0

I want to create a kaplan meier plot. Most tutorials focus on a binary survival (death/survival), but in my case, I would like to plot the time to an event (1,2,3 times switch in medication). Is this possible?

Here is some example data to illustrate my point.

library (data.table)
data <- data.table( ID                   = c(1,1,2,3,4,5,6,6,7,7,7,8,9,10,11,11,11,12,12,13,14,14,14,15,15,16),
                    episode              = c(1,2,1,1,1,1,1,2,1,2,3,1,1,1,1,2,3,1,2,1,1,2,3,1,2,1), 
                    time_till_next_ep    = c(0,5,0,0,0,0,0,8,0,4,14,0,0,0,0,6,17,0,9,0,0,4,9,0,8,0))

First, we have a variable indicating the participant ID.

Then we have a variable indicating the medication episode. This means if a participant takes a certain type of medication within a certain timeframe. Most participants only take the medication once (episode=1), but several take the medication again a second (episode = 2) or third time (episode = 3).

The time between the episodes is calculated in days in the time_till_next_ep variable. Since there is no time between episodes for the 1st case, there are a lot of zeroes in the data.

I want to create a Kaplan-Meier-Graph: on the x-axis the time until the next episode and on the y-axis the number of episodes. With every new episode beginning for a patient, the graph would go one step down.

I've tried using the survival package to create this, but it does not seem to be working

library(survival)

    S_input <- Surv(time = data[,as.numeric(time_till_next_ep)], event = data[,episode])
    
    km_fit <- survfit(S_input ~ 1, 
                  data = data, 
                  type = "kaplan-meier")
    
    km_fig <- ggsurvplot(km_fit, 
                         data,
                         tables.theme = theme_cleantable())

If I check the input S_input data, it seems to have added questionmarks next to my time data

> S_input 
 [1]  0   5?  0   0   0   0   0   8?  0   4? 14?  0   0   0   0   6? 17?  0   9?  0   0   4?  9?  0   8?  0

And the summary data is not really informative:

> summary(km_fit)
Call: survfit(formula = S_input ~ 1, data = data, type = "kaplan-meier")

10 observations deleted due to missingness 
 time n.risk n.event survival std.err lower 95% CI upper 95% CI
    0     16      16        0     NaN           NA           NA

The message says that 10 observations are deleted due to missingness, these seem to be my 10 switches in medication.

Lastly, I get the following error when plotting: Error in f(...) : Aesthetics can not vary with a ribbon

Am I even using the survival analysis correctly in my case or is this really reserved for binary data?

Inkling
  • 469
  • 1
  • 4
  • 19

1 Answers1

2

As you mentioned, your setting cannot be represented in a 'simple' survival model in which subjects die/exit over time due to a unique cause, because there are recurrent events happening in sequence. I think that your setting is best represented in a multistate model, which allows you to explicitly model transitions between events and account for the time intercurrent between them. Here is an excellent tutorial you can refer to on how to model recurrent events. Currently I cannot provide you with a worked example for your specific case, but some code and data are reported in the supplementary material.

Regarding the Kaplan-Meier itself, I don't think you can plot it without reporting the 'survival share' on one of the axes. And Kaplan-Meier plots are straightforward only in all-cause mortality studies. Multistate models give you access to other tools, such as transition probabilities, to plot the risk of experiencing subsequent events.

Alessandro
  • 46
  • 2