Regardless of the censor status, every case in a survival analysis has a survival time, being either the time between diagnosis and event or the time between diagnosis and censor (lost to follow-up, living across the study end date, etc). So, anyway, every case has a survival time. Therefore, I can calculate the mean survival time using the general descriptive statistics. However, I found the result is different from the "mean survival time estimation" obtained from the K-M survival analysis using SPSS. And why? How does the mean survival time estimation in SPSS calculated? I saw the footnote of SPSS saying that "Estimation is limited to the largest survival time if it is censored" but I have no idea what on earth does this sentence mean. Many thanks!
I think this will help you. It is like a weighted average.
http://www-01.ibm.com/support/docview.wss?uid=swg21481019
Best addice is to use median (if it was reached) and not mean survival time.
Different software provides the same medians and different means.
SPSS (Version 16), Stata (Version 10), R (2.8.0), S-Plus 7 calculate of the Kaplan-Meier mean survival as the area under the Kaplan-Meier survival curve up to the longest observation regardless if that is censored observation or failure.
SAS Proc Lifetest computes the area under the curve up to the last failure time. From SAS version 9.1 it can be changed as an option (not default one) and you will obtain the same survival mean as via SPSS.
There are also variants of additional methods how to estimate beyond the
largest study time t_max. Therefore in new version could be implemented something else as the default.
Therefore (due to these inconsistencies) the best advice is not to use mean survival time and use the median survival time only.
Since there seems to be a variation between the computed mean survival time provided in the descriptive statistics and those computed by default when you run the SPSS survival analysis, i would used weighted averages as that would resolve the discrepancies. I currently used JMP Pro 12.2 and that isn't an issue.
The mean survival time is estimated as the area under the survival curve in the interval 0 to tmax (Klein & Moeschberger, 2003).
The median survival is the smallest time at which the survival probability drops to 0.5 (50%) or below. If the survival curve does not drop to 0.5 or below then the median time cannot be computed. The median survival time and its 95% CI is calculated according to Brookmeyer & Crowley, 1982.
So, you can report the one from the general descriptive statistics better.
https://www.medcalc.org/manual/kaplan-meier.php
To calculate the mean survival time it is a must that all persons under follow up develop the event of interest. So, the mean survival time is not interpretable for censored observations. What is calculated by SPSS is the mean follow up of persons. So, those who are censored are given a survival time of (last date of follow up minus date of entering the study). In other words, it is the average of follow up time rather than survival time. The difference in the calculated mean between actuarial life tables and Kaplan Meier is that Kaplan Meier uses exact follow up time while actuarial life tables divide follow up time into intervals. Again remember that the mean for censored data pertains to mean follow up duration rather than mean
survival duration.
The median survival time can be calculated only if 50% of the subjects developed the event of interest, i.e. the duration corresponding to a cumulative survival probability of 0.5.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics