3
$\begingroup$

Let's say I have a database with 21 patients and one of these patients has died. In such a scenario, can we apply overall survival analysis using the Kaplan-Meier model or no? Meaning that does it have any statistical interest or it would be just a waste of time ? why / why not ?

$\endgroup$

2 Answers 2

4
$\begingroup$

Maybe "appropriate" but "not very useful."

The Kaplan-Meier estimator will show a single drop in the survival curve, at the time of the death event. It will only use information from the individuals at risk at that particular time. If $Y$ individuals are at risk at that time, the drop will be to a survival fraction of $(Y-1)/Y$, with a variance (by the Greenwood formula) of $(Y-1)/Y^3$. The Kaplan-Meier curve will be flat at that value thereafter.

That has the merit of displaying your data sample in a useful way, but it wouldn't provide much reason to trust your results beyond your particular data sample.

$\endgroup$
1
  • $\begingroup$ thank you so much $\endgroup$
    – An116
    Commented Feb 12 at 10:36
3
$\begingroup$

Just to give a very intuitive perspective: in survival analysis the main unit we care about is the "case", as in the subject who experienced the event in study. Contrarily, when we analyze a continuous outcome we care about all the subjects with a measure of the outcome.

To make a parallel between the two types of analyses, having only one case in survival it's like having only one observation in a continuous outcome setting. You cannot do much statistics on that..

$\endgroup$
1
  • $\begingroup$ thank you so much $\endgroup$
    – An116
    Commented Feb 12 at 10:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.