Survival Analysis taking into account out-migration

[Stephaney Gyaase]

Hello Everyone,
I am performing a survival analysis on health and demographic surveillance data. The design of the system is that an individual enters into the system by birth or in-migration and exits the system either through death or out-migration. An individual can enter into the system and exit the system multiple times. An example is an individual who enters the system on 1st January 2015 and exits on the 5th September 2015 and then enters the system again on the 1st January 2019 and exits on the 20th September 2019. I applied the stset command in Stata, but I noticed the person years have been over-calculated. For instance, in the above example, I was expecting total person-years of around 352 but it gives me 1720, that is adding the person-years from September 2015 to September 2019 whereas the person was out of the system during those period. Please how do I set my data set to take into account the out-migration period?

Please see below is the the code I used and a sample of the data set after the code

Stata Code : stset enddate,failure(event)id(NewID)origin(startdate) exit(time .)

NewID	startdate	startyr	enddate	endtype	year	_st	_d	_t	_t0
1	1-Jan-15	2012	15-Mar-15	OMG	2015	1	0	73	0
1	1-Jan-19	2015	17-Sep-19	OMG	2019	1	0	1720	73
2	1-Jan-15	2013	15-Mar-15	OMG	2015	1	0	73	0
2	1-Jan-19	2015	17-Sep-19	OMG	2019	1	0	1720	73
3	1-Jan-17	2012	15-Sep-17	OMG	2017	1	0	257	0
3	18-Jun-20	2020	31-Dec-20	NA	2020	1	0	1460	257
4	15-Jun-16	2016	31-Dec-16	NA	2016	1	0	199	0
4	1-Jan-17	2016	31-Dec-17	NA	2017	1	0	564	199
4	1-Jan-19	2016	31-Dec-19	NA	2019	1	0	1294	929

Thank you very much for your help

[Clyde Schechter]

I'm very confused by your data, since you have variables named startyr and year that don't correspond to the year components of startdate or enddate--what are those?

Turning to your question, you need to use the -time0()- option, not the -origin()- option, in your -stset- command in order for Stata to recognize that there are gaps in the observation period. -origin()- refers to the time when the person first becomes at risk for the failure event--and it should have the same value in every observation for that person. The -time0()- option specifies a variable giving the beginning date of the current period of observation. By using that, Stata will see that there is a gap between enddate in one observation and time0() in the next observation.

[Stephaney Gyaase]

Thank you very much for this, it now works very well.