# Re: COND LAPLACE LIKELIHOOD

From: Leonid Gibiansky <LGibiansky>
Date: Tue, 04 Aug 2009 18:01:32 -0400

the model is sensitive to initial values. Try to guess B1, B2, B3 (or at
least B1) based on the observed data. Try B1=-1, B2=1, B3=1,

Also, try to add random effect to B1.

--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566

wangx826
> Leonid,
> Thanks for your suggestion. I totally agree the part you indicated was
> incorrect in my original control stream. But after I revised it, NONMEM
> still stopped running and gave the same error message like " CONDITIONAL
> LIKELIHOOD SET TO NEGATIVE VALUE WITH INDIVIDUAL 2 (IN INDIVIDUAL RECORD
> ORDERING), DATA RECORD 13". Is there anything else wrong in my coding? I
> would appreciate pretty much if you or some one else could point it out
> for me.
>
> Thanks again,
> Tianli
>
> On Aug 4 2009, Leonid Gibiansky wrote:
>
>> something is incorrect here, either model or comments:
>> P0 = C0/(1+C0) ; Probability of Score=>0
>> P1 = C1/(1+C1) ; Probability of Score=>1
>> P2 = C2/(1+C2) ; Probability of Score=>2
>>
>> Probability of Score=>0 is always 1 (because score is 0 or positive
>> number as I can see).
>>
>> My guess is that the model is incorrect, it should be (assuming that
>> the drug increases probability of higher scores while natural state
>> favors score 0)
>>
>> > A1 = B1 + DRUG
>> > A2 = B1 - B2 + DRUG
>> > A3 = B1 - B2 - B3 + DRUG
>>
>> > C1 = EXP(A1)
>> > C2 = EXP(A2)
>> > C3 = EXP(A3)
>> > P1 = C1/(1+C1) ; Probability of Score=>1
>> > P2 = C2/(1+C2) ; Probability of Score=>2
>> > P3 = C3/(1+C3) ; Probability of Score=>3
>> >
>> > PR0 = 1-P1 ; Probability of Score=0
>> > PR1 = P1-P2 ; Probability of Score=1
>> > PR2 = P2-P3 ; Probability of Score=2
>> > PR3 = P3 ; Probability of Score=3
>>
>> > IF (DV.EQ.0) Y=PR0
>> > IF (DV.EQ.1) Y=PR1
>> > IF (DV.EQ.2) Y=PR2
>> > IF (DV.EQ.3) Y=PR3
>>
>> B2 and B3 should be positive
>>
>> --------------------------------------
>> Leonid Gibiansky, Ph.D.
>> President, QuantPharm LLC
>> web: www.quantpharm.com
>> e-mail: LGibiansky at quantpharm.com
>> tel: (301) 767 5566
>>
>>
>>
>>
>> wangx826
>>> Hi Samer,
>>>
>>> \$MODEL
>>> COMP=(ABSO)
>>> COMP=(CENT)
>>> COMP=(PERI)
>>> COMP=(EFFECT)
>>> \$PK
>>> CL =ICL*24
>>> V2 = IVC
>>> Q=...
>>> K23=Q/V2
>>> V3=
>>> K32=Q/V3
>>> K=CL/V2
>>> KA=...
>>> KE0 = THETA(6)
>>> B1 = THETA(1)
>>> B2 = THETA(2)
>>> B3 = THETA(3)
>>> EMAX = THETA(4)
>>> EC50 = THETA(5)*EXP(ETA(1))
>>>
>>> \$DES
>>>
>>> \$ERROR
>>> CE=A(4)
>>> DRUG = EMAX*CE/(EC50+CE)
>>> A0 = B1 + DRUG
>>> A1 = B1 + B2 + DRUG
>>> A2 = B1 + B2 + B3 + DRUG
>>> C0 = EXP(A0)
>>> C1 = EXP(A1)
>>> C2 = EXP(A2)
>>> P0 = C0/(1+C0) ; Probability of Score=>0
>>> P1 = C1/(1+C1) ; Probability of Score=>1
>>> P2 = C2/(1+C2) ; Probability of Score=>2
>>>
>>> PR0 = P0 ; Probability of Score=0
>>> PR1 = P1-P0 ; Probability of Score=1
>>> PR2 = P2-P1 ; Probability of Score=2
>>> PR3 = 1-P2 ; Probability of Score=3
>>> IF (DV.EQ.0) Y=PR0
>>> IF (DV.EQ.1) Y=PR1
>>> IF (DV.EQ.2) Y=PR2
>>> IF (DV.EQ.3) Y=PR3
>>>
>>> \$THETA (-20 -6.3) ; THETA1 B1
>>> \$THETA (-10 -0.3) ; THETA2 B2
>>> \$THETA (-10 2) ; THETA3 B3
>>> \$THETA (0 5) ; THETA4 EMAX
>>> \$THETA (0 50) ; THETA5 EC50
>>> \$THETA (0 1) ; THETA6 KEO
>>>
>>> \$OMEGA 2
>>>
>>> \$ESTIMATION MAXEVAL=9999 PRINT=5 METHOD=COND LAPLACIAN LIKELIHOOD
>>> NOABORT MSFO=MSF1
>>> Here I include the dataset for first two subjects. Since it is a
>>> simultaneous PKPD link model, DV in the data set as follows is
>>> categorical PD data. The dose was given daily and DV was recorded
>>> daily as well.
>>> #SUBJ TIME AMT EVID II ADDL PD KA CL V2
>>> MDV
>>> 1 0 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 0 1 1 1 20 . 9.96 4.12 63.57 1
>>> 1 1 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 2 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 3 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 4 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 5 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 6 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 7 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 8 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 9 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 10 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 11 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 12 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 13 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 14 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 15 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 16 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 17 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 18 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 19 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 20 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 21 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 22 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 23 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 24 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 25 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 26 . 0 . . 0 9.96 4.12 63.57 0
>>> 1 27 . 0 . . 0 9.96 4.12 63.57 0
>>> 2 0 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 0 2 1 1 20 . 1.52 4.68 66.91 1
>>> 2 1 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 2 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 3 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 4 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 5 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 6 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 7 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 8 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 9 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 10 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 11 . 0 . . 0 1.52 4.68 66.91 0
>>> 2 12 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 13 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 14 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 15 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 16 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 17 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 18 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 19 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 20 . 0 . . 1 1.52 4.68 66.91 0
>>> 2 21 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 22 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 23 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 24 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 25 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 26 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 27 . 0 . . 3 1.52 4.68 66.91 0
>>> 2 28 . 0 . . 2 1.52 4.68 66.91 0
>>>
>>> Thanks,
>>> Tianli
>>>
>>>
>>> ******************************************************************************
>>>
>>>
>>> Tianli Wang
>>> University of Minnesota,
>>> Department of Pharmaceutics
>>>
>>> On Aug 4 2009, Samer Mouksassi wrote:
>>>
>>>> Can you include your control (or at least your \$EST bloc)
>>>> And the proportional odds PD data. The likelihood may go wild if some
>>>> categories are very rare or non esistent. You need to gard against over
>>>> or underflow.
>>>>
>>>> -----Original Message-----
>>>> From: owner-nmusers
>>>> [mailto:owner-nmusers
>>>> On Behalf Of wangx826
>>>> Sent: 2009-08-04 13:38
>>>> To: NMUSERS
>>>> Subject: [NMusers] COND LAPLACE LIKELIHOOD
>>>>
>>>> Dear nmusers,
>>>>
>>>> Has anybody seen the following error message from NONMEM:
>>>> "CONDITIONAL LIKELIHOOD SET TO NEGATIVE VALUE
>>>> WITH INDIVIDUAL 2 (IN INDIVIDUAL RECORD ORDERING), DATA RECORD 13"?
>>>> What does it mean?
>>>>
>>>> I am trying to use conditional lapalacian likelihood method to generate
>>>> a proportional odds model dealing with categorical PD data. Here is a
>>>> little piece of my data set:
>>>> SUBJ TIME AMT EVID II ADDL DV CL V
>>>> MDV
>>>> 2 0 . 0 . . 0 4.12 63
>>>> 0
>>>> 2 0 1 1 1 20 . 4.12 63
>>>> 1
>>>> 2 0.005 . 2 . . . 4.12 63
>>>> 1
>>>> 2 0.01 . 2 . . . 4.12 63
>>>> 1
>>>> .
>>>> .
>>>> .
>>>> 2 12 . 0 . . 0 4.12 63
>>>> 0
>>>> 2 13 . 0 . . 1 4.12 63
>>>> 0
>>>> .
>>>> .
>>>> .
>>>> But I don't think the problem exists in the data set because I tried
>>>> deleting the rows indicated in the error message, but the exactly
>>>> same message still came out and NONMEM stopped running. I have no
>>>> idea what's
>>>>
>>>> the problem for my model.
>>>>