Re: [NMusers] Is it possible that IIV (%CV) of final model was higher than IIV of base model?
Dear Luann, "How did you determine that ALT, HGB, and TB should be in the model for TVCL?" I have determined these covariates as significant covariates by stepwise forward selection and backward elimination. Sincerely, I really concerned that I answered you in the right way? kind regards, Pete On Wed, Jul 10, 2019 at 1:23 AM Luann Phillips wrote: > Hi Vichapat, > > I was just asking which method that you had used. Some people do use the > baseline value for the whole dataset even for long studies . I disagree > with this method. I think time-varying covariate values should always be > used for long study periods. It sounds like the way you built your data is > fine. > The ctl stream looks the same whether you use stationary or time-varying > covariates. When NM is fitting the model, it uses which ever covariate > value is available. > > As the differential equation solver in NONMEM steps from TIME=N to > TIME=N+1, the covariate value from TIME=N+1 is used (see example below) > > Example: > > TIME=0 AMT=40 DV=. COVAR=55 > TIME=11.833 AMT=. DV=25 COVAR=55 > TIME=12 AMT=40 DV=. COVAR=55 > TIME=23.833 AMT=. DV=30 COVAR=55 > TIME=24 AMT=40 DV=. COVAR=60 > TIME=35.833 AMT=. DV=25 COVAR=60 > TIME=36 AMT=40 DV=. COVAR=60 > TIME=47.833 AMT=. DV=27 COVAR=60 > etc. > > As NONMEM steps from TIME=0 to TIME=11.833 the value of COVAR=55 from the > TIME=11.833 record is used > As NONMEM steps from TIME=11.833 to TIME=12 the value of COVAR=55 from the > TIME=12 record is used > As NONMEM steps from TIME=12 to TIME=23.833 the value of COVAR=55 from the > TIME=23.833 record is used > As NONMEM steps from TIME=23.833 to TIME=24 the value of COVAR=60 from the > TIME=24 record is used > As NONMEM steps from TIME=24 to TIME=35.833 the value of COVAR=60 from the > TIME=35.833 record is used > As NONMEM steps from TIME=35.833 to TIME=36 the value of COVAR=60 from the > TIME=36 record is used > As NONMEM steps from TIME=36 to TIME=47.833 the value of COVAR=60 from the > TIME=47.833 record is used > etc. > > If the value of COVAR=55 for all records NONMEM still works the same way, > it's just that the value of COVAR will never change. > > So in your case, > TVCL=THETA(1)*EXP(THETA(4)*(ALT/388))*((HGB/10.50)**THETA(5))*((TB/4.7)**THETA(6)) > changes value every time that ALT, HGB, or TB changes value. > The ETA(1) value remains the same for all observations within an > individual but CL will still change with time because ALT, HGB, and TB > change with time. > > How did you determine that ALT, HGB, and TB should be in the model for > TVCL? > > Luann > > > -- > *From: *"Vichapat Tharanon" > *To: *"Luann Phillips" > *Cc: *"nmusers" > *Sent: *Tuesday, July 9, 2019 1:53:18 PM > *Subject: *Re: [NMusers] Is it possible that IIV (%CV) of final model was > higher than IIV of base model? > > Dear Luann, > > (1) My data file was recorded with covariates values changed > each times in according to the lab monitored. Hence, I think I have > time-vary covariates in datafile. Now, I use normal control stream to model > these data. So, you suggested me to put a new value on a record with > matching date and then retain forward to the next covariate sample. From > this suggestion, let me confirm that I should have one column for baseline > covariate (1st Lab monitoring) and another column for exact covariates > recorded on that day? > > (2) Then, how could I code the control stream for the covariate > model with time-varying covarites? (sorry that I have never get into it) > > (3) Btw, I have one doubtful question about stationary > covariates on the data file. Is it possible to model the PPKs of the drugs > with stationary covariates.I mean that is it rationale to use > only one value of each covariates in the model wheres the > concentration+dose were dynamic especially if the study period take quite > long time. > > Thank you so much for your reply, valued comments and > suggestions. > > Kind regards, > Vichapat > > > On Tue, Jul 9, 2019 at 8:58 PM Luann Phillips > wrote: > >> Vichapat, >> >> Your ctl stream appears to be correct. To model with time-varying >> covariates involves a change in the database. >> (A) Did you use the covariate values at the time of each patient's first >> dose (ie, baseline values) in the data? >> or >> (B) Did you use the covariate values each time that they were collected? >> >> (A) is stationary covariates and (B) is time-vary covariates. >> >> To include time-varying covariates in the data, put the new value on a >> record with a matching date and then retain forward to the next covariate >> sample. >> >> Please be aware that the dosing and sample time assumptions (which >> sometimes are required) will also add to unexplained variability. I would >> look at plot of the data prior to running any models and exclude any >> concentrations that look very wrong (ie, collected at a peak instead of a
Re: [NMusers] Is it possible that IIV (%CV) of final model was higher than IIV of base model?
Dear Luann, (1) My data file was recorded with covariates values changed each times in according to the lab monitored. Hence, I think I have time-vary covariates in datafile. Now, I use normal control stream to model these data. So, you suggested me to put a new value on a record with matching date and then retain forward to the next covariate sample. From this suggestion, let me confirm that I should have one column for baseline covariate (1st Lab monitoring) and another column for exact covariates recorded on that day? (2) Then, how could I code the control stream for the covariate model with time-varying covarites? (sorry that I have never get into it) (3) Btw, I have one doubtful question about stationary covariates on the data file. Is it possible to model the PPKs of the drugs with stationary covariates.I mean that is it rationale to use only one value of each covariates in the model wheres the concentration+dose were dynamic especially if the study period take quite long time. Thank you so much for your reply, valued comments and suggestions. Kind regards, Vichapat On Tue, Jul 9, 2019 at 8:58 PM Luann Phillips wrote: > Vichapat, > > Your ctl stream appears to be correct. To model with time-varying > covariates involves a change in the database. > (A) Did you use the covariate values at the time of each patient's first > dose (ie, baseline values) in the data? > or > (B) Did you use the covariate values each time that they were collected? > > (A) is stationary covariates and (B) is time-vary covariates. > > To include time-varying covariates in the data, put the new value on a > record with a matching date and then retain forward to the next covariate > sample. > > Please be aware that the dosing and sample time assumptions (which > sometimes are required) will also add to unexplained variability. I would > look at plot of the data prior to running any models and exclude any > concentrations that look very wrong (ie, collected at a peak instead of a > trough). Perform the modeling and then try re-including the 'wrong' > concentrations to show the impact to the model but I would still make the > final model the one excluding those concentrations. > > Luann > > -- > *From: *"Vichapat Tharanon" > *To: *"Luann Phillips" > *Sent: *Monday, July 8, 2019 10:06:06 PM > *Subject: *Re: [NMusers] Is it possible that IIV (%CV) of final model was > higher than IIV of base model? > > Dear Luann, > >Thank you so much for your valued suggestions. I greatly > appreciated it. By the way, The suggestion given me that mean I should use > "Time varying covariates" on the model? I am really new with NONMEM, If > you do not mind helping me. Could you suggest me how to code the control > file for that model in right way. I really know that my request may disturb > you, but I do not know how to start it. Thank you in advance. > > Best regards, > > PS, This is my original control file for final model. There are 1170 > Tacrolimus concentration from 50 patients (retrospective data) then I > assumed all patient took a drug at same time (every 12 hours: AM, PM on > time) and Trough concentrations were monitored at 11.50 hours (Before the > next morning dose 30 minutes). > Briefly, tacrolimus was reported high inter- & intra-variability and > primarily metabolized by liver via Cytochrome enzyme and eliminated via > bile. > > ;Model Desc: Final model > ;Project Name: step3cov > ;Project ID: NO PROJECT DESCRIPTION > ;Project ID: NO PROJECT DESCRIPTION > > $PROB RUN# ALTHGBTB > $INPUT C ID TIME AMT ADDL II TAD DV MDV EVID BW POD AST ALT ALP GGT TB DB > ALB HGB HCT BUN SCR > $DATA MASTER.CSV IGNORE=C > $SUBROUTINES ADVAN2 TRANS2 > $PK > > > TVCL=THETA(1)*EXP(THETA(4)*(ALT/388))*((HGB/10.50)**THETA(5))*((TB/4.7)**THETA(6)) >CL=TVCL*EXP(ETA(1)) >TVV=THETA(2) >V=TVV*EXP(ETA(2)) >TVKA=THETA(3) >KA=TVKA*EXP(ETA(3)) >S2=V/1000 > > $ERROR > IPRE=F > W= 1 > IRES= DV-IPRE > IWRE=(DV-IPRE)/W >Y = F + ERR(1) > > $EST METHOD=1 INTERACTION PRINT=5 MAX= SIG=3 MSFO=ALTHGBTB.msf > $THETA > (0,20) ;[CL/F] > (0,500) ;[V/F] > (fixed,4.48) ;[KA] > (0.001);[ALT] > (0.001);[HGB] > (0.001);[TB] > > $OMEGA > 0.04 ;[P] omega(1,1) > 0.04 ;[P] omega(2,2) > (fixed,0) ;[A] omega(3,3) > $SIGMA > 0.04 ;[A] sigma(1,1) > > $COV > $TABLE ID CL V KA ETA1 ETA2 ETA3 PRED RES WRES IPRE IWRE CPRED CWRES TIME > AMT ADDL II TAD DV BW POD AST ALT ALP GGT TB DB ALB HGB HCT BUN SCR TIME > ONEHEADER NOPRINT FILE=ALTHGBTB.tab > $TABLE ID TIME CL V KA ETA1 ETA2 ETA3 ONEHEADER NOPRINT FILE=PATABALTHGBTB > $TABLE ID BW POD AST ALT ALP GGT TB DB ALB HGB HCT BUN SCR ONEHEADER > NOPRINT FILE=COTABALTHGBTB > $TABLE ID ONEHEADER NOPRINT FILE=CATABALTHGBTB > $TABLE ID TIME PRED RES WRES IPRE IWRE CPRED CWRES ONEHEADER NOPRINT > FILE=SDTABALTHGBTB > $TABLE ID CL V KA NOAPPEND NOPRINT FILE=ALTHGBTB.par > $TABLE ID ETA1 ETA2 ETA3
Re: [NMusers] Is it possible that IIV (%CV) of final model was higher than IIV of base model?
Dear Pete, The most natural explanation for this finding is if the covariates in question are time varying and vary rather substantially within an individual over time. Given the listed covariates this seems likely (please confirm if that is so). This would explain why you primarily improve residual unexplained variability (RUV) and not IIV with the inclusion of the covariate effects). The base model (without inter-occasion-variability (IOV)) can’t explain time varying changes in the parameters (CL/F and V/F) and hence true changes in parameter estimates over time results in an inflated RUV estimate. If the time varying covariates shows significant trends i.e. general increase/decrease over time, the inclusion of these covariates effects can correct a model misspecification in the base model. Under these circumstances it likely that IIV is underestimated with the misspecified base model and hence the apparent increase in IIV with the covariates included. Depending on how sparse your observed plasma concentration measurement are it may be a good idea to try to first characterize IOV for one or more parameters to better understand how much of the parameter variability that is explained by the covariates. Kind regards, Martin Bergstrand, Ph.D. Principal Consultant Pharmetheus AB +46(0)709 994 396 <+46709%C2%A0994%20396> martin.bergstr...@pharmetheus.com www.pharmetheus.com +46(0)18 513 328 <+4618%20513%20328> U-A Science Park, Dag Hammarskjölds v. 36b 752 37 Uppsala, Sweden *This communication is confidential and is only intended for the use of the individual or entity to which it is directed. It may contain information that is privileged and exempt from disclosure under applicable law. If you are not the intended recipient please notify us immediately. Please do not copy it or disclose its contents to any other person.* On 8 Jul 2019, at 16:21, Vichapat Tharanon wrote: Dear All, I am a hospital pharmacist and I am working on NONMEM as a new user. I have modeled the oral immediate-released tacrolimus (Prograf) in adult liver transplant patients. Most of the data were trough concentration (about 1170 levels) from routine monitoring tacrolimus data in the period of first day post-transplantation to 6 months. The model was constructed by NONMEM 7.2 using FOCE INTERACTION methods with the subroutines ADVAN2 TRANS2 (one compartment model with linear absorption and elimination). The ka could not be estimated and then was fixed at 4.48 h-1. The IIIV and RUV were described by exponential and additive error model, respectively. Forward addition of a liver enzyme (ALT ), Hemoglobin and total bilirubin (TB) on CL/F reduced OFV significantly (delta OFV ~98, 42, 28, respectively) but IIV of CL/F was increased from 37.2% to 38.1%. It was found that no significant covariates influenced to V/F but IIV of V/F was also increased from 55% to 63%. Residual variability was reduced from a SD of 2.80 to 2.65, when compared final model and base model. I feel uncomfortable with these findings. Is it possible that IIV of CL/F and V/F were rising after adding the significant covariates whereas %RSE of the CL/F and V/F estimate as well as IIV of CL/F and IIV of V/F in final model were slightly decreasing. May I have your comment or suggestion; I would really appreciate it. Thank you in advance. Best regards, Pete
