Stability Analysis: How do I back calculate release values necessary to support shelf life?

[#1]

Hello all,

I have recently been working with stability analysis (Degradation) and am running up against a wall with what should be fairly simple (I think).

I have a dataset from multiple lots and the data are poolable (valid for common slope, different intercepts).  I am trying to back calculate what release values would need to be to still be within the lower spec limit at X months.  I have the slope, but I'm not sure how to account for the confidence interval in the model.  I feel like this should be a fairly simple lift but I cannot seem to find how to do this without brute-forcing numbers into the model until I get the right value.

I'm happy to do more legwork...just hoping that someone can point me in the right direction.

Thanks in advance!

Chris

[#2]

There is a data table Reliability/Stability in the Sample Data area.

If you run the degradation script you will see a model with common slopes and separate intercepts.

From the graph you can see the lower confidence bound of batch 2 (green) crosses the spec level at about time 4.5.

Open the Diagnostics and Predictions outline and there is a tab for Inverse Prediction.  Either try and make sense of that graph or from the red triangle choose "Save Crossing Time".

[#3]

Hi Dave,

Thanks for the response.  I think I wasn't explaining my situation correctly.

I have run the degradation script and have the common slope/different intercept parameters.  I also have have the projected expiry based on the ICH limits, etc., from that pooled model.  

I'm wanting to go in the opposite direction.  Say, for example, that I have a lower spec limit of 1 mg/mL...and I want to know - based on the common slope AND accounting for the p=0.25 significance AND the 95%CI of the model - what my starting concentration would need to be for the model confidence interval to not drop below 1 mg/mL at a certain timepoint.  I can get the answer by brute force...but I'm assuming that there's a better approach to doing this.  For example, is it possible to save the model describing the lower 95%CI of the pooled fit and then do an inverse prediction, etc.?

Sorry I didn't go into more detail at first.

Chris

[#4 · Accepted solution]

Hi @ClusterFerret68   The ADG method of example 1 in the attached is how the RL is typically calculated throughout the industry.

http://www.mbswonline.com/upload/presentation_5-25-2012-10-45-1.ppt

[#5]

Thanks @MRB3855 - I'll spend some time on this this evening.

Chris

[#6]

Hi @ClusterFerret68   Using the fit model platform, you’ll have stability time-point and batch as effects. Batch should be set as random. You’ll then have all the inputs you need as shown in example 1.