How to replicate plot with two panels?

Question

I am trying to replicate this plot, here:

Here is the source of this plot, slide 89:

http://www.drizopoulos.com/courses/Int/JMwithR_CEN-ISBS_2017.pdf

The top of the plot is the hazard function over time, whereas the bottom green curve is the fitted linear mixed effects model over time.

I have been able to plot both of these separately, however, cannot seem to combine them using either par(mfrow=c(2,1)) or the gridExtra package (because only one is a ggplot object).

I am using the aids and aids.id datasets (as a part of the JM package) in R.


# Load packages JM and lattice
library("JM")
library("lattice")
library("ggplot2")

#Fit models
lmeFit.aids <- lme(CD4 ~ obstime + obstime:drug,
    random = ~ obstime | patient, data = aids)

coxFit.aids <- coxph(Surv(Time, death) ~ drug, data = aids.id, x = TRUE)

#Plot longitudinal process
p1<-ggplot(data=aids,aes(x=obstime,y=fitted(lmeFit.aids)))
p1<-p1+geom_smooth(se=FALSE)
p1

#Plot survival process 
library(rms)
p2<-psm(Surv(Time,death)~1,data=aids.id)
survplot(p2,what='hazard')

Thank you!

Without your `data`, there's not much we can suggest without making up our own data that many not adequately represent your structure. The following links have several great suggestions for making a question reproducible, namely in the use of `dput(x)` to provide unambiguous sample data. Thanks! https://stackoverflow.com/q/5963269, [mcve], and https://stackoverflow.com/tags/r/info — r2evans, Jul 19 '21 at 14:29
@r2evans, thanks for your comment. I have updated my question accordingly. I am using the 'aids' and 'aids.id' data as a part of the JM package. — statistics123, Jul 19 '21 at 14:44
(I'm a little surprised `JM` passes CRAN's tests: it does not list any of its imported packages in the `DESCRIPTION`, something I thought was a strict requirement.) — r2evans, Jul 19 '21 at 15:01
I loaded the `nlme` and `survival` packages to get `lme` and `coxph`, respectively, but I cannot produce the `p1` plot, complaining: `x has insufficient unique values to support 10 knots: reduce k`. I can kludge it with `jitter` (https://stackoverflow.com/questions/30208670/r-stat-smooth-groups-x-axis), but that comes nowhere close to either of the plots you have. — r2evans, Jul 19 '21 at 15:07
However, you asked about combining ggplot and base graphics, perhaps my answer helps? — r2evans, Jul 19 '21 at 15:07

score 1 · Accepted Answer · answered Jul 19 '21 at 15:07

1

Up front, patchwork allows you to combine ggplot2 and base graphics into one plot. Adapted from ?wrap_elements,

library(ggplot2)
library(patchwork)
gg <- ggplot(mtcars, aes(mpg, disp)) + geom_point()
gg / wrap_elements(full = ~ plot(mtcars$mpg, mtcars$disp))

answered Jul 19 '21 at 15:07

r2evans

141,215
6
77
149

This worked, thank you! I also came up with another solution, which I posted below. – statistics123 Jul 19 '21 at 16:54

score 1 · Answer 2 · answered Jul 19 '21 at 16:53

1

I was able to extract the values of the hazard at various time points using the survest() function. Then, I was able to plot this using ggplot, meaning I could use grid.arrange().

est<-survest(p2,,what='hazard')
hazard<-data.frame(time=est$time, hazard=est$surv)

answered Jul 19 '21 at 16:53

statistics123

33
5

1

`grid.arrange` works. I find `patchwork` to provide all the same options plus more, in case you're looking for more options. – r2evans Jul 19 '21 at 17:48

How to replicate plot with two panels?

2 Answers2