The Kaplan-Meier curve displays time-to-event data. An 'event' is some clearly defined happening that only happens once, such as death, component failure, or similar.
The curve shows the survival rate over time, which starts at 1 (100% at time zero) and falls as time goes on.
In this example, (see it on DataClassroom here) the red arrow shows there is a 50% survival at about 73 weeks.
To display this curve, your dataset needs to be in a standardized format with at least these two columns (variables):
- A numerical column with the Time variable
- One column that is known as Delta which indicates whether a subject experienced the event at their last recorded observation time. A value of 1 indicates the event (often death) occurred, while a 0 indicates that the event did not occur for that test subject during the study and that the test subject has either dropped out of the study or made it through the entire duration of the study without the event occurring.
- One row per event, where Time specifies when the event happened
Then you just need to go to the Advanced -> Survival curve in the left-hand menu:
Display options
You can display the curve optionally with dots for the events, with or without gridlines, and you can also choose to view the data as a table.
Median and confidence interval
Beside the curve, you are also shown the calculated median survival time and 95% confidence intervals for that calculation. For the example above:
The table view also shows these figures for each step-down in survival probability.
Multiple groups
If you have an additional categorical column that groups the events/subjects into categories, you can display several curves at once:
Study drop-outs aka censoring
In such a survival study, there may well be participants or test subjects that leave the study before it ends. To take account of these, you need to add a column that marks whether the row signifies a standard study 'event' like a death, or whether it is a 'censoring' event where the subject left the study.
The convention is that the censoring column be numeric, with a 1 signifying an event, and 0 meaning they left the study at that time. So if no-one left the study, this column should be all-ones.
The term 'censoring' is also just a convention. It does not imply that anything is being deliberately hidden.
You can optionally show these censoring events as small crosses on the curve:
