Jing, I think this is a strategy should be guided by the key questions you are trying to address with your model. If your goal is simply to best approximate the shape of the profile in existing individuals, the strategy you employ could be vastly different than if the next step would be to simulate potential impact of changes to the formulation, in which your placement and justification of covariates could have greater implications on your resulting inferences.
That said, I would suggest considering the mechanism of the formulation release. For example, if these are the similar formulations, with just tweaked adjuvant ratios to control the release rate, you could argue that mtt alone could drive the shape changes. If, on the other hand, these formulations had separate formulations (say different types of microsphere formulations) that could change the mechanism of dispersal *and* the rate of dissolution then you could tie back the contributions in shape to both ntr and mtt. Devin Pastoor Center For Translational Medicine, University of Maryland Baltimore On Sun, Sep 25, 2016 at 9:01 AM Jing Niu <niujin...@gmail.com> wrote: > Dear NMusers, > I am modeling long-acting injectable formulation using transit compartment > model parameterized with mtt (mean transit time) and ntr (number of transit > compartment). As I would like to model multiple formulations using the same > structure model, adding formulation as a categorical covariate on mtt or > ntr separately decreased the OFV significantly (by ~100 units). Considering > mtt and ntr are correlated parameters, should I add formulation on both > mtt and ntr, or adding to either one is sufficient? > Any input is greatly appreciated, > Jing > > PS: I am considering the latter one, due to that the simulations with > changing Ktr or ntr generated similar profiles (Fig1 c and d, > http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3636497/pdf/psp201314a.pdf). > >
