Re: Problems with PD Modelling
You have to use diagnostic plots to study the data. One can plot concentration-time and effect-time curves with the same time scale and see what is going on. If they are changing simultaneously, one can use direct concentration-effect Emax or Hill model. If you see time shift (effect curve is shifted to the right relative to the concentration curve) then you use indirect response. Inhibition of production or stimulation of elimination models are similar (the other pair is inhibition of elimination and stimulation of production). Selection of the model out of these two can be made either on the goodness of fit basis or using biology of the effect.
Histersis plots are sometimes helpful to see that you need indirect-response rather than direct model: you plot effect versus concentration for each subject and connects the points as they are observed in time. If you see loops rather than curves in these plots this is an indication of the time delay that need to be accounted for either by effect compartment model or indirect response model.
Hope this helps
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 2/13/2011 12:01 AM, xin yi wrote:
> Hi,
>
> Thank you for your reply. You mentioned that you cannot see from the
> data whether they correspond to the indirect response model with
> inhibition. Can you advise me how I can infer from the data what kind of
> model this PD data should correspond to?
>
> And if it's not an indirect response model, what other models do u
> suggest I use?
>
> On Sat, Feb 12, 2011 at 11:56 PM, Leonid Gibiansky
> <[email protected] <mailto:[email protected]>> wrote:
>
> The code looks OK, but I cannot see from the data whether they
> indeed correspond to the indirect response with inhibition. Looks
> like random oscillations to me. This could be a reason for error
> messages.
>
> I would try to fix ETA_Kin or ETA_Kout to zero and use only additive
> (or proportional) error. Also, it will not help convergence but it
> is more mechanistic to use
>
> CONC = A(2)/S2
> INH =CONC/(IC50+CONC)
>
> Then IC50 will be in concentrations rather than in amounts.
>
> Also, TOL=3 is to small. Try to use TOL=6 at least (better 7 or 8).
> Same for PK: TOL=3 is not good for the final model.
>
> Regards
> Leonid
>
> --------------------------------------
> Leonid Gibiansky, Ph.D.
> President, QuantPharm LLC
> web: www.quantpharm.com http://www.quantpharm.com
> e-mail: LGibiansky at quantpharm.com http://quantpharm.com
> tel: (301) 767 5566
>
> On 2/11/2011 10:35 AM, xin yi wrote:
>
> Hi all,
>
> I'm new to nonmem and would be grateful for some pointers and
> help. I'm
> trying to do a pk-pd sequential modelling. I have modelled the
> PK data
> but I've encountered problems with PD. My PD model is a indirect
> inhibition response model. I have tried to change my initial
> estimates
> reduce the number of parameters in the model, but nothing seem
> to get me
> a good convergence. At different times, NONMEM gives me
> different error
> messages such as "parameter estimates near its boundary",
> "Minimisation
> successful. However problems occured with the minimization"(blank
> entries for omega or sigma estimates in the results file) ///or
> the "r
> matrix is algorithmically singular and non-positive definite". I
> have a
> few questions:
>
> 1) Are there any errors in the way I input my control and data file?
> 2) Under $ERROR in the control file, did I define it correctly with
> EFF=A(4), Y//=EFF+EFF*ERR(1)+ERR(2)/? Or should it be IPRED=F,
> Y=F+F*ERR(1)+ERR(2).
> 3) Why are they no estimates for sigma and omega in the results
> file? I
> have been constantly changing my initial estimates for omega and
> sigma
> but they always give me nil results.
>
> I appreciate any help on this matter. Thanks!
>
> Regards,
> X.Y. Ng
> /
> *This is an example of my control file:*
>
> $PROB RUN# pd_3_advan6
> $INPUT ID TIME DV AMT CMT ADDL II MDV V2I V3I QI CLI KAI
>
> $DATA FINAL2.2.CSV IGNORE=C
> $SUBROUTINES ADVAN6 TRANS1 TOL=3
> $MODEL
> COMP=DEPOT
> COMP=CENTRAL
> COMP=PERIPH
> COMP=EFFECT ;$MODEL defines the no of compartments in the model
>
> $PK
>
> V2=V2I
> V3=V3I
> Q=QI
> CL=CLI
> KA=KAI
> S2=V2
> S3=V3
> KIN=THETA(1)*EXP(ETA(1))
> KOUT=THETA(2)*EXP(ETA(2))
> IC50=THETA(3)*EXP(ETA(3))
> F4=KIN/KOUT
>
> $DES
> DADT(1)=-KA*A(1)
> DADT(2)=KA*A(1)-Q/V2*A(2)+Q/V3*A(3)-CL/V2*A(2)
> DADT(3)=-Q/V3*A(3)+Q/V2*A(2)
> INH =A(2)/(IC50+A(2))
> DADT(4)=KIN*(1-INH)-KOUT*A(4)
>
> $ERROR
> CP2=A(2)/S2
> CP3=A(3)/S3
> ;IPRED=F
> EFF=A(4)
> Y=EFF+EFF*ERR(1)+ERR(2)
>
> $THETA (0,0.281) ;POPKin
> $THETA (0,0.003) ;POPkout
> $THETA (0,2) ;POPIC50
>
> $OMEGA 0.003 ;BSV Kin
> $OMEGA 0.003 ;BSV Kout
> $OMEGA 0.003 ;BSV IC50
>
> $SIGMA 0.01 ;ERRCCV
> $SIGMA 0.0015 ;ERRADD
>
> $ESTIMATION METHOD=1 INTERACTION NOABORT MAXEVAL=9990 PRINT=10
> POSTHOC
> $COVARIANCE
> $TABLE ID TIME DV AMT CMT NOPRINT ONEHEADER FILE=pd_3_advan6.TAB
>
> *and an example of my data:*
>
> /
> CID TIME DV AMT CMT ADDL II MDV
> V2I V3I QI CLI KAI
> 101 0 0 100 1 2 8 1
> 44.55 11.78 1.07 3.37 0.62
> 101 0 0 1 4 0 0 1
> 44.55 11.78 1.07 3.37 0.62
> 101 3 243 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 7 293 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 11 261 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 15 260 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 19 277 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 23 290 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 35 233 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 39 271 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 43 274 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
> 101 47 276 0 4 0 0 0
> 44.55 11.78 1.07 3.37 0.62
>
> /
> /