Incidence across calendar years

[Naa Naadu]

I have a dataset where I want to calculate incidence by age groups across the calender years
The command I used was:

stset censor, failure(event==1) exit(end) enter(start) id(id) scale(365.25) origin(dob)
stsplit agegroup, at(0(10) 80)

preserve
keep if agecat==0
stset censor, failure(event==1) exit(end enter(start) id(id) scale(365.25)
stsplit calyear, at(40(1)60)
gen calendar_yr= calyear+1960
strate calendar_yr, per (100000)
restore


preserve
keep if agecat==10
stset censor, failure(event==1) exit(end) enter(start) id(id) scale(365.25)
stsplit calyear, at(40(1)60)
gen calendar_yr= calyear+1960
strate calendar_yr, per (100000)
restore


I repeated this for the rest of the age groups.

These codes increased the person-years and gave me wrong incidence.

How do I resolve this? I have attached an example dataset


id event censor start end dob
12. 1. 12/12/2011. 01/01/2010 01/01/2020 07/08/1993
13. 1. 12/12/2014. 07/02/2011 02/02/2019. 01/03/2005
14. 0. 12/04/2011. 06/03/2010 03/03/2018. 02/07/2002
15. 1. 12/08/2012. 05/04/2012 06/04/2017. 03/03/1990
16. 1. 12/12/2011. 04/05/2013 07/05/2016. 04/03/2003
17. 0. 12/09/2013. 03/06/2014 08/06/2016. 05/04/1996
18. 1. 12/12/2011. 01/07/2015 09/07/2019. 06/05/2002
19. 0. 12/12/2011. 08/01/2010 11/08/2020 07/06/2001

Thank you very much

[Paul Dickman]

If you present the example data so it can be easily read into Stata (e.g., using -dataex-) then I would be willing to have a look at it.

Also, can you explain the content of the variables censor and end.

Is it possible you've misunderstood the syntax of stset? have a look at the example with the diet data in the help (Example: Single-record-per-subject with censoring). If you gave me these data without any explanation then, based on variable names, I would use

stset end, failure(event==1) enter(start) id(id) scale(365.25) origin(dob)

Consider your first stset

stset censor, failure(event==1) exit(end) enter(start) id(id) scale(365.25) origin(dob)

Observations contribute person-time between start and censor. For some observations in the example data, start is after censor so those observations will not contribute to the analysis.

Since, in the example data, end > censor for all observations the exit(end) option will have no impact.

Also, you should be able to do this (splitting on multiple timescales) without subsetting on age group (i.e., there's no need to preserve/restore).

You're much more likely to get a response if you provide example data that can be easily read into Stata.