A Question regarding the Kaplan Meier survival curve.

[scerab]

Hello guys

I have a set of epidemiological data for a group of patients whom were followed up for a specific time period. The mortality rate was easily calculated, however I am rather confused on how to perform the Kaplan Meier survival curve.
I have both the SPSS (V-17)and the Epi info (V 3.5.1) software, could anyone please tell me which is the simplest way of getting that survival curve?
I am having trouble filling the variables, I would appreciate some insight on this issue as well.

Thanks in advance guys

[bukharin]

You should have your data in this kind of format:

id  time  died
1   27    1
2   84    0

In this case subject 1 was followed for 27 days (or months or whatever) and then died, whereas subject 2 was followed for 84 days and at the end was still alive, ie they were censored at 84 days.

Then it's simply a case of telling your stats package which variable is the time variable and which is the "failure" (death etc) variable. In SPSS you can do this in Analyze -> Survival -> Kaplan-Meier survival analysis. I haven't used Epi info so I can't tell you how to do it there.

[scerab]

thank you very much your advice really helped.
I was able to make the graph with both Epi info and SPSS. one way I know I am doing things right is that both programs gave me the same curve lol.
so I got this curve after entering the data.
how would you interpret this graph?
isn't it supposed to reach 0 survival probability? ( bottom of the y axis)
Thanks in advance
P.S. The graph is on the attachments.

[bukharin]

In terms of interpretation, it tells you what % were alive at any given time of follow-up (eg survival at 2 years was 93%). Survival will only go to zero if you have followed everyone until death; if there are censored observations then it won't go to zero.

One problem with these curves is that the very right-hand part of the curve can be quite misleading (if there are only a few non-censored survivors), so a rule of thumb is to truncate the x axis at the point where you only have 10 survivors, or 10% of the original cohort, whichever is higher.

I'd suggest re-scaling the y axis to go from 0 to 1 - it would be much easier to interpret. You could also relabel the y axis to go from 0% to 100% for the same reason.

[BelaAli]

This really helped
I was wondering if this curve could be applied in a cross sectional survey as I have read it almost everywhere that it is used in trials

[bukharin]

I suppose it depends on how you've constructed your survey - for example if you've studied the same group of people over time, you can create Kaplan-Meier curves - but you have the problem of "interval censoring" in which you lost a patient but you're not sure when.