Survival Analysis

[Laur Hous]

last year i took STATA's introduction to survival analysis course, but they did not spend much time on putting data together for the analysis prior to setting the data. I need help with understanding when to set an observation as a failure. It is obvious that a failure occurs when a subject leaves the program (or dies), but what if they are still in the program when the program ends. When i took the course, i thought we were told that those observations should be left in (assigned a 0 for failure), then STATA would use the rest of the data to kind of project how long those ID's would have stayed in. But when looking at a colleague's SAS program, it looked like she censored this data. ie assigned a 1 to those observations so the were considered failures. Which is correct? or is the answer, it depends. ?

[Paul Dickman]

Deciding how to code the outcome requires subject-matter knowledge so this wouldn't be prioritized in a Stata course.

i thought we were told that those observations should be left in (assigned a 0 for failure), then STATA would use the rest of the data to kind of project how long those ID's would have stayed in.

This sounds reasonable.

But when looking at a colleague's SAS program, it looked like she censored this data. ie assigned a 1 to those observations so the were considered failures.

Something is not right here. If a survival time is censored then it has NOT resulted in a failure. If a failure has occurred then the survival time is NOT censored.

Note that when specifying the outcome, in Stata we specify the values of the status variable that represent failure (all other values are considered censored) whereas in SAS (if I recall correctly) one specifies the values of the status variable that represent censoring (and all other values are assumed to be failures).

You will find plenty of literature on the topic of how to define the outcome. Look at the section on types of censoring, assumptions about censored data, and the distinction between a censoring and a competing risk occurring.

[Laur Hous]

thanks, your comment is helpful about SAS stating that one specifies the values o the status variable that represent censoring (and all other values are assumed to be failures). That makes sense when i look at the SAS code of my friend's survival analysis. That is one of the things that has been stumping me.... the different ways in one needs to stat things in SAS and STATA.