Edward Perry Dan, Sounds like an informative workshop. I would be curious on people's "take" on sampling vs capability vs validation. For example: you make a feature on a device. This feature can be verified via conventional non-destructive methods to the requirement. Is sampling truly (I know should be based on risk) an option if capable (ie. PPK = 1.33 or greater) OR should suppliers be forced into 100% verification. (Where I feel 100% inspection/verification is never 100% effective due to fatigue, boredom, distractions, etc...). I discuss this with customers a lot to provide a risk based approach and to better understand the critical nature of the feature. It also leads into validation efforts....

Dan O'Leary In the course, I make a sharp distinction between process control and acceptance sampling. The best approach is a capable process, Cpk ≥ 1.33, so the process doesn’t make nonconforming product. Acceptance sampling becomes “insurance”, especially at incoming inspection where you may not know much about an external supplier’s process.

In my opinion, suppliers should not be forced into 100% inspection. However, there is a technique called rectifying inspection in which the Average Outgoing Quality is better than the Average Incoming Quality. It depends on screening rejected lots and replacing nonconforming items with conforming. This technique can protect a customer from a process that is not capable, while the process improvement work continues.

martina lee The legal liability.

Dan O'Leary Martina - I don't usually think of legal or financial liability in terms of the underlying statistics of acceptance sampling. However, I think there is a place for both subjects in the course. Do you more specific issues in mind or examples I could look into?

Hi, Dan, sampling plan in design verification and validation together with process control and production release are interesting topics

How about how to document acceptance sampling and testing in the DHR given changes in ISO13485:2016 as well as ROW requirements such as RDC16?

Dan O'Leary Thank you for the suggestion; it is a good one. There is a problem with the US FDA, where some companies received a Warning Letter for recording an attribute (pass/fail) instead of the actual value measured. This is not a requirement in QSR and runs contrary to statements in the preamble. Moreover, it bears directly on the choice of sampling plan (attribute/variable).

Zoltan Deri I think 2 approaches discussed:Production Process(manufacturer side) and Acceptance Process (Customer/consumer side)). About this latest I would say as in RDC16 chapter 9 under 9.2: Using Z1.4, your sample size is selected based on your lot size. Then, you can have the AQL you identified based on the risk you may take for the process average of percent defective. If you decide not to use Z1.4, use the binomial directly, then you can decide on the AQL and batch tolerance proportion defective (BTPD) first, then calculate a sample size for c=0.

Dan O'Leary In your last point, you are determining a sampling plan that goes through two points on the operating characteristic curve and has an acceptance number of zero. In general, this is a hard problem because it requires the simultaneous solution of two nonlinear equations in integers. I cover this in my course, and provide a tool that can help determine these solutions. In my experience, most people don’t want to do this, and prefer published sampling plans. Those that are interested worry that they might not convince a regulator that this is a statistically valid sampling plan.

Hi Dan, as far as sampling goes, attributes seems to have a great deal more information out there for general use. I would like to see a focus on variables sampling and where it is appropriate to use one or the other.

Outside production, I would love additional training on sample size selection for design verification and validation, particularly with regulatory emphasis on properly attested device reliability. I have been struggling to put together a comprehensive plan for variable sample size selection that is applicable to move than one type of device.

Dan O'Leary Attributes sampling is by far the most the common in acceptance sampling. Variable sampling should be used more often, but the primary source, Z1.9, see very complicated. Primarily, it uses calculation methods for the mean and standard deviation that were developed before calculators and spreadsheets.

The sample size question for design validation, design verification, and process validation will come forward because of the requirements in ISO 13485:2016 to justify sample sizes using statistical techniques.

Chander Bhushan Sharma In india, surgical sutures are covered under drugs and cosmetic act. Not as medical devices. Mostly companies follows UD FDA or CE certification. But customer acceptance feedback is concern.