JMP Stability Test Reports

[#1]

I am new to using the Stability Analysis function (under Reliability and Survival –> Degradation –> Stability Test) in JMP (ver. 14), and I have some general questions about this tool.   Apologies in advance for the rather wide-open “please help” questions. 

Q1: Is this function appropriate to use for on-going monitoring of stability lots, where different lots will be at different timepoints? Or do all timepoints have to be completed before analysis?

Q2: Confidence Intervals. In Minitab Stability Module, the Confidence Interval used would usually be an Upper or Lower One-Sided CI, depending on whether the spec was an Upper or Lower Limit. But in my three examples (attached), the CI always seems to be Lower One-Sided (look in Prediction Settings in the red triangle next to Degradation Data Analysis), no matter the limit or what direction the data is trending. And yet the Overlay Charts seem to be showing 2-Sided (I think – it’s hard to see). It also looks like the Inverse Prediction graphs are showing one-sided confidence intervals, and the Prediction Graph tabs are showing 2-sided. Why aren’t these all the same? [you will have to enter 24 months and select confidence interval from the drop down to see the prediction graph]. How can I tell what is being used for the Stability Test section?

Q3: How do I interpret the Inverse Prediction tab in my Test 2 example? Why are some lines going all the way to zero, but the predicted shelf life is approximately 24 months?

Q4: Also in the Test 2 example, the Prediction Graph shows Lot M confidence interval going beyond the specification at 24 months, but the Stability Tests section indicates the expiration is 24.3282. Why don’t these align?

Q5: Does it make sense to do a shelf life prediction for Test 2? It seems like it is mostly variability and not actual change over time. Is there an objective way to evaluate if a shelf life projection should be done?

My Examples (attached):

Test 1: Upper Spec Limit = 2.9. Increasing Slope. Shelf Life = 24 months.

Test 2: Upper Spec Limit = 14.6. Not much of a Slope in either direction.  Shelf Life = 24 months.

Test 3: : Lower Spec Limit = 78.5, Decreasing Slope. Shelf Life = 24 months.

Thank you!

[#2]

Q1: Looking at stability by lot is convenient, for example adding a local data filter for Lot to your stability report.  A regression control chart would be nice for ongoing stability reporting. JMP doesn't do that our of the box; however there are some discovery talks on the subject.  It might be a good thing to add to the JMP Wish List (tab in the blue bar at the top of the window). 

Q2: The graph in the stability report shows a two one sided 95% confidence interval (alpha 0.1) The crossing time is when the CI crosses the first spec limit for two sided specs. With a one sided spec, the graph is the same, but the crossing time is just for the lower or upper spec limit. The other plots show the same intervals.

Note, when there is no upper spec so the line in the inverse prediction interval goes up to infinity.

Q3: Not sure, maybe post a picture

Q4: Lot M with an upper spec limit of 14-something crosses at 24.3, the lower CI never crosses the upper spec limit. 

Q5: Keep in min the goal of the procedure is to determine whether or not a small number of batches can be pooled or whether the worse case crossing time should be used for determining the shelf life. The one sided 95% CI from the pooled batches with a common slope and common intercept (ignoring the effect of batch or the batch*interaction) results in a narrower CI and therefore a longer crossing time (extended shelf life prediction).

In this case you are looking for batches that are Out Of Trend (OOT). The data presented in the examples appear to remarkably consistent.

I'd be interested to see what some other people might comment.