Hi everyone;
I'd like to get some feed-back from the group about how to assign
occasions for the estimation of inter-occasional variability. The
situation is this: I have a large Phase 1 dataset (both single dose and
multiple dose) where many samples were taken per subject, and "occasion"
is readily assigned by using period. The maximum number of occasions in
these data is equal to 4. However, I also have a dataset from a Phase 3
study where sparse samples were taken. As you would expect, the amount of
data from patient to patient is quite variable, with some subjects having
as few as 2 samples where others have as many as 8. Also, the times
(relative to dosing) are quite variable. What I would like to do is to
combine these two datasets, and keep the same number of occasions that are
in the rich dataset by grouping these sparse samples by time relative to
the first dose. For example (and this is arbitrary) I could define
Occasion 1 as including any sample taken between 0-48 hours, Occasion 2 as
including any sample between 48-72 hrs, etc. up to Occasion 4.
Does anyone see a problem with this? Or, do you have a better idea?
Many thanks for your time,
Mike Fossler
GSK
Assigning occasions to sparse data
Mike
There is no requirement that the number of occasions is the same for all
individuals in the study or for all studies in your dataset. So I don't see
why you want the number of occasions to be the same for all ID numbers.
On a similar note - I believe that the division of within subject
variability (WSV) into between occasion (BOV) and within occasion (WOV),
although helpful and somewhat intuitive, is also arbitrary. Indeed if you
took an asymptotic example where you made the occasion to equal to (say) 1
year and your study had samples over 2 years then BOV would turn out to be
rather inconsequential since the parameter values would be averaged over
each occasion and the random variability between occasions is likely to be
quite small.
So, if WSV = BOV + WOV then as the occasion duration tends to infinity then
WSV --> WOV and BOV --> 0. In converse, as occasion duration tends to zero
then WSV --> BOV and WOV --> 0 [assuming you can accurately estimate BOV].
The value of BOV is therefore design specific and hence any benefit from
interpretation of BOV can only be gained if the occasion is set a priori to
a clinically or biologically meaningful interval (e.g. 1 dose interval).
[i.e. To be (perhaps inappropriately) provocative, if large estimated BOV is
bad for your drug then all you need to do is make the occasion duration to
be very long :-)]
Just some thoughts.
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 913 Dunedin
New Zealand
E: [EMAIL PROTECTED]
P: +64 3 479 5044
F: +64 3 479 7034
Design software: www.winpopt.com
Quoted reply history
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] On Behalf Of
> [EMAIL PROTECTED]
> Sent: Thursday, 23 August 2007 7:04 a.m.
> To: [email protected]
> Subject: [NMusers] Assigning occasions to sparse data
>
>
>
> Hi everyone;
>
> I'd like to get some feed-back from the group about how to
> assign occasions for the estimation of inter-occasional
> variability. The situation is this: I have a large Phase 1
> dataset (both single dose and multiple dose) where many
> samples were taken per subject, and "occasion" is readily
> assigned by using period. The maximum number of occasions in
> these data is equal to 4. However, I also have a dataset from
> a Phase 3 study where sparse samples were taken. As you would
> expect, the amount of data from patient to patient is quite
> variable, with some subjects having as few as 2 samples where
> others have as many as 8. Also, the times (relative to
> dosing) are quite variable. What I would like to do is to
> combine these two datasets, and keep the same number of
> occasions that are in the rich dataset by grouping these
> sparse samples by time relative to the first dose. For
> example (and this is arbitrary) I could define Occasion 1 as
> including any sample taken between 0-48 hours, Occasion 2 as
> including any sample between 48-72 hrs, etc. up to Occasion 4.
>
> Does anyone see a problem with this? Or, do you have a better idea?
>
> Many thanks for your time,
>
>
> Mike Fossler
> GSK
>
Mike,
I agree with what Steve wrote. There is no need to have the same number of
occasions in each study if you choose to pool them -- but it would make sense
to use the same time period for defining an occasion for all studies.
Perhaps the most widely used time period would be a day because when there are
several samples taken after a particular dose the next occasion this is done
would be on a different day. If you use the occasion length as a day then you
can estimate the day to day variability in say clearance.
One of the practical benefits of estimating BOV comes from decreasing model
misspecification of the residual error (Karlsson & Sheiner 1993). The occasion
length is not really too critical for this kind of application.
The other practical benefit is using the size of BOV to understand if an
intervention such as target concentration intervention would be useful (see
Holford 1999). For this purpose the occasion variability will implicitly
related to day to day variability e.g. if you measure concs on one day it will
be the BOV that will determine how reliable you can be in predicting the concs
on another day based on the first day of measurement.
Karlsson MO, Sheiner LB. The importance of modeling interoccasion variability
in population pharmacokinetic analyses. Journal of Pharmacokinetics &
Biopharmaceutics. 1993; 21(6):735-50.
Holford NHG. Target Concentration Intervention: Beyond Y2K. Br J Clin
Pharmacol. 1999;48:9-13.
Stephen Duffull wrote:
>
> Mike
>
> There is no requirement that the number of occasions is the same for all
> individuals in the study or for all studies in your dataset. So I don't see
> why you want the number of occasions to be the same for all ID numbers.
>
> On a similar note - I believe that the division of within subject
> variability (WSV) into between occasion (BOV) and within occasion (WOV),
> although helpful and somewhat intuitive, is also arbitrary. Indeed if you
> took an asymptotic example where you made the occasion to equal to (say) 1
> year and your study had samples over 2 years then BOV would turn out to be
> rather inconsequential since the parameter values would be averaged over
> each occasion and the random variability between occasions is likely to be
> quite small.
>
> So, if WSV = BOV + WOV then as the occasion duration tends to infinity then
> WSV --> WOV and BOV --> 0. In converse, as occasion duration tends to zero
> then WSV --> BOV and WOV --> 0 [assuming you can accurately estimate BOV].
> The value of BOV is therefore design specific and hence any benefit from
> interpretation of BOV can only be gained if the occasion is set a priori to
> a clinically or biologically meaningful interval (e.g. 1 dose interval).
> [i.e. To be (perhaps inappropriately) provocative, if large estimated BOV is
> bad for your drug then all you need to do is make the occasion duration to
> be very long :-)]
>
> Just some thoughts.
>
> Steve
> --
> Professor Stephen Duffull
> Chair of Clinical Pharmacy
> School of Pharmacy
> University of Otago
> PO Box 913 Dunedin
> New Zealand
> E: [EMAIL PROTECTED]
> P: +64 3 479 5044
> F: +64 3 479 7034
>
> Design software: www.winpopt.com
>
>
> > -----Original Message-----
Quoted reply history
> > From: [EMAIL PROTECTED]
> > [mailto:[EMAIL PROTECTED] On Behalf Of
> > [EMAIL PROTECTED]
> > Sent: Thursday, 23 August 2007 7:04 a.m.
> > To: [email protected]
> > Subject: [NMusers] Assigning occasions to sparse data
> >
> >
> >
> > Hi everyone;
> >
> > I'd like to get some feed-back from the group about how to
> > assign occasions for the estimation of inter-occasional
> > variability. The situation is this: I have a large Phase 1
> > dataset (both single dose and multiple dose) where many
> > samples were taken per subject, and "occasion" is readily
> > assigned by using period. The maximum number of occasions in
> > these data is equal to 4. However, I also have a dataset from
> > a Phase 3 study where sparse samples were taken. As you would
> > expect, the amount of data from patient to patient is quite
> > variable, with some subjects having as few as 2 samples where
> > others have as many as 8. Also, the times (relative to
> > dosing) are quite variable. What I would like to do is to
> > combine these two datasets, and keep the same number of
> > occasions that are in the rich dataset by grouping these
> > sparse samples by time relative to the first dose. For
> > example (and this is arbitrary) I could define Occasion 1 as
> > including any sample taken between 0-48 hours, Occasion 2 as
> > including any sample between 48-72 hrs, etc. up to Occasion 4.
> >
> > Does anyone see a problem with this? Or, do you have a better idea?
> >
> > Many thanks for your time,
> >
> >
> > Mike Fossler
> > GSK
> >
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
[EMAIL PROTECTED] tel:+64(9)373-7599x86730 fax:+64(9)373-7090
www.health.auckland.ac.nz/pharmacology/staff/nholford
Hi Steve;
Yes there is no requirement that the number of occasions be the same in
every study (I have both 2 period and 4 period studies in my Phase 1 dat
set). But, from a practical stand-point, how much is too much? I guess
I'd like to keep things simple, both for practical reasons and also that
I'm not completely convinced that modeling 8 occassions (as opposed to 4)
would really result in that much of an improvement.
Mike
"Stephen Duffull" <[EMAIL PROTECTED]>
22-Aug-2007 16:55
To
[EMAIL PROTECTED], [email protected]
cc
Subject
RE: [NMusers] Assigning occasions to sparse data
Mike
There is no requirement that the number of occasions is the same for all
individuals in the study or for all studies in your dataset. So I don't
see
why you want the number of occasions to be the same for all ID numbers.
On a similar note - I believe that the division of within subject
variability (WSV) into between occasion (BOV) and within occasion (WOV),
although helpful and somewhat intuitive, is also arbitrary. Indeed if you
took an asymptotic example where you made the occasion to equal to (say) 1
year and your study had samples over 2 years then BOV would turn out to be
rather inconsequential since the parameter values would be averaged over
each occasion and the random variability between occasions is likely to be
quite small.
So, if WSV = BOV + WOV then as the occasion duration tends to infinity
then
WSV --> WOV and BOV --> 0. In converse, as occasion duration tends to
zero
then WSV --> BOV and WOV --> 0 [assuming you can accurately estimate BOV].
The value of BOV is therefore design specific and hence any benefit from
interpretation of BOV can only be gained if the occasion is set a priori
to
a clinically or biologically meaningful interval (e.g. 1 dose interval).
[i.e. To be (perhaps inappropriately) provocative, if large estimated BOV
is
bad for your drug then all you need to do is make the occasion duration to
be very long :-)]
Just some thoughts.
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 913 Dunedin
New Zealand
E: [EMAIL PROTECTED]
P: +64 3 479 5044
F: +64 3 479 7034
Design software: www.winpopt.com
Quoted reply history
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] On Behalf Of
> [EMAIL PROTECTED]
> Sent: Thursday, 23 August 2007 7:04 a.m.
> To: [email protected]
> Subject: [NMusers] Assigning occasions to sparse data
>
>
>
> Hi everyone;
>
> I'd like to get some feed-back from the group about how to
> assign occasions for the estimation of inter-occasional
> variability. The situation is this: I have a large Phase 1
> dataset (both single dose and multiple dose) where many
> samples were taken per subject, and "occasion" is readily
> assigned by using period. The maximum number of occasions in
> these data is equal to 4. However, I also have a dataset from
> a Phase 3 study where sparse samples were taken. As you would
> expect, the amount of data from patient to patient is quite
> variable, with some subjects having as few as 2 samples where
> others have as many as 8. Also, the times (relative to
> dosing) are quite variable. What I would like to do is to
> combine these two datasets, and keep the same number of
> occasions that are in the rich dataset by grouping these
> sparse samples by time relative to the first dose. For
> example (and this is arbitrary) I could define Occasion 1 as
> including any sample taken between 0-48 hours, Occasion 2 as
> including any sample between 48-72 hrs, etc. up to Occasion 4.
>
> Does anyone see a problem with this? Or, do you have a better idea?
>
> Many thanks for your time,
>
>
> Mike Fossler
> GSK
>
Hi, Mike,
That doesn't follow from my experience. If the data is good, "contains the
information that would support the tested model feature", then more is
better. If the data is bad, "excessive noise or doesn't contain the
information expected", then it does not matter how much more of data you
add. The way to discriminate, of course, is to test the models.
Bill
_____
Quoted reply history
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of [EMAIL PROTECTED]
Sent: Thursday, August 23, 2007 8:28 AM
To: Stephen Duffull
Cc: [email protected]
Subject: RE: [NMusers] Assigning occasions to sparse data
Hi Steve;
Yes there is no requirement that the number of occasions be the same in
every study (I have both 2 period and 4 period studies in my Phase 1 dat
set). But, from a practical stand-point, how much is too much? I guess I'd
like to keep things simple, both for practical reasons and also that I'm not
completely convinced that modeling 8 occassions (as opposed to 4) would
really result in that much of an improvement.
Mike
"Stephen Duffull" <[EMAIL PROTECTED]>
22-Aug-2007 16:55
To
[EMAIL PROTECTED], [email protected]
cc
Subject
RE: [NMusers] Assigning occasions to sparse data
Mike
There is no requirement that the number of occasions is the same for all
individuals in the study or for all studies in your dataset. So I don't see
why you want the number of occasions to be the same for all ID numbers.
On a similar note - I believe that the division of within subject
variability (WSV) into between occasion (BOV) and within occasion (WOV),
although helpful and somewhat intuitive, is also arbitrary. Indeed if you
took an asymptotic example where you made the occasion to equal to (say) 1
year and your study had samples over 2 years then BOV would turn out to be
rather inconsequential since the parameter values would be averaged over
each occasion and the random variability between occasions is likely to be
quite small.
So, if WSV = BOV + WOV then as the occasion duration tends to infinity then
WSV --> WOV and BOV --> 0. In converse, as occasion duration tends to zero
then WSV --> BOV and WOV --> 0 [assuming you can accurately estimate BOV].
The value of BOV is therefore design specific and hence any benefit from
interpretation of BOV can only be gained if the occasion is set a priori to
a clinically or biologically meaningful interval (e.g. 1 dose interval).
[i.e. To be (perhaps inappropriately) provocative, if large estimated BOV is
bad for your drug then all you need to do is make the occasion duration to
be very long :-)]
Just some thoughts.
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 913 Dunedin
New Zealand
E: [EMAIL PROTECTED]
P: +64 3 479 5044
F: +64 3 479 7034
Design software: www.winpopt.com
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] On Behalf Of
> [EMAIL PROTECTED]
> Sent: Thursday, 23 August 2007 7:04 a.m.
> To: [email protected]
> Subject: [NMusers] Assigning occasions to sparse data
>
>
>
> Hi everyone;
>
> I'd like to get some feed-back from the group about how to
> assign occasions for the estimation of inter-occasional
> variability. The situation is this: I have a large Phase 1
> dataset (both single dose and multiple dose) where many
> samples were taken per subject, and "occasion" is readily
> assigned by using period. The maximum number of occasions in
> these data is equal to 4. However, I also have a dataset from
> a Phase 3 study where sparse samples were taken. As you would
> expect, the amount of data from patient to patient is quite
> variable, with some subjects having as few as 2 samples where
> others have as many as 8. Also, the times (relative to
> dosing) are quite variable. What I would like to do is to
> combine these two datasets, and keep the same number of
> occasions that are in the rich dataset by grouping these
> sparse samples by time relative to the first dose. For
> example (and this is arbitrary) I could define Occasion 1 as
> including any sample taken between 0-48 hours, Occasion 2 as
> including any sample between 48-72 hrs, etc. up to Occasion 4.
>
> Does anyone see a problem with this? Or, do you have a better idea?
>
> Many thanks for your time,
>
>
> Mike Fossler
> GSK
>
Mike,
Estimating BOV from 8 occasions or from 4 occasions will take essentially the
same computation time (if you use BLOCK (*) SAME).
But assuming ETA is the same on two of the occasions (which are collapsed
together from having 4 occasions instead of 8) will necessarily induce model
misspecification on those occasions.
The "how much improvement" will depend on the size of BOV and residual error
and other model related issues. But whether there is an improvement or not
costs essentially nothing. Why add this problem when it can be avoided simply
by using BOV on 8 occasions?
Nick
[EMAIL PROTECTED] wrote:
>
> Hi Steve;
>
> Yes there is no requirement that the number of occasions be the same in every
> study (I have both 2 period and 4 period studies in my Phase 1 dat set). But,
> from a practical stand-point, how much is too much? I guess I'd like to keep
> things simple, both for practical reasons and also that I'm not completely
> convinced that modeling 8 occassions (as opposed to 4) would really result in
> that much of an improvement.
>
> Mike
>
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
[EMAIL PROTECTED] tel:+64(9)373-7599x86730 fax:+64(9)373-7090
www.health.auckland.ac.nz/pharmacology/staff/nholford