Thanks for getting back to me. The setup is as follows:
Subjects receive subcutaneous injections of a drug every 6 months (with flip-flop kinetics). The absorption rate (and clearance, etc.) varies across subjects. However, because absorption is affected by where each injection is given (e.g., depth of injection, vascularization of injection site, etc.), there is within-subject (occasion to occasion) variability in absorption. The random-effects model for absorption is then: Ka_ij = a_i + a_ij , where a_i is an across subjects random effect and a_ij is a (nested within subject) random effect for dosing event j within subject i.
From what I can tell about Phoenix NLME (and I am a relatively new user!), the omega matrix does not readily permit such nested effects. As you mention, I could let ID correspond to unique dosing events, but I would need drug clearance (as an example) not to vary by dosing event but by subject, so that coding would open up another problem.
perhaps you can share more details of the data that you want to fit?
As a first response, if you have within-subject effects I would suggest you try inter-occasion variable where you can label each individual dosing event with a different id. I am not sure why you would need a nested random effect, perhaps you mean a correlation between two random effects. This is typically setup in the omega matrix where you can ask for estimation of off-diagonal elements.
Hope, this helps for a start, we might give more guidance when we have more details.