KNOWLEDGEBASE - ARTICLE #1800

Bug: Median survival can be incorrect when a curve ends at 50% survival.

Definition of median survival

Median survival is the time at which a survival curve crosses the 50% survival point. This is somewhat ambiguous in two cases shown in the following example.

The blue curve never gets down to a survival of 50% so the median survial is simply not defined. You could say that the median survival is greater than 150 days, the last time point at which there are data. 

The red curve hits a survival of 50% at 50 days (that both values are 50 is just a coincidence) and remains there until 127 days.  What is the median survival? It isn't really defined. Prism follows a convention to average those two values (50 and 127) so reports that the median survival is 88.5 days. Since this curve really does plot survival ("survival" curves can be used for any one-time event), one could argue that it is more meaningful to say that 50% of the patients survive until day 127. But it seems to be standard to report the average of the first and last time where survival is 50%. 

The bug

Prism, up to 5.04 and 5.0d, reports the incorrect value for median survival when one data set ends with a horizontal line at Y=50%, and another data set has data beyond that point. Conisder this example:

The survival for the control group drops to 50% on day 1 and the survival curve ends on day 2. The median survival ought to be reported, therefore, as 1.5 days. Instead, Prism averages 1 (the day the survival went down to 50%) and 6 (the end of the curve for all data sets) so reports a median survival of 3.5 days.

It is easy to bypass the bug by manual calculations. The bug only occurs when one survival curve ends with a horizontal line at Y=50% and another survival curve in the same table extends to longer time points. 

Fixed in Prism 6.

Prism file for this example.

Explore the Knowledgebase

Analyze, graph and present your scientific work easily with GraphPad Prism. No coding required.