[R-sig-ME] A model for repeated treatments and repeated outcomes

Fri Feb 12 20:16:01 CET 2016

Thanks a lot for the advice Peter.

Just to make sure I got it correctly. Will the model would be this multivariate model:

gfrs_post_op ~ gfr_pre_op1 + gfr_pre_op2 + gfr_pre_op3 + SamplePeriod + (SamplePeriod | Patient)

where gfrs_post-op is a matrix of patient x gfr_post_op

Thanks a lot,
Dan

--------------------------------------------
On Fri, 2/12/16, Peter Claussen <dakotajudo at mac.com> wrote:

 Subject: Re: [R-sig-ME] A model for repeated treatments and repeated outcomes
 To: "Daniel Rubi" <daniel_rubi at ymail.com>
 Cc: "r-sig-mixed-models at r-project.org" <r-sig-mixed-models at r-project.org>
 Date: Friday, February 12, 2016, 12:43 PM

 Dan,

 If I may jump in.

 If I read this correctly, you have two random
 samples per patient - a random sample of pre-operation GRF
 and a random sample post-operation. I think these should be
 considered random since you state the points are not evenly
 spaced in either case.

 So
 you add a factor for sampling period (SamplePeriod =
 c(“Pre”,”Post”)) and model random effect as
 (SamplePeriod | Patient).

 That’s my first thought - I’m not sure
 given three points per sample period you can reasonably
 measure trends.

 Peter

 > On Feb 11, 2016, at 7:01
 PM, Daniel Rubi via R-sig-mixed-models <r-sig-mixed-models at r-project.org>
 wrote:
 > 
 > Dear
 Thierry,
 > 
 > Thanks a
 lot for the help.
 > 
 >
 Please allow me to explain better.
 > 
 > What I'm trying to estimate is whether
 the trend in pre-operation GFR affects the levels of
 post-operation GFR.
 > The GFR is a
 measurement that assesses kidney function based on a blood
 sample and for that reason it reflects the functions of both
 kidneys.
 > The operation is either
 partial or total removal of one kidney - that is affected.
 Therefore it is supposed to prevent GFR from escalating.
 > 
 > What I'm
 interesting in estimating is whether the trend of
 pre-operation GFR (along the time pre-operation time points)
 is a good predictor of the post-operatoin GFR. It's not
 clear at which post-operation time point it is correct to
 measure GFR and therefore I have 3 time points.
 > 
 > The trivial thing to
 do is to estimate the slope of pre-operation GFR and
 individually regress each of the post-operation GFRs against
 thos.
 > 
 > But I'm
 wondering whether there's anything better?
 > 
 > Thanks a lot,
 > Dan
 > 
 > 
 >
 --------------------------------------------
 > On Wed, 2/10/16, Thierry Onkelinx <thierry.onkelinx at inbo.be>
 wrote:
 > 
 > Subject:
 Re: [R-sig-ME] A model for repeated treatments and repeated
 outcomes
 > To: "Daniel Rubi"
 <daniel_rubi at ymail.com>
 > Cc: "r-sig-mixed-models at r-project.org"
 <r-sig-mixed-models at r-project.org>
 > Date: Wednesday, February 10, 2016, 3:54
 AM
 > 
 > Dear
 > Dan,
 > You'll need
 to
 > provide more information. What is
 the global pattern that
 > you expect
 (linear? quadratic? non-linear?) How to you thing
 > that the operation can effect the GFR? You
 need to answer
 > those kind of questions
 so that you can make a sensible
 > fixed
 effects part of the model.
 > The random
 effect is probably just
 > ~1|Patient. And
 a corCAR1(form = ~ Time) can handle the
 >
 temporal correlation within the patient.
 > Best regards,
 > ir.
 Thierry Onkelinx
 > Instituut voor natuur-
 en bosonderzoek /
 > Research Institute
 for Nature and Forest 
 > team Biometrie
 & Kwaliteitszorg / team
 > Biometrics
 & Quality Assurance 
 > Kliniekstraat
 25
 > 1070
 >
 Anderlecht
 > Belgium
 >

 > To call in the statistician after
 the
 > experiment is done may be no more
 than asking him to perform
 > a
 post-mortem examination: he may be able to say what the
 > experiment died of. ~ Sir Ronald Aylmer
 Fisher
 > The plural of anecdote is not
 data. ~ Roger
 > Brinner 
 > The combination of some data and an
 > aching desire for an answer does not
 ensure that a
 > reasonable answer can be
 extracted from a given body of
 > data. ~
 John Tukey
 > 
 >
 2016-02-09 0:55 GMT+01:00
 > Daniel Rubi
 via R-sig-mixed-models <r-sig-mixed-models at r-project.org>:
 >  I have the following experimental
 > design:Measurements of kidney function (in
 units called GFR)
 > taken at several time
 points pre-operation (time points not
 >
 evenly spaced) and at several time points post-operation
 > (neither evenly spaced).
 > 
 > Here's an
 example of my data in R code:
 > 
 > set.seed(1)df <- data.frame(patient =
 letters[1:10],   
 >       
      gfr_ten_days_prop = rnorm(10,5,1),
 > gfr_five_days_prop = rnorm(10,10,1),
 gfr_three_days_prop =
 > rnorm(10,12,1), 
               gfr_one_day_postop =
 > rnorm(10,10,1), gfr_one_day_postop =
 rnorm(10,5,1),
 > gfr_one_day_postop =
 rnorm(10,2,1))
 > 
 >

 > 
 > I'm looking
 for a model which will estimate the effect
 > of pre-operation GFR on post-operation
 GFR, taking into
 > account the different
 times at which GFRs were measured pre-
 >
 and post-operation.One additional possible caveat - my
 data
 > contain missing values (NAs).
 > 
 > I'm having a
 hard time seeing how a mixed-effects model
 > fits this problem since in all the
 examples of repeated
 >
 measures/longitudinal data I came across in each time
 point
 > the response is measured whereas
 here it is more a
 > predictive question -
 how strong does each pre-operation GFR
 >
 predict pos-operation GFR, where the time at which GFRs
 were
 > measured may matter.
 > 
 > Thanks a lot,Dan
 > 
 >     
    [[alternative HTML version deleted]]
 > 
 > 
 > 
 >
 _______________________________________________
 > 
 > R-sig-mixed-models at r-project.org
 > mailing list
 > 
 > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
 >

 >
 _______________________________________________
 > R-sig-mixed-models at r-project.org
 mailing list
 > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models