Feature brainstorming: Automatic steady-state?

[billdenney]

As you can tell from the previous two issues that I just submitted, I'm having a lot of thoughts about steady-state. I'm running a clinical trial simulation where the system comes to steady-state in about 2 weeks. After that, it's all just time within the dosing interval and residual variability differences for how each measurement behaves.

It would be handy if I could tell mrgsolve to try to find steady-state automatically in a simple way.

What I have done for the current simulation is a lot of mucking about and playing with the time variable so that it takes advantage of the speed of steady-state calculation while taking advantage of the simplicity of different data management.

The specific algorithm that I'm thinking of is something like the following (with an indicator of SS=-1 in the dataset to take advantage of these automatic features):

• If SS == 0 or SS == 1, perform with current specifications.
• If SS == -1
  • From the beginning to steady-state time simulate as normal (i.e. allow the system to come to steady-state and show the values as the system is coming to steady-state as though SS == 0)
  • From when steady-state is achieved (using the current detection of steady-state rules for SS == 1) through the end of the dataset for the individual or until the first record where SS != -1, assume that the only difference for all the remaining records is the time after dose. Simulate a single dosing interval with all time after dose values in that interval, and resample from the residual error distribution for each additional value.
  • If there are additional values after the last dosing interval at steady-state, simulate from the end of the dosing interval to the last sample as though SS == 0.

[billdenney]

The first issue I see with this is something like auto-regressive residual error, but that would be a problem for normal SS == 1 records, anyway, so it shouldn't be a specific issue here.