GraphPad Statistics Guide

Determining the median followup time

Determining the median followup time

Previous topic Next topic No expanding text in this topic  

Determining the median followup time

Previous topic Next topic JavaScript is required for expanding text JavaScript is required for the print function Mail us feedback on this topic!  

Survival analysis often deals with experimental designs where different subjects are followed for different durations. How can one quantify the median followup time? Survival analysis (in Prism and other programs) tells you the median survival time. But what about the median time of followup?

Note the distinction between the median survival time and the median time that research subjects were followed (the topic of this page).

Prism presents you with a table of number of subjects at risk over time. One thought is to look at this table and see how long it takes for the number to drop to half the starting value. But there are two reasons why the number-at-risk drops over time: a subject can die or his data can be censored. Looking merely at the number-at-risk table treats those two situations identically. If someone dies, you don't know how long they would have been followed. From the point of view of tracking followup time, the roles of deaths and censoring are sort of reversed.

Schemper and Smith (1) followed that idea to its conclusion and devised a clever method to obtain the median followup time.  Run the data through the Kaplan-Meier analysis again, but with the meaning of the status indicator reversed. The end point is loss-of-followup (which is usually considered censoring). If the patient died, you can't know how long they would have been followed. So death censors the true but unknown observation time of an individual. So create a Kaplan Meier curve where loss of followup is the event being followed, and a death is treated as censoring the data.

In Prism:

1.From the survival analysis results, click New, then Duplicate sheet.

2.OK the dialog that lists the data being analyzed.

3.On the parameters dialog, swap the two indicator variables. The default is for 1 to denote death and zero to denote censoring. Reverse this convention in the dialog (but leave the data alone).Tell Prism that 0 denotes "death" and 1 denotes "censoring".

4.OK from the dialog and look at the results page. Ignore the log rank test and its P value. These values cannot be interpreted. Instead, look at the "median survival". Since you swapped the meaning of survival and censored, this value is really the median followup time.

5.The Kaplan-Meier graph created from this analysis tracks the number of patients being followed over time. It is distinct from the Kaplan-Meier graph that tracks percent survival over time.  

For the sample data comparing two groups, the results (with some polishing) look like this:

1. M Schemper and TL Smith. A note on quantifying follow-up in studies of failure time. Controlled clinical trials (1996) vol. 17 (4) pp. 343-346