Re: Modeling Cmin and Tmin

On 4/6/2011 11:00 AM, Martin Bergstrand wrote: > Dear Friederike, > > Below are an example of code for obtaining Cmax and Tmax for a simple PK model (I know there are analytical solutions for this, it is only intended as a simple example of a general approach). This can easily be manipulated to instead obtain Cmin and Tmin with another kind of model (contact me again if you run into any problems with that). This solution is based on the addition of two compartments (compartment 3 and 4 in the example) and a COMRES variable (COM(1)). One important thing to consider is that the example code only works if the model predicts a single peak (or in your case a single trough). If the model/design predicts several Cmax (or in your case Cmin) the code needs to be extended to account for that (requires additional compartments and COMRES variables). Temporary Cmax and Tmax (Cmin and Tmin) estimates are outputted in a table file and the individual estimates should be read for the last row of each individual. A good advise can be to add a dummy row (EVID = 2) for each individual at a time well past the last observation for any individual in the dataset and use that for read out. > > $ABBREVIATED COMRES=1 > > $SUBS ADVAN6 TOL=5 > > $DES > > DADT(1) = -KA*A(1) ; Amount in depot compartment > > DADT(2) = KA*A(1)-KE*A(2) ; Amount in central compartment > > X = KA*A(1)-KE*A(2) ; Derivative for the amount in central compartment > > COM(1)=0 > > IF(X.GT.0) COM(1)=1 ; A2 increasing > > IF(X.EQ.0) COM(1)=2 ; A2 maximum > > IF(X.LT.0) COM(1)=3 ; A2 decreasing > > IF(COM(1).LT.3)THEN ; Time to Cmax > > DADT(3) = 1 > > ELSE > > DADT(3) = 0 > > ENDIF > > IF(COM(1).LT.3)THEN ; Maximum A2 > > DADT(4) = X > > ELSE > > DADT(4) = 0 > > ENDIF > > $ERROR > > TMAX = A(3) > > CMAX = A(4)/V ; Amount divided by the central volume of distribution => Cmax > > $TABLE ID TIME TMAX CMAX ONEHEADER NOPRINT FILE=mytab > > ; The individual CMAX and TMAX is found on the last row for each individual in mytab > > Kind regards, > > Martin Bergstrand > > ----------------------------------------------- > > Pharmacometrics Research Group, > > Department of Pharmaceutical Biosciences, > > Uppsala University > > ----------------------------------------------- > > [email protected] <mailto:[email protected]> > > ----------------------------------------------- > > *From:* [email protected] [ mailto: [email protected] ] *On Behalf Of *Friederike Kanefendt > > *Sent:* Wednesday, April 06, 2011 3:05 PM > *To:* Joachim Grevel; [email protected] > *Subject:* AW: [NMusers] Modeling Cmin and Tmin > > Dear Joachim, > > thank you for your explanations. My plan was not to determine the typical value or distribution for this parameters (as THETA or ETA) but to determine the minimum biomarker concentration for each ID with the appropriate time for further correlation studies with outcome etc. I tried to use NONMEM to determine Cmin while individual parameters/profiles are estimated. I read something in NMusers about Cmax and thought that this is maybe also possible for Cmin and Tmin > > http://www.cognigencorp.com/nonmem/current/2008-January/0733.html > > http://www.cognigencorp.com/nonmem/current/2007-December/0716.html > > If this is not possible I would try this using simulations and R > > Thanks > > King regards, Friederike > > *Von:* [email protected] [ mailto: [email protected] ] *Im Auftrag von *Joachim Grevel > > *Gesendet:* Mittwoch, 6. April 2011 11:49 > *An:* Friederike Kanefendt; [email protected] > *Betreff:* RE: [NMusers] Modeling Cmin and Tmin > > Dear Friederike, > > Please, step back for one minute and try to understand what you want to find out: You are applying a nonlinear mixed-effects modelling program to obtain parameter values for your model. All parameters in your model are random variables. Some have a mean of zero (ETAs, EPSs) some should have a mean <> 0 (THETAs). Your results are not point estimates but distributions. Therefore you should approach the question of Cmin and Tmin through simulation and report the confidence interval (or other stats) for your parameters of interest. Thus construct a simulation data set with dense time points around the expected Tmin, and run > > $SIMULATION (240311) ONLYSIMULATION SUBPROBLEMS=1000 > > $TABLE ID TIME ...... NOHEADER NOAPPEND NOPRINT FILE=xxxx.tab > > with all your final parameter estimates as initial values in your control file. Then analyse your 100 (or 1000) simulated data files with R or SAS and report the distribution of the Cmin and Tmin you found. > > Please, come back with more questions in case I have confused you, > > Joachim > > *Joachim Grevel, PhD* > > Scientific Director > > *BAST Inc Limited* > > BioCity Nottingham > > Pennyfoot Street > > Nottingham, NG1 1GF > > Tel: +44 (0)115 8120497 > > *From:* [email protected] [ mailto: [email protected] ] *On Behalf Of *Friederike Kanefendt > > *Sent:* 06 April 2011 09:56 > *To:* [email protected] > *Subject:* [NMusers] Modeling Cmin and Tmin > > Dear NMUser, > > I am a beginner in NONMEM and I have a question regarding modeling Cmin and Tmin. > > I have data of a biomarker which decreases after drug intake and I want to determine the minimum change from baseline (Cmin) as well as the corresponding time (Tmin). I tried different models but the estimated values for these parameters are not plausible or always zero. > > Has someone experience with this or any idea to solve this problem? > > Thanks a lot > > Best regards > > Friederike > > Attached I send you a part of my code: > > $SUBROUTINE ADVAN6 TOL=3 > > ... > > $DES > > ... > > DADT(5) = KIN*INH-KOUT*A(5) ; Biomarker CMT > > IF(TIME.EQ.0) THEN > > CMIN = BASE ; Baseline conc also estimated by NM > > TMIN=0 > > ENDIF > > CC=A(5) > > IF(CC.LT.CMIN) THEN > > CMIN= CC > > TMIN= T > > ENDIF > > REL = CMIN/BASE ; change relative to baseline > > ... > > $ESTIMATION SIG=2 PRINT=1 METHOD=1 INTER MAXEVAL=0 NOABORT > > Friederike Kanefendt > > Pharmacist, PhD-Student > > University of Bonn > > -Clinical Pharmacy- > > An der Immenburg 4 > > D-53121 Bonn > > Phone: +49 (0)228 73-5781 > > Fax: +49(0) 228 73-9757 > > [email protected]