ggforest error - undefined columns selected

Question

I am trying to make a forrest plot for my model with ggforest(). Here is the code to create mock data to reproduce the problem. Data is formatted according to Therneau for time dependent covariates. I guess this might be the reason why ggforest does not operate properly.

library(survival)
library(survminer)

set.seed(1)
repetitions<-floor(sample(rnorm(1:10, 10)))
id<-rep(1:10, times=repetitions )
age<-rep(floor(sample(18:80,10)),times=repetitions)
diabetes<-rep(sample(0:1,10,replace=TRUE), times=repetitions)
bil<-sample(4:60,length(id), replace=TRUE)


status<-rep(1,length(id))
indices<-vector(length=10)
for(i in 1:10){
       indices[i]<-sum(repetitions[1:i])
   }
status[indices]<-2

daystart <- vector()
a<-vector()
for(i in 1:10){
  if(i==1){ daystart<-1:indices[i]
  } else {a<-1:(indices[i]-indices[i-1])
  }
  daystart<-c(daystart,a)
}

dayend<-daystart+1

mock_data<-cbind.data.frame(id,age,diabetes, bil, daystart, dayend, status)
mock_data$agegroup<-cut(mock_data$age, 2)



fit2<-coxph(Surv(daystart,dayend, status)~bil+diabetes+strata(agegroup), data=mock_data)
ggforest(fit2 , data=mock_data)

I get

Error in [.data.frame(data, ,var ) : undefined columns selected.

I tried installing previous version of package broom ( version 0.5.6) as, as suggested in previous threads, but it didnt resolve the issue. R versions 3.6.1 and 4.1.1 were used. Any ideas?

EDIT: So, the ggforest() gets confused with +strata(). Removing +strata() produces a plot.

If you solved the problem, consider add the solution as an answer to help other users identify the solution. — Daniels118, Oct 14 '21 at 14:59
I added the solution, which I hope is appropriate. I am not too confident in my understanding of all the methods used here. — Leonardo Kelava, Oct 22 '21 at 11:07

score 0 · Answer 1 · answered Oct 22 '21 at 11:03

So, the problem was in this row in ggforest() function.

terms <- attr(model$terms, "dataClasses")[-1]

I just did the quick fix and copy-pasted the body of the function in the new function I created, adding index "-4", in order not to add strata attributes to terms.

I guess the original function might be changed to accomodate this and exclude strata from terms, but I should stress that I am not great at math or statistics, so I am not 100 % sure if stratifying the data for time-varying cox proportional hazards analysis is valid if I stratify by continuous variable such as age. That would end up with each strata containing the data for only several individuals with the same age, each having repeated measurements values.

ggforest error - undefined columns selected

1 Answers1