Least squares mean in BE

[skyee1]

How to use API to obtain the least squares mean of rbe results

[PharmCat]

Hi!
You can see Metida.jl. You can find it more general.

In ReplicateBE you can get model coefficients with coef or you can get an estimate for your contrast matrix with estimate

Can you provide SAS code for your task?

[skyee1]

Hi!
Thank you for your reply!
Follow is my SAS code for my task:
sas proc mixed data=be; class seq trt per subject ; model lcmax=seq trt per/ddfm=satterth; random trt/sub=subject type=fa0(2); repeated /group=trt sub=subject; lsmeans trt ; ods output Estimates=est; ods output LsMeans=lsm; ods output Tests3=anova; ods output CovParms=res(rename=(estimate=var)); run;

[PharmCat]

proc mixed ...

Actually, I understand what you want. Your code leads to some warnings with output. I make a few changes (check the level of reference ):

proc mixed data=be; 
class seq trt(ref="R") per subject; 
model lcmax=seq trt per/ddfm=satterth solution; 
random trt/sub=subject type=fa0(2); 
repeated /group=trt sub=subject; 
lsmeans trt ; 
ESTIMATE 'diff' trt 1 -1/CL ALPHA=0.1;
ods output Estimates=est; 
ods output LsMeans=lsm; 
ods output Tests3=anova; 
ods output CovParms=res(rename=(estimate=var)); 
run;

So for LsMeans in ReplicateBE at first you should make L matrix (1xp), where p - length of fixed effect vector coef
If you have 2 sequences, 4 periods, 2 treatments (your model is 1 + seq + per + trt) you will have 6 coefficients. and for T and R treatment you will have these matrices:

#1 
[1 0.5 0.25 0.25 0.25 0]
#2 
[1 0.5 0.25 0.25 0.25 1]

then you can get an estimate:

estimate(rbe, [1 0.5 0.25 0.25 0.25 0]; name = "Estimate 1", alpha = 0.05)

There is no automatic creation of L-matrix by the factor in ReplicateBE. But this can be done easily:

Set 1 for Intercept coefficient, set 1 for one level of the factor of interest (or 0 for base level). For other sets 1/n where n - number of levels in a factor. For the example above: 1st - intercept, 2nd - 1/2 (2 sequences), 3-5 - 1/4 (4 periods), 6 - 1 for T 0 for R.

[skyee1]

Thanks for your reply!

This package is very useful for my research. I also recommended it to my colleagues. this tutorial has perfectly solved my problem for your reply.

I have starred this depot. However, I found that it takes a long time to run data for the first time (sample size is 27). , and it may take 80 seconds. But the second run will be very fast.

I would like to know what caused this. I use this package in R through JuliaConnectoR. In fact, it also has this problem when running in julia (it takes a long time).

Hope to get your reply.

My computer configuration is as follows.

• Session info --------------------------------------------------------
  setting value
  version R version 4.0.5 (2021-03-31)
  os Windows 10 x64
  system x86_64, mingw32
  ui RStudio
  language (EN)
  collate Chinese (Simplified)_China.936
  ctype Chinese (Simplified)_China.936
  tz Asia/Taipei
  date 2022-01-05

[PharmCat]

Hi! Glad if this repo helps you! "First time run" is a known feature (problem) of Julia - this time used for compilation. Second run - is real work time. Also other packages can take time for compilation. ReplicateBE using ForwardDiff for REML gradient and hessian - it add little more compile time. In future development ReplicateBE will use Metida for fitting, it can help to make better fitting and validation.
Also you can read this: https://discourse.julialang.org/t/time-to-first-plot-clarification/58534
And if you use julia image with some precompiled modules https://docs.julialang.org/en/v1/devdocs/sysimg/

One more question. Why are you using Julia under R? Is any specific package that is not present in Julia?