JMP DOE and survivability

[#1]

This may seem like a really obscure question, but I actually think at least a decent number of biologists deal with this: sacrificing organisms during the study. I have used some of the platforms under "reliability and survival" under "Analyze," and they are well set up for situations in which your sample size inherently decreases over the duration of your study because you are either removing organisms for analyses or they are dying due to the treatment. What I am wondering is: can JMP's DOE platform be used for setting up such experiments? From playing around with it today, it seems like you would have to manually enter in "constraints" (e.g., genotype A will NOT be there at time point X because it would have already been removed.) but then again you may not actually know ahead of time when and under which situations an experimental organism may die and therefore cease to serve as a replicate. Are there dedicated DOE platforms for these sorts of studies? Toxicologists much use them, and I imagine others may, as well.

[#2]

Brilliant surgestion by abmayfield!

[#3]

I'm not aware of any specific DOE families that are tailored to the scenario that you describe. One wild thought would be to look at covering arrays? Supported in JMP Pro. These designs are meant to be efficient wrt to determining and 'covering' possible interactions in a system where determining 'failure' in the system is coming from factor interactions. So wouldn't 'death' of an organism be somewhat analogous to a 'failure' in a covering array context? Just a wild thought. Love to hear others thoughts too.

[#4]

I agree with @P_Bartell that a life test might be worth considering. The DOE > Special Purpose > Accelerated Life Test Design is the place to start. It is accelerated because you introduce a factor that changes lifetime. You can start by reading about it here. They are more complicated to design because of the non-linear model used for the analysis.

[#5]

Thanks everyone. I had seen "Accelerated life test" as a DOE option, and the "life" term got me to suspect that this could indeed be the platform I desire. I will read the article Mark suggested. It's strange because, for toxicologists simply looking to kill off organisms in a study, responses such as "time to death" or essentially half-life are easily modeled in the reliability and survival analyses of JMP, it's just that when the study KEEPS GOING that things get tricky (e.g., some organisms survive the treatment and others do not). To be fair, I have never once had a reviewer call me out for this because they are dealing with the same issues, though it would be nice if there was a formalized way to handle these types of situations (as much as you could expect to given what you might know ahead of time about your organisms and their physiologies).