NONMEM Users Network Archive

Hosted by Cognigen

Re: AW: Time to Event Model

From: Ron Keizer <ron.keizer>
Date: Mon, 15 Aug 2011 16:26:22 +0200

dear Friederike,

I quickly ran your model (with updates from NH / JG) and got no errors
and reasonable results with a recreated dataset, so I don't think there
is an error in the model file. (To focus only on the estimation of
hazard, I removed the PK, and replaced DP with DPRG = TIME/100.)

A few suggestions that may help:
- try to get the hazard model running first (so without PK/DP), and add
these after you get this running.
- do some sanity checks on the model. E.g. simulate all components (PK,
DP, hazard) separately and check if it matches the observed dataset.
Maybe some initial estimates are too far off.
- IIV: I'm not sure you are able to estimate an eta on drug/disease
effect (BETA) in this hazard model with these kind of data (don't know
all details of the dataset). First try with all eta's fixed to 0.

hope this helps,
Ron


On 08/15/2011 02:13 PM, Friederike Kanefendt wrote:
> Dear Nick and Joachim,
>
> thanks a lot for your reply. I corrected the code to TOL=7 and NSIG=2/SIGL=6
> as well as to
> DADT(5)=BLHAZ*EXP(BETA*DPRG) ; CUMULATIVE HAZARD (instead of my assumption of HAZ)
>
> but I have still the ERROR message:
> CONDITIONAL LIKELIHOOD SET TO NEGATIVE VALUE
> WITH INDIVIDUAL 1 (IN INDIVIDUAL RECORD ORDERING), DATA RECORD 1
>
> Do you have any idea what that means?
> Is it possible that I have insufficient information to make a TTE analysis?
>
> Best regards,
> Friederike
>
>
>
> -----Ursprüngliche Nachricht-----
> Von: owner-nmusers
> Gesendet: Montag, 15. August 2011 11:55
> An: nmusers
> Betreff: Re: [NMusers] Time to Event Model
>
> Friederike,
>
> I have not tried to understand all the details of your model but the
> most obvious problem is the way you obtain the cumulative hazard.
>
> Remove the initialization A_0(5)=BLHAZ. The cumulative hazard starts at
> zero.
>
> The hazard in $DES must be the same as the hazard for HAZNOW.
>
> Change
> DADT(5)=BETA*DPRG ; HAZARD RATE
> to
> DADT(5)=BLHAZ*EXP(BETA*DPRG) ; HAZARD RATE
>
> Nick
>
> On 15/08/2011 8:55 p.m., Friederike Kanefendt wrote:
>> Dear NMusers,
>>
>> I try to model the influence of an intervention on Time-To-Event
>> (disease progression) data with NM7 based on the presentation of Nick
>> Holford (PAGE 2011).
>>
>> One problem might be that I have only data from 21 patients with 11
>> event data (DV=1) and 10 right censored data (DV=0)...
>>
>> The treatment influences the hazard rate.
>>
>> (h(t)=h_0 (t)*exp(BETA*X).
>>
>> For X I tested disease progression (DPRG) -affected by ON or OFF
>> treatment-, free drug concentration (C), or the AUC at steady state
>> (AUC_SS).
>>
>> Unfortunately, the estimation aborted with different ERROR messages:
>>
>> 1) for Concentration
>>
>> NUMERICAL DIFFICULTIES WITH INTEGRATION ROUTINE.
>>
>> NO. OF REQUIRED SIGNIFICANT DIGITS IN SOLUTION VECTOR
>>
>> TO DIFFERENTIAL EQUATIONS, 4, MAY BE TOO LARGE.
>>
>> 0PROGRAM TERMINATED BY OBJ
>> --> setting
>> TOL to lower values has no influence
>>
>> 2) for AUC_SS and DPRG
>>
>> CONDITIONAL LIKELIHOOD SET TO NEGATIVE VALUE
>>
>> WITH INDIVIDUAL 1 (IN INDIVIDUAL RECORD ORDERING), DATA RECORD 1
>>
>> Does someone have experience with that kind of error or have any idea
>> what could be the problem?
>>
>> Attached you find my control file and a part of the structure of the
>> data set
>>
>> Thanks in advance.
>>
>> Best regards,
>>
>> Friederike
>>
>> I have a data set with dosing events and dummy observations for the PK
>> model as well as one row with event/exclusion
>>
>> ID TIME AMT DV MDV CLx Vx ...
>>
>> 1, 0, 50, 0, 1, 30, 2000
>>
>> 1, 23.83, 0, 0, 1, 30, 2000
>>
>> 1, 24, 50, 0, 1, 30, 2000
>>
>> 1, 47.83, 0, 0, 1, 30, 2000
>>
>> 1, 48, 50, 0, 1, 30, 2000
>>
>> ...
>>
>> 1, 8280, 0, 1, 0, 30, 2000 ; progression (event)
>>
>> 2, 0, 50, 0, 1, 35, 1800
>>
>> 2, 23.83, 0, 0, 1, 35, 1800
>>
>> 2, 24, 50, 0, 1, 35, 1800
>>
>> 2, 47.83, 0, 0, 1, 35, 1800
>>
>> 2, 48, 50, 0, 1, 35, 1800
>>
>> ...
>>
>> 2, 4236, 0, 0, 0, 35, 1800 ; censored
>>
>> ...
>>
>> $INPUT ID TIME AMT DV MDV CLx Vx ...
>>
>> $DATA data.csv
>>
>> $SUBROUTINE ADVAN6 TOL=4
>>
>> $MODEL
>>
>> NCOMP=5
>>
>> COMP=(DEPOT)
>>
>> COMP=(CENTRAL)
>>
>> COMP=(PERI)
>>
>> COMP=(MET)
>>
>> COMP=(HAZ)
>>
>> $THETA (0,0.5) ; 1 TH_BLHAZ - Baseline Hazard
>>
>> $THETA (0.01) ; 2 TH_BETA - Factor
>>
>> $THETA (5) ; 3 TH_EFFECT
>>
>> $THETA (0,5) ; 4 TH_INTRI
>>
>> $THETA (0,0.5) ; 5 TH_SLOPE
>>
>> $OMEGA 0 FIX ; 1 ETA_HAZ
>>
>> $OMEGA 0.1 ; 2 ETA_BETA
>>
>> $OMEGA 0.1 ; 3 ETA_EFFECT
>>
>> $OMEGA 0.1 ; 4 ETA_INTRI
>>
>> $OMEGA 0.1 ; 5 ETA_SLOPE
>>
>> $PK
>>
>> ;HAZARD
>>
>> TVBLHAZ = THETA(1)
>>
>> BLHAZ = TVBLHAZ*EXP(ETA(1))
>>
>> TVBETA = THETA(2)
>>
>> BETA = TVBETA*EXP(ETA(2))
>>
>> ;SYMPTOMATIC TREATMENT EFFECT
>>
>> EFFECT = THETA(3)*EXP(ETA(3)) ; TREATMENT EFFECT FACTOR
>>
>> ;DISEASE PROGRESS
>>
>> INTRI = THETA(4)*EXP(ETA(4)) ; INTERCEPT OF DISEASE PROGRESSION
>>
>> SLOPE = THETA(5)*EXP(ETA(5)) ; SLOPE OF DISEASE PROGRESSION
>>
>> ;PHARMACOKINETIC
>>
>> ...
>>
>> ;EXPOSURE OF TOTAL DRUG AT STEADY-STATE
>>
>> AUC_SS = DOSE/CLx+DOSE/CLM
>>
>> A_0(5)=BLHAZ
>>
>> $DES
>>
>> ...
>>
>> C=A(2)/V1+A(4)/VM
>>
>> IF(C.GE.50) THEN ; EFFECTIVE CONCENTRATION
>>
>> TREA = 1
>>
>> ELSE
>>
>> TREA = 0
>>
>> ENDIF
>>
>> INTRC = INTRI-EFFECT*TREA
>>
>> DPRG = INTRC+SLOPE*T
>>
>> DADT(5)=BETA*DPRG ; HAZARD RATE
>>
>> ;DADT(5)=BETA*C
>>
>> ;DADT(5)=BETA*AUC_SS
>>
>> $ERROR
>>
>> CUB = A(2)/V1+A(4)/VM
>>
>> CUMHAZ = A(5) ; CUMULATIVE HAZARD
>>
>> ;EFFECTIVE CONCENTRATION
>>
>> IF(CUB.GE.50) THEN ; CONC EFFECTIVE
>>
>> TREAT = 1
>>
>> ELSE
>>
>> TREAT = 0
>>
>> ENDIF
>>
>> INTR = INTRI-EFFECT*TREAT ; INTERCEPT OF DISEASE PROGRESSION
>>
>> DISPRG = INTR+SLOPE*TIME
>>
>> SURV = EXP(-CUMHAZ) ; SURVIVAL FUNCTION -
>> probability not to have an event
>>
>> IF(DV.EQ.1) THEN ; EVENT
>>
>> HAZNOW = BLHAZ*EXP(BETA*DISPRG) ; HAZARD RATE AT THAT TIME
>>
>> Y = HAZNOW*SURV ; PDF - PROBABILITY
>> DENSITY FUNCTION
>>
>> ELSE ;
>> CENSORED
>>
>> Y = SURV
>>
>> ENDIF
>>
>> $ESTIMATION SIG=3 SIGL=9 MAXEVAL=9990 METHOD=COND LAPLACE LIKE PRINT=1
>>
>> $COVARIANCE PRINT=E
>>
>> $TABLE ID TIME BLHAZ SURV HAZNOW CUMHAZ NOPRINT ONEHEADER NOAPPEND
>> FILE=PATAB
>>
>> Friederike Kanefendt
>>
>> - PhD-Student -
>>
>> University of Bonn, Germany
>>
>> -Clinical Pharmacy-
>>
>> Phone: +49 (0)228 73-5781
>>
>> Fax: +49(0) 228 73-9757
>>
>> friederike.kanefendt
>>


--
-----------------------------------
Ron Keizer, PharmD PhD
Post-doctoral fellow
Pharmacometrics Research Group
Uppsala University
-----------------------------------
Received on Mon Aug 15 2011 - 10:26:22 EDT

The NONMEM Users Network is maintained by ICON plc. Requests to subscribe to the network should be sent to: nmusers-request@iconplc.com.

Once subscribed, you may contribute to the discussion by emailing: nmusers@globomaxnm.com.