Re: [NMusers] Can you post this question?

From: Jakob Ribbing <jakob.ribbing_at_pharmetheus.com>
Date: Tue, 2 Aug 2016 16:16:40 +0200

Hi Ali,

If drop out generally only occurs at visits (or as in your case, where =
the exact time of dropout is unknown) it can sometimes make sense to use =
interval-censored TTE, and I would agree in your case the best would be =
to assume that the time of dropout was between subjects last visit and =
the next planned visit.

In either case the dropout record should be placed after the last PD =
observation. For subjects that miss intermediate visits, but that are =
not discontinued from the study, such missing records do not count as a =
dropout event, so this does not pose a great problem.
Ideally, VPC evaluations should allow any simulated subject that do not =
drop out (in simulation) to present data until the planned end of study, =
so for such evaluations you may want to add dummy records (or depending =
on how simulations are preformed: one dummy record, at the planned end =
of study), in subjects that dropped out in your original analysis data =
set.

With regards to modelling dropout, do you need a simultaneous fit of PD =
and dropout, or is it sufficient to combine the two for simulations? =
(i.e. PD estimates are not much affected by the incorporation of a =
dropout model, whereas PD simulations are)
The IPRED for PD would then be fixed according to your final PD model, =
when you subsequently develop the TTE model for dropout.
This may simplify a great deal, and still allows realistic simulations =
(e.g. for a VPC evaluation), that can confirm that your final PD model =
is adequate for simulations.

Finally, I noticed in your data set LOCF contains the previous DV value =
(for DVID=1). Normally, by LOCF, you only carry forward the previous =
value if a planned measurement is missing, so this is not what normally =
is called LOCF.
But maybe this is just an unfortunate name, and what you need from this =
variable is the previous DV value?

I did not find time to check your control steam, but if there is any =
obvious error I am sure someone else will spot that.

Best wishes

Jakob





Jakob Ribbing, Ph.D.

Senior Consultant, Pharmetheus AB



Cell/Mobile: +46 (0)70 514 33 77

Jakob.Ribbing_at_Pharmetheus.com

www.pharmetheus.com



Phone, Office: +46 (0)18 513 328

Uppsala Science Park, Dag Hammarskjölds väg 52B

SE-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 02 Aug 2016, at 14:51, Ali Alhadab <alhad009_at_umn.edu> wrote:

> Please post this Jointing Modeling of Dropout
>
>
>
> I am trying to jointly model PD and dropout and I have few questions =
for those who have experience with this model.
>
> Does it matter if I use right or interval censored event model? What =
is the advantage of using one over the other?
>
> The dropout record should be the last record for each patient in a =
dataset, isn’t it? If I want to do interval censored and I only know =
the last time a subject is known to be in the trial, can I use the last =
time observed and time of next scheduled visit for my interval? Or I =
only can do right censoring in this case.
>
> If dropout out data was not collected and I need to account for that, =
I can develop criteria to define dropout after the fact that study is =
over, can’t I? For example, subjects who did not show up for at least =
three visits (first thee visits) are considered missing, or subjects who =
have been in the study less than 14 days are considered missing. What if =
a subject made it to the 1st and 5th visit but missed those in between? =
Any suggestions how to do that?
>
> When I run my joint model, I get the following error: ”SQUARED" =
WEIGHTED INDIVIDUAL RESIDUALS IS INFINITE” or “NO. OF REQUIRED =
SIGNIFICANT DIGITS IN SOLUTION VECTOR TO DIFFERENTIAL EQUATIONS, 5, MAY =
BE TOO LARGE”. I tried to use different ADVAN (6,8,9) and reduce TOL =
(6,5,4,3,2,1) but that did not solve the problem. Any idea what the =
problem is?
>
> ID
> TIME
> DV
> LOCF
> DVID
> CMT
> 1
> 0
> 37
> 0
> 1
> 1
> 1
> 14
> 18
> 37
> 1
> 1
> 1
> 30
> 14
> 18
> 1
> 1
> 1
> 58
> 7
> 14
> 1
> 1
> 1
> 62
> 0
> 7
> 2
> 1
> 2
> 0
> 22
> 0
> 1
> 1
> 2
> 0
> 0
> 0
> 3
> 2
> 2
> 25
> 23
> 22
> 1
> 1
> 2
> 34
> 1
> 22
> 2
> 1
>
> $SUB ADVAN=6 TOL=9
> $MODEL COMP=(HAZARD)
> $PK
> ;;; PD COUNT MODEL ;;;
> S0=THETA(1)*EXP(ETA(1))+THETA(10)*(SEX-1)
> IF (REGION.EQ.6) THEN
> REG=THETA(12)
> ELSE
> REG=0
> ENDIF
>
> IF (TX.EQ.1) =
SSS=THETA(2)*EXP(ETA(2))+THETA(10)*(SEX-1)+REG+THETA(13)*MISSTOT
> IF (TX.EQ.2) =
SSS=THETA(3)*EXP(ETA(2))+THETA(10)*(SEX-1)+REG+THETA(13)*MISSTOT
> IF (TX.EQ.3) =
SSS=THETA(4)*EXP(ETA(2))+THETA(10)*(SEX-1)+REG+THETA(13)*MISSTOT
> IF (TX.EQ.4) =
SSS=THETA(5)*EXP(ETA(2))+THETA(10)*(SEX-1)+REG+THETA(13)*MISSTOT
>
> IF (TX.EQ.1) =
TPROG=THETA(6)*EXP(ETA(3))+(B_INF/25.86)**THETA(11)+REG+SEASON**THETA(13=
)
> IF (TX.EQ.2) =
TPROG=THETA(7)*EXP(ETA(3))+(B_INF/25.86)**THETA(11)+REG+SEASON**THETA(13=
)
> IF (TX.EQ.3) =
TPROG=THETA(8)*EXP(ETA(3))+(B_INF/25.86)**THETA(11)+REG+SEASON**THETA(13=
)
> IF (TX.EQ.4) =
TPROG=THETA(9)*EXP(ETA(3))+(B_INF/25.86)**THETA(11)+REG+SEASON**THETA(13=
)
>
> ;;; DROPOUT MODEL ;;;
> BASE = THETA(14)
> SHP = THETA(15)
> LAM = BASE*SHP
> BETA = SHP-1
> BETA1 = THETA(16)
>
> $DES
> ;;; PD COUNT MODEL ;;;
> DCOUNT=S0+(SSS-S0)*(1-EXP(-LN2/TPROG*T))
>
> ;;; DROPOUT MODEL ;;;
> IF(T.GT.0)THEN
> DADT(1) = LAM*EXP(BETA*LOG(BASE*T)+DCOUNT*BETA1)
> ELSE
> DADT(1) = 0
> ENDIF
>
> $ERROR
> COUNT=S0+(SSS-S0)*(1-EXP(-LN2/TPROG*TIME)) ;RENAME IPRED
>
> CHZ = A(1) ;rename old cumulative hazard
> SUR = EXP(-CHZ) ;survival probability
>
> IF(TIME.GT.0)THEN
> HAZNOW=LAM*EXP(BETA*LOG(BASE*TIME)+COUNT*BETA1)
> ELSE
> HAZNOW = 0
> ENDIF
>
> IF(DVID.EQ.1) THEN
> F_FLAG=0
> Y=COUNT+ERR(1) ;COUNT PREDICTION
> ENDIF
>
> IF(DVID.EQ.2.AND.DV.EQ.1) THEN
> F_FLAG=1
> Y=SUR*HAZNOW ;DROP OUT EVENT
> ENDIF
>
> IF(DVID.EQ.2.AND.DV.EQ.0) THEN
> F_FLAG=1
> Y=SUR ;RIGHT CENSORED EVENT
> ENDIF
>
> $EST MAXEVAL=9990 METHOD=COND LAPLACIAN
>
>
> Thanks
>
> Ali Alhadab, PharmD | PhD student
> University of Minnesota College of Pharmacy
> Department of Experimental & Clinical Pharmacology
> E-mail: alhad009_at_umn.edu | Cell:541-740-7991



Received on Tue Aug 02 2016 - 10:16:40 EDT

This archive was generated by hypermail 2.3.0 : Fri Sep 27 2019 - 16:51:22 EDT