Dear colleagues,
could someone give me an advice about the rational of using IOV in a particular circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a chemotherapy and the patients have received up to 7 cures. The issue is that the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as for 2nd and 3rd). Shall we use as many occasions as cures (7 in our dataset)? In that case, how many patients by occ is relevant as a minimum? For certain occ we may have few patients. Combining cures is hazardous and has no clinical justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
Rational of using IOV
Nicolas,
Within subject variability in a parameter (e.g. clearance) can be predictable ('fixed effect') e.g. associated with maturation of renal function in young children, or can be unpredictable ('random effect') i.e. apparently varying randomly from occasion to occasion.
IOV is usually used to describe the random between occasion variability. Because this is by assumption a random source of variability it makes no difference which occasion (or treatment) you associate it with. In your case you have 3 occasions for each patient. Just number the occasion 1,2 and 3 for every patient and estimate IOV in the usual way as a random effect with variance described by OMEGA.
I assume you use the word 'cure' to mean treatment rather than complete remission of the disease. If you think there may be non-random differences between occasions e.g. due to different treatments on each occasion, then you can model these with a predictable ('fixed effect') treatment covariate model with each treatment type representing a different value of the treatment covariate. If you only have a few patients for a particular treatment type then of course you are unlikely to detect a treatment type effect and should be very cautious about accepting any effect you find as being real (Ribbing & Johnson 2004).
Best wishes,
Nick
Ribbing J, Jonsson EN. Power, Selection Bias and Predictive Performance of the Population Pharmacokinetic Covariate Model. Journal of Pharmacokinetics and Pharmacodynamics. 2004;31(2):109-34.
Quoted reply history
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
> Dear colleagues,
>
> could someone give me an advice about the rational of using IOV in a particular circumstance?
>
> We have data from a clin trial with 3 occasions for each patient. It was a chemotherapy and the patients have received up to 7 cures. The issue is that the 3 occasions differ from one patient to another.
>
> Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on cure 2, 5 and 6 or whatever...
>
> It seems to me that combining the 1st occ of all patients is meaningless (as for 2nd and 3rd). Shall we use as many occasions as cures (7 in our dataset)? In that case, how many patients by occ is relevant as a minimum? For certain occ we may have few patients. Combining cures is hazardous and has no clinical justification.
>
> Best regards
> Nicolas
>
> Pr Nicolas SIMON
> Universite de la Mediterranee (Aix-Marseille II)
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology& Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
Hi Nicolas
My short answer would be another question: "what is the aim of your analysis ?"
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring interesting
insight into the mechanisms explaining variability. But if you want to use your
results for prediction, e.g. to devise a TDM program, you won't be able to draw
relevant information from IOV: a blood sample obtained in a patient on a
certain occasion won't inform you on your patient's behavior on another
occasion; in this situation, a model merely distinguishing inter-individual and
intra-individual variability components is easier to exploit. Thus, there may
be good reasons not to use IOV even when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
Quoted reply history
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a particular
circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case, how
many patients by occ is relevant as a minimum? For certain occ we may have few
patients. Combining cures is hazardous and has no clinical justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
Hi Thierry,
Actually, devising a TDM program is precisely when you should be evaluating
whether you have substantial IOV. If IOV is considerably greater than IIV
then there is little benefit in a TDM program as you point out since a
concentration from one occasion may not contain much information about the
next occasion. In this setting a TDM program would be "chasing noise".
However, simply fitting a model that does not partition out IOV doesn't mean
that IOV doesn't exist and you will still have problems using such a model
in a TDM program. I do agree there are times when you don't need to
partition out IOV (particularly if IOV is small) but if you know you plan to
use the model in a TDM program that is one of the reasons for evaluating
whether IOV is a big contributor to the total variation.
Ken
Kenneth G. Kowalski
President & CEO
A2PG - Ann Arbor Pharmacometrics Group, Inc.
110 E. Miller Ave., Garden Suite
Ann Arbor, MI 48104
Work: 734-274-8255
Cell: 248-207-5082
Fax: 734-913-0230
[email protected]
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Buclin Thierry
Sent: Monday, November 01, 2010 7:56 AM
To: [email protected]
Subject: [NMusers] Rational of using IOV
Hi Nicolas
My short answer would be another question: "what is the aim of your analysis
?"
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring
interesting insight into the mechanisms explaining variability. But if you
want to use your results for prediction, e.g. to devise a TDM program, you
won't be able to draw relevant information from IOV: a blood sample obtained
in a patient on a certain occasion won't inform you on your patient's
behavior on another occasion; in this situation, a model merely
distinguishing inter-individual and intra-individual variability components
is easier to exploit. Thus, there may be good reasons not to use IOV even
when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a
particular circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case,
how many patients by occ is relevant as a minimum? For certain occ we may
have few patients. Combining cures is hazardous and has no clinical
justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
Dear Thierry,
The article below nicely describes the importance of incorporating IOV since
it otherwise might lead to a falsely optimistic impression of the potential
value of TDM.
Karlsson MO, Sheiner LB. The importance of modeling interoccasion
variability in population
pharmacokinetic analyses. J Pharmacokinet Biopharm. 1993 Dec;21(6):735-50.
Best regards,
Ulrika
Ulrika Simonsson, PhD
Assoc Prof of Pharmacometrics
Uppsala Pharmacometrics
Department of Pharmaceutical Biosciences
Uppsala University
BMC, Box 591, 751 24 Uppsala
Sweden
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Buclin Thierry
Sent: den 1 november 2010 14:35
To: [email protected]
Subject: RE: [NMusers] Rational of using IOV
Dear James,
I always thought that intra-individual variability (IIV) classically
represented the immovable limit on the gains to be expected from TDM IOV
being indeed used only in a minority of population PK analyses. Both intra-
and inter-occasion variability actually represent nuisance. We agree on the
point that specifying an IOV term in a model will decrease the residual IIV.
But wouldnt this precisely give a falsely favorable impression about
potential gains from a TDM program? Am I wrong to think so?
Kind regards
Thierry
De : James G Wright [mailto:[email protected]]
Envoyé : lundi, 1 novembre 2010 14:04
À : Buclin Thierry
Objet : Re: [NMusers] Rational of using IOV
Dear Thierry,
I hope you are well. I think you are right to highlight the importance of
IOV for TDM, but I would argue it is very important to include it in the
model. This is because IOV places an immovable limit on the gains from TDM.
The classic error is to develop a TDM strategy mistakenly lumping IOV with
IIV, and drastically over-estimating the utility of TDM.
Best regards, James
On 01/11/2010 11:55, Buclin Thierry wrote:
Hi Nicolas
My short answer would be another question: what is the aim of your analysis
?
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring
interesting insight into the mechanisms explaining variability. But if you
want to use your results for prediction, e.g. to devise a TDM program, you
wont be able to draw relevant information from IOV: a blood sample obtained
in a patient on a certain occasion wont inform you on your patients
behavior on another occasion; in this situation, a model merely
distinguishing inter-individual and intra-individual variability components
is easier to exploit. Thus, there may be good reasons not to use IOV even
when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a
particular circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case,
how many patients by occ is relevant as a minimum? For certain occ we may
have few patients. Combining cures is hazardous and has no clinical
justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
--
James G Wright PhD,
Scientist, Wright Dose Ltd
Tel: UK (0)772 5636914
Thierry,
Between subject variability (BSV aka IIV) and within subject variability (WSV aka IOV) describe the limits of what we can identify as sources of variability.
I don't consider this a nuisance -- it is an opportunity for learning. The random assumption used for estimation of WSV is a convenient way of describing the size of the problem. If we recognize there is a large element of WSV then it may stimulate thinking and further investigation to try and understand it.
Ignoring WSV will give a false impression about what can be gained from TCI (aka TDM). TCI can only hope to remove the BSV part of unpredictable variability.
See Holford NH. Target concentration intervention: beyond Y2K. Br J Clin Pharmacol. 1999;48(1):9-13.
Best wishes,
Nick
Quoted reply history
On 2/11/2010 2:34 a.m., Buclin Thierry wrote:
> Dear James,
>
> I always thought that intra-individual variability (IIV) classically represented the immovable limit on the gains to be expected from TDM -- IOV being indeed used only in a minority of population PK analyses. Both intra- and inter-occasion variability actually represent nuisance. We agree on the point that specifying an IOV term in a model will decrease the residual IIV. But wouldn't this precisely give a falsely favorable impression about potential gains from a TDM program? Am I wrong to think so?
>
> Kind regards
>
> Thierry
>
> *De :*James G Wright [mailto:[email protected]]
> *Envoyé :* lundi, 1 novembre 2010 14:04
> *À :* Buclin Thierry
> *Objet :* Re: [NMusers] Rational of using IOV
>
> Dear Thierry,
>
> I hope you are well. I think you are right to highlight the importance of IOV for TDM, but I would argue it is very important to include it in the model. This is because IOV places an immovable limit on the gains from TDM. The classic error is to develop a TDM strategy mistakenly lumping IOV with IIV, and drastically over-estimating the utility of TDM.
>
> Best regards, James
>
> On 01/11/2010 11:55, Buclin Thierry wrote:
>
> Hi Nicolas
>
> My short answer would be another question: "what is the aim of your analysis ?"
>
> IOV is fine to split variability into inter-individual, intra-individual-inter-occasion and intra-individual-intra-occasion random components. This is excellent for data description, and can bring interesting insight into the mechanisms explaining variability. But if you want to use your results for prediction, e.g. to devise a TDM program, you won't be able to draw relevant information from IOV: a blood sample obtained in a patient on a certain occasion won't inform you on your patient's behavior on another occasion; in this situation, a model merely distinguishing inter-individual and intra-individual variability components is easier to exploit. Thus, there may be good reasons not to use IOV even when it would be possible.
>
> Kind regards
>
> Thierry
>
> Thierry Buclin, MD, PD,
>
> Division of Clinical Pharmacology and Toxicology
>
> University Hospital of Lausanne (CHUV)
>
> Lausanne - SWITZERLAND
>
> tel +41 21 314 42 61
>
> fax +41 21 314 42 66
>
> On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
>
> Dear colleagues,
>
> could someone give me an advice about the rational of using IOV in a particular circumstance?
>
> We have data from a clin trial with 3 occasions for each patient. It was a chemotherapy and the patients have received up to 7 cures. The issue is that the 3 occasions differ from one patient to another.
>
> Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on cure 2, 5 and 6 or whatever...
>
> It seems to me that combining the 1st occ of all patients is meaningless (as for 2nd and 3rd). Shall we use as many occasions as cures (7 in our dataset)? In that case, how many patients by occ is relevant as a minimum? For certain occ we may have few patients. Combining cures is hazardous and has no clinical justification.
>
> Best regards
> Nicolas
>
> Pr Nicolas SIMON
> Universite de la Mediterranee (Aix-Marseille II)
>
> --
> James G Wright PhD,
> Scientist, Wright Dose Ltd
> Tel: UK (0)772 5636914
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology& Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
Nick
While I agree that BOV is not solely a nuisance parameter it is a design
specific parameter and hence can be somewhat of a nuisance. By design specific
we can formulate settings in which the design of the study changes the estimate
of BOV.
To estimate the variance between occasions the duration of the occasion needs
to be defined (a priori). If the occasion is long then the estimate of BOV
will tend to zero since the integral over the occasion to get the average
parameter value will integrate over the random variability. If the occasion is
short then it will tend to a larger positive number. Imagine an occasion of 1
hour versus 1 year. I realise that most tend to use a dose interval as an
occasion but this is also arbitrary as is clinic visits. The duration of the
occasion would need to be indexed to the substantive inferences of the model to
ensure that any influence that BOV has can be assessed in terms of model
predictions.
Given that BOV is design specific then how should this be interpreted in any
given circumstance? Note that being design specific doesn't preclude the
benefit of BOV in its role as an estimable but nuisance parameter (i.e. to
reduce bias in estimates of the population mean parameter values).
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 56 Dunedin
New Zealand
E: [email protected]<mailto:[email protected]>
P: +64 3 479 5044
F: +64 3 479 7034
W: http://pharmacy.otago.ac.nz/profiles/stephenduffull
Design software: http://www.winpopt.com
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Nick Holford
Sent: Tuesday, 2 November 2010 8:33 a.m.
To: [email protected]
Subject: Re: [NMusers] Rational of using IOV
Thierry,
Between subject variability (BSV aka IIV) and within subject variability (WSV
aka IOV) describe the limits of what we can identify as sources of variability.
I don't consider this a nuisance -- it is an opportunity for learning. The
random assumption used for estimation of WSV is a convenient way of describing
the size of the problem. If we recognize there is a large element of WSV then
it may stimulate thinking and further investigation to try and understand it.
Ignoring WSV will give a false impression about what can be gained from TCI
(aka TDM). TCI can only hope to remove the BSV part of unpredictable
variability.
See Holford NH. Target concentration intervention: beyond Y2K. Br J Clin
Pharmacol. 1999;48(1):9-13.
Best wishes,
Nick
On 2/11/2010 2:34 a.m., Buclin Thierry wrote:
Dear James,
I always thought that intra-individual variability (IIV) classically
represented the immovable limit on the gains to be expected from TDM - IOV
being indeed used only in a minority of population PK analyses. Both intra- and
inter-occasion variability actually represent nuisance. We agree on the point
that specifying an IOV term in a model will decrease the residual IIV. But
wouldn't this precisely give a falsely favorable impression about potential
gains from a TDM program? Am I wrong to think so?
Kind regards
Thierry
De : James G Wright [mailto:[email protected]]
Envoyé : lundi, 1 novembre 2010 14:04
À : Buclin Thierry
Objet : Re: [NMusers] Rational of using IOV
Dear Thierry,
I hope you are well. I think you are right to highlight the importance of IOV
for TDM, but I would argue it is very important to include it in the model.
This is because IOV places an immovable limit on the gains from TDM. The
classic error is to develop a TDM strategy mistakenly lumping IOV with IIV, and
drastically over-estimating the utility of TDM.
Best regards, James
On 01/11/2010 11:55, Buclin Thierry wrote:
Hi Nicolas
My short answer would be another question: "what is the aim of your analysis ?"
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring interesting
insight into the mechanisms explaining variability. But if you want to use your
results for prediction, e.g. to devise a TDM program, you won't be able to draw
relevant information from IOV: a blood sample obtained in a patient on a
certain occasion won't inform you on your patient's behavior on another
occasion; in this situation, a model merely distinguishing inter-individual and
intra-individual variability components is easier to exploit. Thus, there may
be good reasons not to use IOV even when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a particular
circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case, how
many patients by occ is relevant as a minimum? For certain occ we may have few
patients. Combining cures is hazardous and has no clinical justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
--
James G Wright PhD,
Scientist, Wright Dose Ltd
Tel: UK (0)772 5636914
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]<mailto:[email protected]>
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
Hi Steve,
I think we can apply some mechanistic insight into IOV in most cases. For
example for absorption parameters, it is difficult to see that anything but
each dosing occasion would constitute a separate occasion. There may however
be situations where it is difficult to judge properly and IOV may not be
modeled ideally. Then however, not only IOV, but also IIV and RV become
nuisance parameters as they will not represent the true IIV or RV.
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Sweden
Postal address: Box 591, 751 24 Uppsala, Sweden
Phone +46 18 4714105
Fax + 46 18 4714003
Quoted reply history
From: [email protected] [mailto:[email protected]] On
Behalf Of Stephen Duffull
Sent: Monday, November 01, 2010 10:49 PM
To: Nick Holford; [email protected]
Subject: RE: [NMusers] Rational of using IOV
Nick
While I agree that BOV is not solely a nuisance parameter it is a design
specific parameter and hence can be somewhat of a nuisance. By design
specific we can formulate settings in which the design of the study changes
the estimate of BOV.
To estimate the variance between occasions the duration of the occasion
needs to be defined (a priori). If the occasion is long then the estimate
of BOV will tend to zero since the integral over the occasion to get the
average parameter value will integrate over the random variability. If the
occasion is short then it will tend to a larger positive number. Imagine an
occasion of 1 hour versus 1 year. I realise that most tend to use a dose
interval as an occasion but this is also arbitrary as is clinic visits. The
duration of the occasion would need to be indexed to the substantive
inferences of the model to ensure that any influence that BOV has can be
assessed in terms of model predictions.
Given that BOV is design specific then how should this be interpreted in any
given circumstance? Note that being design specific doesnt preclude the
benefit of BOV in its role as an estimable but nuisance parameter (i.e. to
reduce bias in estimates of the population mean parameter values).
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 56 Dunedin
New Zealand
E: [email protected]
P: +64 3 479 5044
F: +64 3 479 7034
W: http://pharmacy.otago.ac.nz/profiles/stephenduffull
Design software: www.winpopt.com
From: [email protected] [mailto:[email protected]] On
Behalf Of Nick Holford
Sent: Tuesday, 2 November 2010 8:33 a.m.
To: [email protected]
Subject: Re: [NMusers] Rational of using IOV
Thierry,
Between subject variability (BSV aka IIV) and within subject variability
(WSV aka IOV) describe the limits of what we can identify as sources of
variability.
I don't consider this a nuisance -- it is an opportunity for learning. The
random assumption used for estimation of WSV is a convenient way of
describing the size of the problem. If we recognize there is a large element
of WSV then it may stimulate thinking and further investigation to try and
understand it.
Ignoring WSV will give a false impression about what can be gained from TCI
(aka TDM). TCI can only hope to remove the BSV part of unpredictable
variability.
See Holford NH. Target concentration intervention: beyond Y2K. Br J Clin
Pharmacol. 1999;48(1):9-13.
Best wishes,
Nick
On 2/11/2010 2:34 a.m., Buclin Thierry wrote:
Dear James,
I always thought that intra-individual variability (IIV) classically
represented the immovable limit on the gains to be expected from TDM IOV
being indeed used only in a minority of population PK analyses. Both intra-
and inter-occasion variability actually represent nuisance. We agree on the
point that specifying an IOV term in a model will decrease the residual IIV.
But wouldnt this precisely give a falsely favorable impression about
potential gains from a TDM program? Am I wrong to think so?
Kind regards
Thierry
De : James G Wright [mailto:[email protected]]
Envoyé : lundi, 1 novembre 2010 14:04
À : Buclin Thierry
Objet : Re: [NMusers] Rational of using IOV
Dear Thierry,
I hope you are well. I think you are right to highlight the importance of
IOV for TDM, but I would argue it is very important to include it in the
model. This is because IOV places an immovable limit on the gains from TDM.
The classic error is to develop a TDM strategy mistakenly lumping IOV with
IIV, and drastically over-estimating the utility of TDM.
Best regards, James
On 01/11/2010 11:55, Buclin Thierry wrote:
Hi Nicolas
My short answer would be another question: what is the aim of your analysis
?
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring
interesting insight into the mechanisms explaining variability. But if you
want to use your results for prediction, e.g. to devise a TDM program, you
wont be able to draw relevant information from IOV: a blood sample obtained
in a patient on a certain occasion wont inform you on your patients
behavior on another occasion; in this situation, a model merely
distinguishing inter-individual and intra-individual variability components
is easier to exploit. Thus, there may be good reasons not to use IOV even
when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a
particular circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case,
how many patients by occ is relevant as a minimum? For certain occ we may
have few patients. Combining cures is hazardous and has no clinical
justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
--
James G Wright PhD,
Scientist, Wright Dose Ltd
Tel: UK (0)772 5636914
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
Thanks to all for your comments, I didn't expect so much.
If the occasion correspond to a specific dose (i.e. 1st dose or later) then the question is, shall we expect a variability related to the duration of the treatment? In that case, we will have as many occasions as doses. The number of occasion can inflate rapidely and likely with few patients per occ. That doesn't sound good for me. Should we expect something like an overparameterization?
However, from a clinical point of view, a dose = an occasion, seems reasonable when the pathology may interfere with the PK (i.e. renal function and antibiotics for pyelonephritis, cancer and overall well being). If we combine observations coming from different doses or chemotherapy cures (in my example), I am not sure that it's relevant and even that could lead to false interpretation.
Best regards
Nicolas
----- Message de [email protected] ---------
Date : Tue, 2 Nov 2010 03:10:48 +0100
De : mats karlsson <[email protected]>
Répondre à : [email protected]
Objet : RE: [NMusers] Rational of using IOV
À : 'Stephen Duffull' < [email protected] >, 'Nick Holford' < [email protected] >, [email protected]
> Hi Steve,
>
> I think we can apply some mechanistic insight into IOV in most cases. For
> example for absorption parameters, it is difficult to see that anything but
> each dosing occasion would constitute a separate occasion. There may however
> be situations where it is difficult to judge properly and IOV may not be
> modeled ideally. Then however, not only IOV, but also IIV and RV become
> nuisance parameters as they will not represent the true IIV or RV.
>
> Best regards,
>
> Mats
>
> Mats Karlsson, PhD
>
> Professor of Pharmacometrics
>
> Dept of Pharmaceutical Biosciences
>
> Uppsala University
>
> Sweden
>
> Postal address: Box 591, 751 24 Uppsala, Sweden
>
> Phone +46 18 4714105
>
> Fax + 46 18 4714003
>
Quoted reply history
> From: [email protected] [mailto:[email protected]] On
> Behalf Of Stephen Duffull
> Sent: Monday, November 01, 2010 10:49 PM
> To: Nick Holford; [email protected]
> Subject: RE: [NMusers] Rational of using IOV
>
> Nick
>
> While I agree that BOV is not solely a nuisance parameter it is a design
> specific parameter and hence can be somewhat of a nuisance. By design
> specific we can formulate settings in which the design of the study changes
> the estimate of BOV.
>
> To estimate the variance between occasions the duration of the occasion
> needs to be defined (a priori). If the occasion is long then the estimate
> of BOV will tend to zero since the integral over the occasion to get the
> average parameter value will integrate over the random variability. If the
> occasion is short then it will tend to a larger positive number. Imagine an
> occasion of 1 hour versus 1 year. I realise that most tend to use a dose
> interval as an occasion but this is also arbitrary as is clinic visits. The
> duration of the occasion would need to be indexed to the substantive
> inferences of the model to ensure that any influence that BOV has can be
> assessed in terms of model predictions.
>
> Given that BOV is design specific then how should this be interpreted in any
> given circumstance? Note that being design specific doesnt preclude the
> benefit of BOV in its role as an estimable but nuisance parameter (i.e. to
> reduce bias in estimates of the population mean parameter values).
>
> Steve
>
> --
>
> Professor Stephen Duffull
>
> Chair of Clinical Pharmacy
>
> School of Pharmacy
>
> University of Otago
>
> PO Box 56 Dunedin
>
> New Zealand
>
> E: [email protected]
>
> P: +64 3 479 5044
>
> F: +64 3 479 7034
>
> W: http://pharmacy.otago.ac.nz/profiles/stephenduffull
>
> Design software: www.winpopt.com
>
> From: [email protected] [mailto:[email protected]] On
> Behalf Of Nick Holford
> Sent: Tuesday, 2 November 2010 8:33 a.m.
> To: [email protected]
> Subject: Re: [NMusers] Rational of using IOV
>
> Thierry,
>
> Between subject variability (BSV aka IIV) and within subject variability
> (WSV aka IOV) describe the limits of what we can identify as sources of
> variability.
>
> I don't consider this a nuisance -- it is an opportunity for learning. The
> random assumption used for estimation of WSV is a convenient way of
> describing the size of the problem. If we recognize there is a large element
> of WSV then it may stimulate thinking and further investigation to try and
> understand it.
>
> Ignoring WSV will give a false impression about what can be gained from TCI
> (aka TDM). TCI can only hope to remove the BSV part of unpredictable
> variability.
>
> See Holford NH. Target concentration intervention: beyond Y2K. Br J Clin
> Pharmacol. 1999;48(1):9-13.
>
> Best wishes,
>
> Nick
>
> On 2/11/2010 2:34 a.m., Buclin Thierry wrote:
>
> Dear James,
>
> I always thought that intra-individual variability (IIV) classically
> represented the immovable limit on the gains to be expected from TDM IOV
> being indeed used only in a minority of population PK analyses. Both intra-
> and inter-occasion variability actually represent nuisance. We agree on the
> point that specifying an IOV term in a model will decrease the residual IIV.
> But wouldnt this precisely give a falsely favorable impression about
> potential gains from a TDM program? Am I wrong to think so?
>
> Kind regards
>
> Thierry
>
> De : James G Wright [mailto:[email protected]]
> Envoyé : lundi, 1 novembre 2010 14:04
> À : Buclin Thierry
> Objet : Re: [NMusers] Rational of using IOV
>
> Dear Thierry,
>
> I hope you are well. I think you are right to highlight the importance of
> IOV for TDM, but I would argue it is very important to include it in the
> model. This is because IOV places an immovable limit on the gains from TDM.
> The classic error is to develop a TDM strategy mistakenly lumping IOV with
> IIV, and drastically over-estimating the utility of TDM.
>
> Best regards, James
>
> On 01/11/2010 11:55, Buclin Thierry wrote:
>
> Hi Nicolas
>
> My short answer would be another question: what is the aim of your analysis
> ?
>
> IOV is fine to split variability into inter-individual,
> intra-individual-inter-occasion and intra-individual-intra-occasion random
> components. This is excellent for data description, and can bring
> interesting insight into the mechanisms explaining variability. But if you
> want to use your results for prediction, e.g. to devise a TDM program, you
> wont be able to draw relevant information from IOV: a blood sample obtained
> in a patient on a certain occasion wont inform you on your patients
> behavior on another occasion; in this situation, a model merely
> distinguishing inter-individual and intra-individual variability components
> is easier to exploit. Thus, there may be good reasons not to use IOV even
> when it would be possible.
>
> Kind regards
>
> Thierry
>
> Thierry Buclin, MD, PD,
>
> Division of Clinical Pharmacology and Toxicology
>
> University Hospital of Lausanne (CHUV)
>
> Lausanne - SWITZERLAND
>
> tel +41 21 314 42 61
>
> fax +41 21 314 42 66
>
> On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
>
> Dear colleagues,
>
> could someone give me an advice about the rational of using IOV in a
> particular circumstance?
>
> We have data from a clin trial with 3 occasions for each patient. It was a
> chemotherapy and the patients have received up to 7 cures. The issue is that
> the 3 occasions differ from one patient to another.
>
> Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
> cure 2, 5 and 6 or whatever...
>
> It seems to me that combining the 1st occ of all patients is meaningless (as
> for 2nd and 3rd).
> Shall we use as many occasions as cures (7 in our dataset)? In that case,
> how many patients by occ is relevant as a minimum? For certain occ we may
> have few patients. Combining cures is hazardous and has no clinical
> justification.
>
> Best regards
> Nicolas
>
> Pr Nicolas SIMON
> Universite de la Mediterranee (Aix-Marseille II)
>
> --
> James G Wright PhD,
> Scientist, Wright Dose Ltd
> Tel: UK (0)772 5636914
>
> --
> Nick Holford, Professor Clinical Pharmacology
> Dept Pharmacology & Clinical Pharmacology
> University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
> tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
> email: [email protected]
> http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
----- Fin du message de [email protected] -----
--
Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
http://annuaire.univmed.fr/showuser.php?uid=simon.n
Hi Mats
I agree that if a mechanistic quality can be used to determine BOV then I think
this would provide a very strong argument for duration of the occasion. I
believe that this falls into my overarching statement "The duration of the
occasion would need to be indexed to the substantive inferences of the model to
ensure that any influence that BOV has can be assessed in terms of model
predictions.".
My interpretation of nuisance parameter is perhaps slightly different from your
use - I was using a more general sense to indicate a non-ignorable parameter
for which the value was of limited or no interest. As a general rule, although
I am sure there are exceptions, BSV and RUV are non-design specific. The
ability to estimate these parameters accurately and reliably is of course
related to the design. The value of BOV as well as the ability to estimate the
value is design specific and hence I am more inclined to include BOV in the
non-ignorable but (design specific) interest. I would not consider BSV to be
nuisance, RUV is equivocal. I suspect I am splitting hairs at this stage.
Regards
Steve
--
Quoted reply history
From: mats karlsson
[mailto:[email protected]]<mailto:[mailto:[email protected]]>
Sent: Tuesday, 2 November 2010 3:11 p.m.
To: Stephen Duffull; 'Nick Holford';
[email protected]<mailto:[email protected]>
Subject: RE: [NMusers] Rational of using IOV
Hi Steve,
I think we can apply some mechanistic insight into IOV in most cases. For
example for absorption parameters, it is difficult to see that anything but
each dosing occasion would constitute a separate occasion. There may however be
situations where it is difficult to judge properly and IOV may not be modeled
ideally. Then however, not only IOV, but also IIV and RV become "nuisance"
parameters as they will not represent the true IIV or RV.
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Sweden
Postal address: Box 591, 751 24 Uppsala, Sweden
Phone +46 18 4714105
Fax + 46 18 4714003
From: [email protected]<mailto:[email protected]>
[mailto:[email protected]]<mailto:[mailto:[email protected]]>
On Behalf Of Stephen Duffull
Sent: Monday, November 01, 2010 10:49 PM
To: Nick Holford; [email protected]<mailto:[email protected]>
Subject: RE: [NMusers] Rational of using IOV
Nick
While I agree that BOV is not solely a nuisance parameter it is a design
specific parameter and hence can be somewhat of a nuisance. By design specific
we can formulate settings in which the design of the study changes the estimate
of BOV.
To estimate the variance between occasions the duration of the occasion needs
to be defined (a priori). If the occasion is long then the estimate of BOV
will tend to zero since the integral over the occasion to get the average
parameter value will integrate over the random variability. If the occasion is
short then it will tend to a larger positive number. Imagine an occasion of 1
hour versus 1 year. I realise that most tend to use a dose interval as an
occasion but this is also arbitrary as is clinic visits. The duration of the
occasion would need to be indexed to the substantive inferences of the model to
ensure that any influence that BOV has can be assessed in terms of model
predictions.
Given that BOV is design specific then how should this be interpreted in any
given circumstance? Note that being design specific doesn't preclude the
benefit of BOV in its role as an estimable but nuisance parameter (i.e. to
reduce bias in estimates of the population mean parameter values).
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 56 Dunedin
New Zealand
E: [email protected]<mailto:[email protected]>
P: +64 3 479 5044
F: +64 3 479 7034
W: http://pharmacy.otago.ac.nz/profiles/stephenduffull
Design software: http://www.winpopt.com
From: [email protected]<mailto:[email protected]>
[mailto:[email protected]]<mailto:[mailto:[email protected]]>
On Behalf Of Nick Holford
Sent: Tuesday, 2 November 2010 8:33 a.m.
To: [email protected]<mailto:[email protected]>
Subject: Re: [NMusers] Rational of using IOV
Thierry,
Between subject variability (BSV aka IIV) and within subject variability (WSV
aka IOV) describe the limits of what we can identify as sources of variability.
I don't consider this a nuisance -- it is an opportunity for learning. The
random assumption used for estimation of WSV is a convenient way of describing
the size of the problem. If we recognize there is a large element of WSV then
it may stimulate thinking and further investigation to try and understand it.
Ignoring WSV will give a false impression about what can be gained from TCI
(aka TDM). TCI can only hope to remove the BSV part of unpredictable
variability.
See Holford NH. Target concentration intervention: beyond Y2K. Br J Clin
Pharmacol. 1999;48(1):9-13.
Best wishes,
Nick
On 2/11/2010 2:34 a.m., Buclin Thierry wrote:
Dear James,
I always thought that intra-individual variability (IIV) classically
represented the immovable limit on the gains to be expected from TDM - IOV
being indeed used only in a minority of population PK analyses. Both intra- and
inter-occasion variability actually represent nuisance. We agree on the point
that specifying an IOV term in a model will decrease the residual IIV. But
wouldn't this precisely give a falsely favorable impression about potential
gains from a TDM program? Am I wrong to think so?
Kind regards
Thierry
De : James G Wright [mailto:[email protected]]
Envoyé : lundi, 1 novembre 2010 14:04
À : Buclin Thierry
Objet : Re: [NMusers] Rational of using IOV
Dear Thierry,
I hope you are well. I think you are right to highlight the importance of IOV
for TDM, but I would argue it is very important to include it in the model.
This is because IOV places an immovable limit on the gains from TDM. The
classic error is to develop a TDM strategy mistakenly lumping IOV with IIV, and
drastically over-estimating the utility of TDM.
Best regards, James
On 01/11/2010 11:55, Buclin Thierry wrote:
Hi Nicolas
My short answer would be another question: "what is the aim of your analysis ?"
IOV is fine to split variability into inter-individual,
intra-individual-inter-occasion and intra-individual-intra-occasion random
components. This is excellent for data description, and can bring interesting
insight into the mechanisms explaining variability. But if you want to use your
results for prediction, e.g. to devise a TDM program, you won't be able to draw
relevant information from IOV: a blood sample obtained in a patient on a
certain occasion won't inform you on your patient's behavior on another
occasion; in this situation, a model merely distinguishing inter-individual and
intra-individual variability components is easier to exploit. Thus, there may
be good reasons not to use IOV even when it would be possible.
Kind regards
Thierry
Thierry Buclin, MD, PD,
Division of Clinical Pharmacology and Toxicology
University Hospital of Lausanne (CHUV)
Lausanne - SWITZERLAND
tel +41 21 314 42 61
fax +41 21 314 42 66
On 1/11/2010 10:53 a.m., Nicolas SIMON wrote:
Dear colleagues,
could someone give me an advice about the rational of using IOV in a particular
circumstance?
We have data from a clin trial with 3 occasions for each patient. It was a
chemotherapy and the patients have received up to 7 cures. The issue is that
the 3 occasions differ from one patient to another.
Patient X may have be seen on cure 3, 5 and 7 while patient X+1 was seen on
cure 2, 5 and 6 or whatever...
It seems to me that combining the 1st occ of all patients is meaningless (as
for 2nd and 3rd).
Shall we use as many occasions as cures (7 in our dataset)? In that case, how
many patients by occ is relevant as a minimum? For certain occ we may have few
patients. Combining cures is hazardous and has no clinical justification.
Best regards
Nicolas
Pr Nicolas SIMON
Universite de la Mediterranee (Aix-Marseille II)
--
James G Wright PhD,
Scientist, Wright Dose Ltd
Tel: UK (0)772 5636914
--
Nick Holford, Professor Clinical Pharmacology
Dept Pharmacology & Clinical Pharmacology
University of Auckland,85 Park Rd,Private Bag 92019,Auckland,New Zealand
tel:+64(9)923-6730 fax:+64(9)373-7090 mobile:+64(21)46 23 53
email: [email protected]<mailto:[email protected]>
http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford
Dear all,
Just as you thought the discussion was over I feel that I have to point out
that a less "arbitrary" implementation of IOV (BOV) was introduced by Mats
Karlsson at ACoP 2009. The new approach is described in the presentation:
Connecting to the Other Side on slide 40-49
http://2009.go-acop.org/acop2009/program (the ACoP link to the presentation
is currently broken but I have uploaded a copy to my ftp:
http://www.anst.uu.se/marbe710/KarlssonACoP2009.pdf). I have used the
suggested implantation for two projects with very satisfactory results.
For certain PK parameters like CL, IOV typically represents an
auto-correlated within subject variability (WSV) for other typical PK
parameters like KA and lag-time the IOV is typically fairly independent
between doses. The implementation suggested by Mats can describe the WSV in
both these types of parameters (estimates the appropriate shape of the WSV).
In my opinion this is a good way of making IOV less design dependent. Below
is a short coding example if you want to try it out. If you want to make the
code entirely design independent you can either implement the code in $DES
(with T instead of TIME) or use the MTIME function to update TIME more
frequently than at each observation in the dataset [1]. From what I have
heard stochastic differential equations offer an even more sophisticated
solution to WSV in parameters [2].
[1] Petersson KJ, Friberg LE, Karlsson MO. Transforming parts of a
differential equations system to difference equations as a method for
run-time savings in NONMEM. J Pharmacokinet Pharmacodyn. 2010
Oct;37(5):493-506. Epub 2010 Sep 29.
[2] Stochastic differential equations in NONMEM: implementation,
application, and comparison with ordinary differential equations. Tornøe CW,
Overgaard RV, Agersø H, Nielsen HA, Madsen H, Jonsson EN. Pharm Res. 2005
Aug;22(8):1247-58. Epub 2005 Aug 3.
$PK
; ----- Inter Occasion Code ------------------------------------------
OCCL = THETA(7) ; Length between centre of independent occasions
(days)
GAM = THETA(8) ; Shape parameter
SW = OCCL/2
DAY = TIME/24
O1 = ETA(7)/(((OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O2 = ETA(8)/(((2*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O3 = ETA(9)/(((3*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O4 = ETA(10)/(((4*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O5 = ETA(11)/(((5*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O6 = ETA(12)/(((6*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O7 = ETA(13)/(((7*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O8 = ETA(14)/(((8*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O9 = ETA(15)/(((9*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
O10 = ETA(16)/(((10*OCCL-OCCL/2-DAY)**2/SW**2)**GAM+1)
IOV = O1+O2+O3+O4+O5+O6+O7+O8+O9+O10
; --------------------------------------------------------------------
TVCL = THETA(6)
CL = TVCL*EXP(ETA(6))*EXP(IOV)
$THETA (5,10) ; 6_CL
$THETA (10,14) ; 7_OCCL (Lower boundary = Max observed time / #
Eta in IOV code)
; If OCCL goes towards the lower boundary an O11, O12 etc. can be added
and the lower boundary hence lowered.
$THETA (0,1) ; 8_GAM_IOV
$OMEGA 0.4 ; 6_IIV_CL
$OMEGA BLOCK(1) 0.15 ; 7_IOV_CL
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
$OMEGA BLOCK(1) SAME
; --------------------------------------------------------------------
Best regards,
Martin Bergstrand, MSc, PhD student
-----------------------------------------------
Pharmacometrics Research Group,
Department of Pharmaceutical Biosciences,
Uppsala University
-----------------------------------------------
[email protected]
-----------------------------------------------
Work: +46 18 471 4639
Mobile: +46 709 994 396
Seems like the mail below from last week never appeared on the nmusers.
Probably the discussion became too long (or too hair-splitting.).
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Sweden
Postal address: Box 591, 751 24 Uppsala, Sweden
Phone +46 18 4714105
Fax + 46 18 4714003
Quoted reply history
From: mats karlsson [mailto:[email protected]]
Sent: Tuesday, November 02, 2010 11:31 AM
To: 'Stephen Duffull'; 'Nick Holford'; '[email protected]'
Subject: RE: [NMusers] Rational of using IOV
Hi Steve,
When you say BSV is not design-specific, do you with BSV mean the
variability in parameter value between subjects at any given instance, or
the variability in their average parameter value over 1 hour or over 1 year?
With each of these definitions and in the presence of time-varying
parameters, BSV is different. Whenever we have variability in parameters
over time, our ability to capture and distinguish between random effects
will be dependent on our design. If you have IIV, IOV and RV present, all
three will become design-dependent in the sense I think you used the word.
I would say that none of them are design-dependent, but the information in
data and our ability in postulating appropriate models is under many
situations not sufficient to capture all variability adequately. In
particular parameter time-variation is difficult to both capture and model
appropriately.
I have no definition of nuisance parameters, I just tried to echo your use.
With respect to your definition of them, I would say that no parameter has
unlimited interest, but in a model I am interested in, no estimated
parameter has no interest. (and now I'm quite sure I'm splitting hairs.)
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Dept of Pharmaceutical Biosciences
Uppsala University
Sweden
Postal address: Box 591, 751 24 Uppsala, Sweden
Phone +46 18 4714105
Fax + 46 18 4714003
From: Stephen Duffull [mailto:[email protected]]
Sent: Tuesday, November 02, 2010 9:05 AM
To: [email protected]; 'Nick Holford'; [email protected]
Subject: RE: [NMusers] Rational of using IOV
Hi Mats
I agree that if a mechanistic quality can be used to determine BOV then I
think this would provide a very strong argument for duration of the
occasion. I believe that this falls into my overarching statement "The
duration of the occasion would need to be indexed to the substantive
inferences of the model to ensure that any influence that BOV has can be
assessed in terms of model predictions.".
My interpretation of nuisance parameter is perhaps slightly different from
your use - I was using a more general sense to indicate a non-ignorable
parameter for which the value was of limited or no interest. As a general
rule, although I am sure there are exceptions, BSV and RUV are non-design
specific. The ability to estimate these parameters accurately and reliably
is of course related to the design. The value of BOV as well as the ability
to estimate the value is design specific and hence I am more inclined to
include BOV in the non-ignorable but (design specific) interest. I would
not consider BSV to be nuisance, RUV is equivocal. I suspect I am splitting
hairs at this stage.
Regards
Steve
--