alternative way to compute or infer median from survival curve I have used the survival package in R to generate survival plots. However for one of my sub groups, the graph does not fall below 0.5 survival. Good news for them....but not for me as I cannot compute median survival time.
I would like to know if there is any way I can compute the median survival time? or infer it somehow. I am convinced this is impossible so this is a bit of a last ditch attempt at finding some wisdom on this topic. 
As a statistician, would you not include this sub-group in your analysis, or would you find an alternative method of computing median survival time?
 A: You cannot calculate the median survival time if less than half of the individuals have an event. You just know that it is larger than the time point of the last event in that specific sample. If this happens because the group is very small (for example it contains only 3 individuals, and 2 are censored, or the follow-up is very short in the group), then the group should not really be by itself and it does not contain a lot of information. If this just happened by chance, then I would definitely not exclude it, and I would report that the median survival time in this group is larger than the last event time. 
It's as if from 100 cars followed for two weeks, 10 break down during this time and the rest of 90 do not, and you would ask: "by what time do 50 cars break down?". This you can not estimate from the data. 
A way around it is to extrapolate with a parametric assumption or from the other groups (the last which you can achieve with a Cox model, with a proportional hazards assumption). 
