Survival analysis is often used in tumor clinical trials, and there are usually two estimations that appear in the report: the median survival time and the median follow-up time.
I was also confused about the distinction between these two estimations when I was new to the drug trials and requested to calculate the follow-up time.
For the median survival time, I suppose many people know how to deal with that through the Kaplan-Meier curve. But how about the median follow-up time? As we know, if we want to calculate the follow-up time, we can not guarantee that all subjects are ongoing. Thus we should think of a reasonable way to deal with the completed subjects, otherwise if we directly calculate the median value, the result will result in an underestimate.
Refer to Schemper and Smith, a very clear way to use the reverse Kaplan-Meier curve to calculate the median follow-up time. Given a tumor trial in which the event of interest is actually the loss-of-followup, it's easy to understand that we can not know how long they would have been followed if that event didn't happen. To make this calculation more analytical, Schemper and Smith suggest using the Kaplan-Meier curve with the reversed status indicator (if you use R, it can be seen that
1 to indicate the subject who is censored, and
0 to indicate the event.), where the median survival time can actually be interpreted as the median follow-up time.
Suppose we have a set of OS survival data where 0 indicates the death status. In R that can be used as below.
library(survival) fit <- survfit(Surv(time, status==0) ~ 1, data = os_data) surv_median(fit)
- M Schemper and TL Smith. A note on quantifying follow-up in studies of failure time. Controlled clinical trials (1996) vol. 17 (4) pp. 343-346
- Determining the median followup time