that would be a very low risk process/performance requirement and the question would be why you'd need to validate it in the first place...