I'm doing cancer research and I want to model the risk of having a cancer recurrence on various sites in the body, with other sites and death as competing risks.
My variables are:
diagdate ("Date of Diagnosis")
followupdate ("Date of last follow-up on vital status")
recurrencelocation ("Location of first recurrence, 3 sites/values")
deathdate ("Date of death within the follow-up period")
and a number of prognostic variables to be adjusted for in the model.
So what I want is to analyse each recurrence location separately with a recurrence in any of the other 2 locations as a competing risk and also death of any cause during the follow-up period for those that never had a recurrence. If you die before having a recurrence, you can't get a recurrence, thus a competing risk and not a censoring event. If you get a recurrence in another location other than the one analysed you can't have a recurrence in the location being analysed (only the first recurrence is recorded).
My problem is getting the syntax down, and setting the right time frames. I could generate a new supporting variable "Death before recurrence" and set that binary but then I would loose the time-aspect of that variable. I want my subjects to be under risk for as long as they are actually under risk, and not destroy data by removing the time-aspect. I don't know how to set different time-frames under risk for different categories of patients. If you never died or had a recurrence your time under risk is the same as the follow-up period. If you died before you had a recurrence your time under risk is how long you lived after diagnosis. If you had a recurrence your time under risk is how long you where recurrence-free after diagnosis and exit the analysis at that time point, even though you might live longer than the follow-up period.
Am I making any sense? This link http://www.stata.com/features/overview/competing-risks-regression/ pretty much describes my problem but I have problem decipher the different elements of the syntax since I have two types of competing events ('death before recurrence during follow-up' and 'other recurrences').
Managed to solve it myself using this guide: http://www.stata.com/manuals13/ststcrreg.pdf#ststcrreg
If you generate a new variable with any possible of several events, for example var_any_outcome and events can be no_event=0, local_recurrence=1, distant_recurrence=2 and death_before_any_recurrence=3
then the stset command and stcrreg will be (variables in italic):
stset followupdate, failure(var_any_outcome==1) enter(diagdate) origin(diagdate) scale(365.25)
stcrreg variable_1 variable_2 variable_3 variable_etc, compete(var_any_outcome == 2 3)
since Stata will treat 0 and missing as no event and thus censored when lost-to-follow-up, so make sure your no event are coded as 0 and not as 1 or any other number.
Using generate and replace it is easy to make a compound variable with all possible outcomes even if your raw data has outcomes split on several variables.
- Badges
- Science topic
- Similar topics
- Medicine
- Oncology
- Cancer Research