Best design for device delivery component testing?

Question

Hi,
I've been asked to support a device delivery component design project. The goal is to demonstrate that multiple types and multiple lots of three components—syringes, tubing, and stopcocks—consistently yield endotoxin results within our specification. Since testing is expensive, we're trying to minimize the number of runs while still generating data that would be acceptable to the agency.
I initially set this up using a custom DOE with categorical variables. For each of the three components, I included 9 levels to represent the 3 types and 3 lots per type. This resulted in a minimum of 25 runs, which seems manageable from a testing perspective.
That said, I’m not sure this is the most effective approach. Modeling each component with 9 levels loses visibility into the effect of type versus lot, since they’re all lumped together. I’m also unsure how best to handle the response data, as we expect a mix of <LOD values and numerical results.
This seems like a fairly common type of study, but I haven’t had much luck finding similar examples in the JMP literature—possibly because I’m not using the right search terms.
Any guidance or suggestions would be greatly appreciated!

Hartebeest8 · Answer

Thanks for the response.  I now realize I should clarify that the goal is to test the final assembled configuration of the device, since that’s what the agency will evaluate.

Regarding your question about agency expectations, Section III of the FDA’s Endotoxins Testing for Medical Devices: Questions and Answers guidance outlines some key principles. It emphasizes the importance of demonstrating that devices consistently meet endotoxin limits, and specifically calls for testing across multiple lots and component types to ensure reproducibility. It also notes that testing should be performed on the final configuration of the device as it will be used clinically, which is a critical consideration for our study design.

I'm not 100% sure what you mean by using a hypothesis, we do expect the different components to vary some which is why we need to do this study but don't have any reason to believe one component or lot would be more than another. We essentially want to prove that any combination would see have us below spec. We do plan to test 3 different lots of the 3 different types of the 3 different components but of course are not able to test every iteration of that so looking for the most appropriate way to come up with those combinations/study.

statman · Answer

I don't completely understand your particular situation, nor do I know "agency" requirements, but DOE would not be the approach I would use.  Instead I would use directed sampling.  I would do components of variation studies and use hypotheses to suggest a rational sampling plan so the samples you get are REPRESENTATIVE of future conditions.  Likely a nested study that will allow for assessing the consistency of the incoming product.  To make the study more efficient, use hypotheses you develop to explain why the product might vary and then ensure your study includes instances where those vary.  For example, let's say you hypothesize the tubing will vary with different lots of the material used by the supplier due to the changing chemical makeup of the rubber.  Then you will need to ensure that your samples of tubing are over multiple lots of material.