Microbial activity and shelf-life estimation

[#1]

I realize that this is a little bit off-topic, but still relevant with respect to JMP's degradation/stability platform. I was wondering how to estimate shelf-life from microbiological tests (aerobic plate count for example). Literature and ICH only tell me that "Qualitative attributes and microbiological attributes are not amenable to this kind of statistical analysis" (read " not amenable to linear regression & poolability testing"). So far I have been quite content by just commenting on the microbiological activity observed in individual stability studies without performing any statistical tests. Usually I'd just provide an x-y plot where x = time and y = microbe colonies counted. Is this the way this is done in the food, nutritional supplement & pharmaceutical industries, i.e. treating time-dependent data series of microbiological parameters like a qualitative result? What are other users' experiences?

[#2 · Accepted solution]

I cannot speak for those industries as a whole, or what would be accepted in papers, but to your question whether 'treating time-dependent data series of microbiological parameters like a qualitative result' is valid, I believe the answer is yes.  If you can build an empirical model to explain the data you see from your process, and you can be confident that it is valid for the conditions you are trying to estimate, then I think could build a useful model.  You must be careful though.  As you referenced, a linear model which has a great fit over part of your data might not be valid when extrapolated past either end or your existing data, thus you need to be careful about what model you use.