Dear all
Assuming the population PK or PD data are log-normally (or normally)
distributed, if you have the mean and standard deviation of a readout at each
timepoint but do not have the actual raw data (assuming all pateints are with
the same dosing regimen, etc), is it possible to establish a well fitted
population PK or PD model? How would one get about doing it?
Your help is very much appreciated
Penny
Question of fitting population PK model using summary statistics of data instead of raw data
Re: FW: Question of fitting population PK model using summary statistics of data instead of raw data
It is likely impossible without strong assumptions. I would first fit the population model (fixed effects only) and then start to simulate with different assumption trying to match observed SD or CV for peaks and troughs. You may need to assume the structure and the magnitude of the error model and the structure of the IIV model (ETAs on CL, or V, or both equal, etc.). You may get some rough idea about the magnitude of the IIV but you may need strong assumptions about the residual and IIV model.
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 9/10/2015 2:06 PM, Penny Zhu wrote:
> Dear Dinko
> Thank you for the suggestion. It seems this NAD approach only uses the mean
> data and does not estimate inter-subject variability using the standard
> deviation data.
>
> My intention is to establish a population PK/PD model with appropriate
> estimation of intersubject variability based on the mean and standard deviation
> data at each timepoint.
>
> A major assumption is that we have good knowledge of the base structure of the
> model (e.g. biexponential), and won't run the risk mistaking 2 mono exponential
> models for a biexponential model
>
> Your help and discussions will be very much appreciated.
>
> Penny
>
> -----Original Message-----
> From: Rekic, Dinko [mailto:[email protected]]
> Sent: Thursday, September 10, 2015 10:41 AM
> To: Zhu, Penny
> Subject: RE: [NMusers] Question of fitting population PK
> model using summary statistics of data instead of raw data
>
> See the link and text below.
>
> http://accp1.org/pharmacometrics/theory_popmeth.htm#npd
>
> Naive averaged data approach (NAD)
>
> A model without BSV and BOV is fitted to the
> mean data from all individuals.
>
> Features
>
> -Specialized software not
> necessary.
>
> Disadvantages
>
> -Does not distinguish between
> BSV and WSV.
>
> -Inappropriate means lead to
> biased parameter estimates.
>
> -May produce model distortion
> i.e., 2 mono exponential equations averaged together can
> yield a biexponential.
>
> -Covariate modeling cannot be
> performed.
>
> Kind regards
> Dinko
> _________________________________
> Dinko Rekić, Ph.D., MSc(Pharm)
> Pharmacometrics reviewer
> Division of Pharmacometrics
> Office of Clinical Pharmacology
> Office of Translational Science
> Center for Drug Evaluation and Research
> U.S. Food and Drug Administration
> 10903 New Hampshire Ave
> Silver Spring, MD 20993
> WO Bldg 51, Rm 3122
> Office phone: (8)240 402-3785
>
> "The contents of this message are mine personally and do not
> necessarily reflect any position of the Government or the
> Food and Drug Administration."
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]]
> On Behalf Of Penny Zhu
> Sent: Thursday, September 10, 2015 9:49 AM
> To: [email protected]
> Subject: [NMusers] Question of fitting population PK model
> using summary statistics of data instead of raw data
>
> Dear all
> Assuming the population PK or PD data are log-normally (or
> normally) distributed, if you have the mean and standard
> deviation of a readout at each timepoint but do not have the
> actual raw data (assuming all pateints are with the same
> dosing regimen, etc), is it possible to establish a
> well fitted population PK or PD model? How would one
> get about doing it?
>
> Your help is very much appreciated
>
> Penny
I would argue that it is impossible. With mean data and SD's at each
time-point, it is impossible to separate between and with-subject variability.
However, a model with "strong assumptions", as Leonid has aptly put it, may
still be very useful. It should not be too difficult to come up with reasonable
BSV estimates, and the residual error estimates could probably be taken from
the assay information, once you have fit the mean curve. You should not expect
too much from such a model, but it may be helpful in simulating some future
studies.
Sent from my iPhone
Quoted reply history
> On Sep 10, 2015, at 6:00 PM, Leonid Gibiansky <[email protected]>
> wrote:
>
> It is likely impossible without strong assumptions. I would first fit the
> population model (fixed effects only) and then start to simulate with
> different assumption trying to match observed SD or CV for peaks and troughs.
> You may need to assume the structure and the magnitude of the error model and
> the structure of the IIV model (ETAs on CL, or V, or both equal, etc.). You
> may get some rough idea about the magnitude of the IIV but you may need
> strong assumptions about the residual and IIV model.
> Leonid
>
>
> --------------------------------------
> Leonid Gibiansky, Ph.D.
> President, QuantPharm LLC
> web: www.quantpharm.com
> e-mail: LGibiansky at quantpharm.com
> tel: (301) 767 5566
>
>
>
>> On 9/10/2015 2:06 PM, Penny Zhu wrote:
>> Dear Dinko
>> Thank you for the suggestion. It seems this NAD approach only uses the mean
>> data and does not estimate inter-subject variability using the standard
>> deviation data.
>>
>> My intention is to establish a population PK/PD model with appropriate
>> estimation of intersubject variability based on the mean and standard
>> deviation data at each timepoint.
>>
>> A major assumption is that we have good knowledge of the base structure of
>> the model (e.g. biexponential), and won't run the risk mistaking 2 mono
>> exponential models for a biexponential model
>>
>> Your help and discussions will be very much appreciated.
>>
>> Penny
>>
>>
>> -----Original Message-----
>> From: Rekic, Dinko [mailto:[email protected]]
>> Sent: Thursday, September 10, 2015 10:41 AM
>> To: Zhu, Penny
>> Subject: RE: [NMusers] Question of fitting population PK
>> model using summary statistics of data instead of raw data
>>
>> See the link and text below.
>>
>> http://accp1.org/pharmacometrics/theory_popmeth.htm#npd
>>
>>
>> Naive averaged data approach (NAD)
>>
>> A model without BSV and BOV is fitted to the
>> mean data from all individuals.
>>
>> Features
>>
>> -Specialized software not
>> necessary.
>>
>> Disadvantages
>>
>> -Does not distinguish between
>> BSV and WSV.
>>
>> -Inappropriate means lead to
>> biased parameter estimates.
>>
>> -May produce model distortion
>> i.e., 2 mono exponential equations averaged together can
>> yield a biexponential.
>>
>> -Covariate modeling cannot be
>> performed.
>>
>> Kind regards
>> Dinko
>> _________________________________
>> Dinko Rekić, Ph.D., MSc(Pharm)
>> Pharmacometrics reviewer
>> Division of Pharmacometrics
>> Office of Clinical Pharmacology
>> Office of Translational Science
>> Center for Drug Evaluation and Research
>> U.S. Food and Drug Administration
>> 10903 New Hampshire Ave
>> Silver Spring, MD 20993
>> WO Bldg 51, Rm 3122
>> Office phone: (8)240 402-3785
>>
>> "The contents of this message are mine personally and do not
>> necessarily reflect any position of the Government or the
>> Food and Drug Administration."
>>
>> -----Original Message-----
>> From: [email protected]
>> [mailto:[email protected]]
>> On Behalf Of Penny Zhu
>> Sent: Thursday, September 10, 2015 9:49 AM
>> To: [email protected]
>> Subject: [NMusers] Question of fitting population PK model
>> using summary statistics of data instead of raw data
>>
>> Dear all
>> Assuming the population PK or PD data are log-normally (or
>> normally) distributed, if you have the mean and standard
>> deviation of a readout at each timepoint but do not have the
>> actual raw data (assuming all pateints are with the same
>> dosing regimen, etc), is it possible to establish a
>> well fitted population PK or PD model? How would one
>> get about doing it?
>>
>> Your help is very much appreciated
>>
>> Penny
>>
________________________________
Notice: This e-mail message, together with any attachments, contains
information of Trevena, Inc., 1018 West 8th Avenue, King of Prussia, PA 19406,
USA. This information may be confidential, proprietary, copyrighted and/or
legally privileged.
It is intended solely for use by the individual or entity named on this
message. If you are not the intended recipient, and have received this message
in error, please notify us immediately and delete it and any attachments from
your system.
FW: FW: Question of fitting population PK model using summary statistics of data instead of raw data
Dear Dr Gibiansky
Thank you very much for the suggestion. I largely agree with you that it seems
to be an trial and error thing to make the variability match the model
prediction if we have a strong assumption about the model structure.
I was also wondering whether it is possible to simulate individual patient data
at each timepoint based on the mean, steandard deviation, and an using an
assumption that within patients (especially in adjacent timepionts) the Pk
concentrations are more correlated compared to between patients. Then use
these simulated data to fit the population PK model.
Best regards.
Penny Zhu
Quoted reply history
> -----Original Message-----
> From: Leonid Gibiansky [mailto:[email protected]]
>
> Sent: Thursday, September 10, 2015 5:10 PM
> To: Zhu, Penny; [email protected]
> Subject: Re: FW: [NMusers] Question of fitting population PK
> model using summary statistics of data instead of raw data
>
> It is likely impossible without strong assumptions. I would
> first fit the population model (fixed effects only) and then
> start to simulate with different assumption trying to match
> observed SD or CV for peaks and troughs. You may need to
> assume the structure and the magnitude of the error model
> and the structure of the IIV model (ETAs on CL, or V, or
> both equal, etc.). You may get some rough idea about the
> magnitude of the IIV but you may need strong assumptions
> about the residual and IIV model.
> Leonid
>
>
> --------------------------------------
> Leonid Gibiansky, Ph.D.
> President, QuantPharm LLC
> web: www.quantpharm.com
> e-mail: LGibiansky at quantpharm.com
> tel: (301) 767 5566
>
>
>
> On 9/10/2015 2:06 PM, Penny Zhu wrote:
> > Dear Dinko
> > Thank you for the suggestion. It seems this NAD
> approach only uses the mean data and does not estimate
> inter-subject variability using the standard deviation
> data.
> >
> > My intention is to establish a population PK/PD model
> with appropriate estimation of intersubject variability
> based on the mean and standard deviation data at each
> timepoint.
> >
> > A major assumption is that we have good knowledge of
> the base
> > structure of the model (e.g. biexponential), and won't
> run the risk
> > mistaking 2 mono exponential models for a biexponential
> model
> >
> > Your help and discussions will be very much
> appreciated.
> >
> > Penny
> >
> >
> > -----Original Message-----
> > From: Rekic, Dinko [mailto:[email protected]]
> > Sent: Thursday, September 10, 2015
> 10:41 AM
> > To: Zhu, Penny
> > Subject: RE: [NMusers] Question of
> fitting population PK
> > model using summary statistics of data
> instead of raw data
> >
> > See the link and text below.
> >
> > http://accp1.org/pharmacometrics/theory_popmeth.htm#npd
> >
> >
> > Naive averaged data approach (NAD)
> >
> > A model without BSV and
> BOV is fitted to the
> > mean data from all individuals.
> >
> > Features
> >
> >
> -Specialized software not
> > necessary.
> >
> > Disadvantages
> >
> > -Does not
> distinguish between
> > BSV and WSV.
> >
> >
> -Inappropriate means lead to
> > biased parameter estimates.
> >
> > -May
> produce model distortion
> > i.e., 2 mono exponential equations
> averaged together can
> > yield a biexponential.
> >
> > -Covariate
> modeling cannot be
> > performed.
> >
> > Kind regards
> > Dinko
> > _________________________________
> > Dinko Rekić, Ph.D., MSc(Pharm)
> > Pharmacometrics reviewer
> > Division of Pharmacometrics
> > Office of Clinical Pharmacology
> > Office of Translational Science
> > Center for Drug Evaluation and
> Research
> > U.S. Food and Drug Administration
> > 10903 New Hampshire Ave
> > Silver Spring, MD 20993
> > WO Bldg 51, Rm 3122
> > Office phone: (8)240 402-3785
> >
> > "The contents of this message are mine
> personally and do not
> > necessarily reflect any position of
> the Government or the
> > Food and Drug Administration."
> >
> > -----Original Message-----
> > From: [email protected]
> > [mailto:[email protected]]
> > On Behalf Of Penny Zhu
> > Sent: Thursday, September 10, 2015
> 9:49 AM
> > To: [email protected]
> > Subject: [NMusers] Question of fitting
> population PK model
> > using summary statistics of data
> instead of raw data
> >
> > Dear all
> > Assuming the population PK or PD data
> are log-normally (or
> > normally) distributed, if you have the
> mean and standard
> > deviation of a readout at each
> timepoint but do not have the
> > actual raw data (assuming all pateints
> are with the same
> > dosing regimen, etc), is it
> possible to establish a
> > well fitted population PK or PD
> model? How would one
> > get about doing it?
> >
> > Your help is very much appreciated
> >
> > Penny
> >
>
I think that the procedure that you suggested (simulation - correlation - popPK model) may not be reliable in general case as it assumes (or will result in the data) that the profiles of individual subjects are ordered at all time points. It correspond to some special (but explicit) assumptions about random effect structure. I would rather use explicit assumptions.
Leonid
--------------------------------------
Leonid Gibiansky, Ph.D.
President, QuantPharm LLC
web: www.quantpharm.com
e-mail: LGibiansky at quantpharm.com
tel: (301) 767 5566
Quoted reply history
On 9/11/2015 1:38 PM, Penny Zhu wrote:
> Dear Dr Gibiansky
> Thank you very much for the suggestion. I largely agree with you that it seems
> to be an trial and error thing to make the variability match the model
> prediction if we have a strong assumption about the model structure.
>
> I was also wondering whether it is possible to simulate individual patient data
> at each timepoint based on the mean, steandard deviation, and an using an
> assumption that within patients (especially in adjacent timepionts) the Pk
> concentrations are more correlated compared to between patients. Then use
> these simulated data to fit the population PK model.
>
> Best regards.
>
> Penny Zhu
>
> > -----Original Message-----
> > From: Leonid Gibiansky [mailto:[email protected]]
> >
> > Sent: Thursday, September 10, 2015 5:10 PM
> > To: Zhu, Penny; [email protected]
> > Subject: Re: FW: [NMusers] Question of fitting population PK
> > model using summary statistics of data instead of raw data
> >
> > It is likely impossible without strong assumptions. I would
> > first fit the population model (fixed effects only) and then
> > start to simulate with different assumption trying to match
> > observed SD or CV for peaks and troughs. You may need to
> > assume the structure and the magnitude of the error model
> > and the structure of the IIV model (ETAs on CL, or V, or
> > both equal, etc.). You may get some rough idea about the
> > magnitude of the IIV but you may need strong assumptions
> > about the residual and IIV model.
> > Leonid
> >
> > --------------------------------------
> > Leonid Gibiansky, Ph.D.
> > President, QuantPharm LLC
> > web: www.quantpharm.com
> > e-mail: LGibiansky at quantpharm.com
> > tel: (301) 767 5566
> >
> > On 9/10/2015 2:06 PM, Penny Zhu wrote:
> >
> > > Dear Dinko
> > > Thank you for the suggestion. It seems this NAD
> >
> > approach only uses the mean data and does not estimate
> > inter-subject variability using the standard deviation
> > data.
> >
> > > My intention is to establish a population PK/PD model
> >
> > with appropriate estimation of intersubject variability
> > based on the mean and standard deviation data at each
> > timepoint.
> >
> > > A major assumption is that we have good knowledge of
> >
> > the base
> >
> > > structure of the model (e.g. biexponential), and won't
> >
> > run the risk
> >
> > > mistaking 2 mono exponential models for a biexponential
> >
> > model
> >
> > > Your help and discussions will be very much
> >
> > appreciated.
> >
> > > Penny
> > >
> > > -----Original Message-----
> > > From: Rekic, Dinko [mailto:[email protected]]
> > > Sent: Thursday, September 10, 2015
> >
> > 10:41 AM
> >
> > > To: Zhu, Penny
> > > Subject: RE: [NMusers] Question of
> >
> > fitting population PK
> >
> > > model using summary statistics of data
> >
> > instead of raw data
> >
> > > See the link and text below.
> > >
> > > http://accp1.org/pharmacometrics/theory_popmeth.htm#npd
> > >
> > > Naive averaged data approach (NAD)
> > >
> > > A model without BSV and
> >
> > BOV is fitted to the
> >
> > > mean data from all individuals.
> > >
> > > Features
> >
> > -Specialized software not
> >
> > > necessary.
> > >
> > > Disadvantages
> > >
> > > -Does not
> >
> > distinguish between
> >
> > > BSV and WSV.
> >
> > -Inappropriate means lead to
> >
> > > biased parameter estimates.
> > >
> > > -May
> >
> > produce model distortion
> >
> > > i.e., 2 mono exponential equations
> >
> > averaged together can
> >
> > > yield a biexponential.
> > >
> > > -Covariate
> >
> > modeling cannot be
> >
> > > performed.
> > >
> > > Kind regards
> > > Dinko
> > > _________________________________
> > > Dinko Rekić, Ph.D., MSc(Pharm)
> > > Pharmacometrics reviewer
> > > Division of Pharmacometrics
> > > Office of Clinical Pharmacology
> > > Office of Translational Science
> > > Center for Drug Evaluation and
> >
> > Research
> >
> > > U.S. Food and Drug Administration
> > > 10903 New Hampshire Ave
> > > Silver Spring, MD 20993
> > > WO Bldg 51, Rm 3122
> > > Office phone: (8)240 402-3785
> > >
> > > "The contents of this message are mine
> >
> > personally and do not
> >
> > > necessarily reflect any position of
> >
> > the Government or the
> >
> > > Food and Drug Administration."
> > >
> > > -----Original Message-----
> > > From: [email protected]
> > > [mailto:[email protected]]
> > > On Behalf Of Penny Zhu
> > > Sent: Thursday, September 10, 2015
> >
> > 9:49 AM
> >
> > > To: [email protected]
> > > Subject: [NMusers] Question of fitting
> >
> > population PK model
> >
> > > using summary statistics of data
> >
> > instead of raw data
> >
> > > Dear all
> > > Assuming the population PK or PD data
> >
> > are log-normally (or
> >
> > > normally) distributed, if you have the
> >
> > mean and standard
> >
> > > deviation of a readout at each
> >
> > timepoint but do not have the
> >
> > > actual raw data (assuming all pateints
> >
> > are with the same
> >
> > > dosing regimen, etc), is it
> >
> > possible to establish a
> >
> > > well fitted population PK or PD
> >
> > model? How would one
> >
> > > get about doing it?
> > >
> > > Your help is very much appreciated
> > >
> > > Penny