We are conducting a pharmacological trial with repeated measurements of the drug concentration in the blood over several weeks. Due to the imprecision of the method analysing the drug we have small concentration values even before administering the drug (time 0). The problem is that these values have no (theoretical or practical) correlation to the rest of the measurements. The rest of the measurements have some sort of declining correlation in time (autoregressive or other) so how do we model the covariance structure with no correlation between any pair with baseline measurements and a covariance structure between all other pairs? We are using af linear mixed model and use mostly SAS but any help will be appreciated.

(I have posted the queston in here because i think the question is somewat general and not only for biostatistics. Correct me if i am wrong)