Variance components analysis for lyophilization process

Question

Hello all,

I would like to perform a variance components analysis to characterize the robustness of a lyophilization process.

Here is the background:

* I have performed three lyophilization batches with three different cycle parameters.
* I have intentionally placed product vials in defined locations across the shelf for each batch.
* At the end of each batch, I have taken 10 samples from each of those defined locations and analyzed them. Each sample of the 10 samples is a replicate sample.
* I want to understand the potential variation of analytical results due to the lyophilization batch and shelf location

Considering this information, is it appropriate to carry out a variance components analysis or are there other statistical methods that would be more appropriate e.g. one-way ANOVA? How can this be done in JMP?

Thanks in advance.

P_Bartell · Answer

To add a bit to @statman 's advise you may want to take a look at this On Demand webinar hosted by my former (I'm now retired) JMP colleague @JerryFish . Measurement Systems Analysis Jerry shows the Wheeler EMP method which is essentially the control chart method referred to by @statman

statman · Answer

I am not an SME for your situation, but I have some thoughts:

1. The situation you describe appears to be a nested/systematic component of variation study.  The hierarchical layers are:

Batch (although this appears to be manipulated)

Shelf (Systematic locations, though you do not specify how many)

Samples (Confounded with the Sample component is the measurement system variation)

2. How confident are you in the measurement system? Why do you need 10 samples? You are somewhat biasing this layer.

3. Robustness is the measure of how consistent the lyophilization process is over changing noise.  I would be suspicious that in 3 batches you have exposed all of the possible noise effects.  You will not be able to assess the consistency of the batch-to-batch variation.

4. For analysis you may be able to use the control chart method or ANOVA method.  I would start with variability plots (make sure the data types are nominal) and let JMP calculate the quantitative components.