Can I simulate different scenarios for different covariates?

[WXX155367]

As the title shown, I want to simulate different covariates in different scenarios. I remember there is a code to simulate covariates and I also need to do weight-based simulation. I would like to know if I can add nSud=10,20,30 in different covariates in below code or only to add nSud in "sim" function? Look forward to your reply.

set.seed(42)
rxSetSeed(42)
library(dplyr)
nsub=30

Simulate Weight based on age and gender

AGE<-round(runif(nsub,min=18,max=70))
SEX<-round(runif(nsub,min=0,max=1))
HTm<-round(rnorm(nsub,176.3,0.17sqrt(4482)),digits=1)
HTf<-round(rnorm(nsub,162.2,0.16sqrt(4857)),digits=1)
WTm<-round(exp(3.28+1.92*log(HTm/100))exp(rnorm(nsub,0,0.14)),digits=1)
WTf<-round(exp(3.49+1.45log(HTf/100))*exp(rnorm(nsub,0,0.17)),digits=1)
WT<-ifelse(SEX==1,WTf,WTm)
CRCL<-round(runif(nsub,30,140))

[john-harrold]

I think what you would need to do is create a different set of parameters/covariates for each scenario. There are subtilties in how you would do this depending on whether you're using post hoc estimates or subjects generated from IIV terms. What I've done in the past is say generate 1000 subjects using IIV and then append the covariates to those generated subjects using, for example the random values you have above (do this for each scenario). I think you can then send these 1000 subjects into rxSolve and specify nSub =100 and nStud=10. I believe it will then do 10 studies where it samples 100 subjects from the original 1000 each time (again for each scenario).

@mattfidler please correct me if I'm wrong here.

[WXX155367]

And I also would like to ask if the covariates themselves are correlated? For example, some biomarkers are correlated. How can I add that in the simulation? I assume there is correlation between body weight and BMI.

[WXX155367]

I mean I find there are correlations among covariates, such as BW and BMI, serum total protein (STP) and serum sodium (SS). To more closely reflect data and real-world situations, I feel it is better to create a covariate matrix to simulate.
Can I use the below code to create covariate correlation matrix? And this matrix can input in rxSolve?
rm(list=ls())
library(mvtnorm)
library(knitr)
#dimension
d <- 3
#number of simulation
n <- 1000
#correlation matrix
RHO <- rbind(c(1,0.5,0.75), c(0.5,1,0.25), c(0.75,0.25,1))
print(RHO)

[mattfidler]

rxSolve typically uses covariance matrices, but you cannot directly use these in covariate simulations right now. Rather you use any method you want to generate the covariates and then attach them to the dataset like @john-harrold mentioned.

rxode2::rxRmvn() is what rxode2 uses internally when simulating (based on https://academic.oup.com/jrsssb/article/79/1/125/7041924)

[john-harrold]

The params and events inputs to rxSolve just represent those individual subjects. So if you specify BW, STP, and SS it should either come from actual population data or simulated data that take those correlations into account. If you know the statistical details then you should use those to generate the values.

[WXX155367]

Thanks for answering. I expressed wrongly. I mean I want to create virtual population based on correlation of covariates, and when I want to simulate the model, virtual population, event and covariance matrix, I want to use rxSolve to do the simulation. So I just want to know if I can use rxSolve to do the simulation with virtual population.