By the way, just for the people that want me to answer to a personal conversation, I cannot do it for some unknown reason(Simon, any idea?).
Now about baseline as a covariate, what do you write as covariate relationship for kin.
Do you estimate aslo baseline by defining it as a structural parameter. I see a sort a circular reference here as of course baseline covariate is highly correlated to baseline as a structural parameter.
Can you send the project?
The relationship kin/kout=baseline applies only to steady state assumption without the drug. Are you under that assumption or do you estimate baseline without that assumption. If you do not make any assumption about baseline, estimate it and then put Kin correlated to baseline, then I do not see any issue. However doing both assumptions is to me problematic. I would need the project.
Thank you for your willingness to help, please see the attached link below with my project.
The final model in the workflow has baseline used as a covariate on kin, while the one just before that does not have this.
Baseline=kin/kout , therefore is it acceptable to set baseline as a covariate on kin from a PK and modeling perspective? Does this make sense to do so, clearly as you can see in my model it is significantly lowering the log likelihood (90points).
We have no idea about the baseline effects , no suspicion, we actually assume baseline has no effect. Please also see the covariate search that I did in my first post above (the continuous covariate plots whereby baseline and kin have a strong relationship), after enabling baseline as a covariate on kin, the etas lose the relationship.
We do assume the baseline is at steady state without drug, but the variability in the population is huge. To give you an idea what we are measuring as baseline, it is the lipoprotein concentration in plasma (mg/dL) in healthy subjects from a Phase 1 study prior to study initiation, it is quite variable.
Is Baseline always identified as significant on Kin by the covariate search, or is this just happening for my particular model? Please see the image in my first post.
I will be happy to include baseline on kin as a covariate, especially since it improves model diagnostics and LL, I just want to be able to justify doing so.
Thanks for your expert help. The link to my model is below (last 2 in the workflow).
there is an “observed” Baseline which was measured with error. You can use the measured baseline as a covaraite on your model based baseline: Kin/Kout or if you parameterize the model by baseline even better but you still have to include a random effect around it since otherwise you assume that you know the baseline without any uncertainty.
Again please read the refered article it studies how to handle the baseline information and what method is best.
One interesting finding, when baseline is not set to a covariate on kin, kin is slightly greater but almost equal to kout (basically the ratio is nearly 1 to 1)
After setting baseline as a covariate on kin, the (ratio of kin to kout becomes 48 fold).
It seems that kin as the rate constant for production of Lp(a) in plasma is significantly increased while all other PD parameters (k1e, IC50, kout) remain nearly identical after baseline is added on kin as a covariate.
We do know that the subjects have huge variability in baseline, i did set the baseline on kin as median. When we arrange the baseline values from our dataset the median baseline is 54, the range is from 5 to 259.4.
Glad I found this discussion! I am doing something similar, and I would like advice on how to implement steady-state individual estimation like the Karlsson paper referenced above.
Previously sequence{Kin/Kout} was ballpark accurate, so I have tried sequence{G_SS} and stparm(G_SS = tvG_SS * exp(nG_SS)), while parameterising Kin/Kout in terms of G_SS.