Product Shelf-Life Extension Based on Statistical Analysis of the Collected Stability Data

[#1]

I am unable to find any reasonable resources on how to use the 'Degradation' feature of JMP.  In older versions (and 13), I've simply been using Fit X by Y and performing linear regression with 95% confidence intervals shown.   

I'm trying to learn the new degrdation features, but there seem to be no good resources that step you through how to fill out the fields and they are vaguely named.  I keep getting a JMP error of "System ID role and Covariates role cannot both be empty", but I have no idea how to solve this.   There is one field in the Degradation Data Analysis Stability Test tab that is titled "Label, System ID" so I am guessing that is the "system ID" in the error message.   I have no idea wha the covariates role is.   Is there a video that step users through an example 'click by click'.    The resource I found from Dr Meeker were great in theoretical information, but short on the 'step by step' use from a practical stand point.   (although his webcast was very informative on the theory--thank you Dr. Meeker).

[#2]

See Help > Books > Reliability and Survival > Chapter 7: Degradation. In particular, see pages 189-191. (References specific to JMP 13 but similar material exists for previous versions.)

[#3]

You are supposed to be measuring the response along the degradation path so JMP needs to know which measurements are for the same measurement unit over time. That is the purpose of the Label, System ID analysis role. This might represent a batch, vial, or actual device. The Covariate role is used if you varied some factor, like temperature,during  in the collection of the data. For example, an accelerated life test might use four elevated temperatures (above recommended) to accelerate the degradation. This covariate can be included in the model so that you can extrapolate to recommended temperature and time.

[#4]

Mark,  thanks for your replys.   I was able to by pass the Label, system ID by using the Destructive Degradation Tab.   While I am getting representative plots, I am unable to filter the data by the X category.   In otherwords, if I have vial types "upright" and "inverted", I am unable to look just at upright and just at inverted by filtering.     I know it may seem rather basic and intuitive in the menus, but I'm so used to Fity X by Y and how to navigate those menus that I simply can't get the graph to be visualized the same ways in the Destructive Degradation plots.    If possible, it would be great if one of the staff could do a YouTube video on this topic.   Just a thought.....

Again, thanks for helping me try to solve this problem.   I think I will show the folks I'm training (novice users) both ways.  

[#5]

So your measurement destroys the sample and it can't be measured again? You have not said what you are testing but vial suggests something biological or chemical. That said, you might produce many vials from a batch of material. If so, we don't consider the testing of individual vials as destructive. You are measuring the degradation path of the 'batch,' not the 'vial.'

You might want to use the vial type in the By role to produce separate analyses. The covariate analysis role is intended for something that has an effect and you want to use in predictions (interpolation, extrapolation). The by analysis role is a convenient way to perform the same analysis on more than one group of data. Since you are filtering on vial type, you are essentially taking it out of the model and performing a separate analysis.

[#6]

Mark, I really appreciate your help.   Unfortunately, I am a visual learner and I am having difficulting interpreting the response.  I am performing Stability Shelf Life Analysis on a Monoclonal Antibody that has some vials "inverted" and some vials "upright".   I wish to compare their slopes.   I will try to reach out to one of my old mentors who originally taught me the X by Y (Dr. Tom Little) back in the early 2000's,  before this function was available.    I think part of the problem is that the explanation is more 'statistical' instead of step by step practical.   Sorry, but I'm in the "remedial" class.  

[#7]

Tell me the names of the data columns and I will prescribe the analysis for you using Degradation and using Fit Least Squares. That way you can try them for yourself and decide which one suits you better. (It would be great if you could share the data with me, but I understand completely if that is not allowable.)

I strongly urge you to learn the more complex analysis method because of the complex nature of the data and the questions. An overly simple analysis could be misleading. I will help you learn!

[#8]

How do I send you the data file ?

[#9]

The Bivariate platform (from Fit Y by X) is not be the best analysis for determining expiry. It models a single degradation path so it is (1) less powerful (divide the data into small sets) and (2) can't test for inter-batch or device effects. For example, with just the minimum of three production lots you want to determine if the degradation path is the same, on average, or different between batches. Bivariate can't help with that determination. Use Analyze > Fit Model and include the time variable, the batch/system variable, and their cross term (interaction) as effects. This dialog will launch the Fit Least Squares platform. It provides the necessary tests to decide if you can use a common slope and intercept as well as determine expiry using inverse prediction.

The Stability Test in the Degradation platform automates much of this process and provides additional information.

[#10]

Mark,
Thank you for your patience and kindness. Your help made it possible for me to easily understand the new (?) degradation feature of JMP. I practiced and matter the there feature and was able to teach them to colleagues. Thanks for helping me and our company bee successful with our stability studies.