Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Poisson regression of SIR in aggregated data with external standard

    I want to analyse differences in incidence of appendicitis between first and second generation immigrants and also compare with the native population. The incidence of appendicitis is strongly related to age, differs between the sexes and there are also secular trends. To overcome some of these differences I have use stsplit by age and year and merged external national reference incidence for each sex, age and year strata. I can thus estimate Standardised Incidence Ratio which describe the differences in the incidence rate between the groups I study and the national incidence rate. I know want to compare the SIR between the 1st and 2nd generation immigrants, I guess it will be the ratio of the SIRs. I assume this can be done with poisson regresion, but can not figure out how, and also describe what I have done (what is the ratio of two SIRs?)

    This is my do.file for perforated appendicitis

    stset doexit, failure(perorated) enter(time doenter) exit(time doexit) origin(time dobirth) scale(365.25) id(lopnr)
    stsplit ageband, at(0(5)100)
    stsplit yearband, after(time=d(1/1/1900)) at(0 65(5)110)
    replace yearband = yearband + 1900

    sort kon ageband yearband

    merge kon ageband yearband using D:...........incperforated.dta
    gen pyrs= _t - _t0
    gen E = pyrs*incperforated

    strate countryoforigin1stgen, smr(incperforated) per(100000)
    stsum, by(countryoforigin1stgen)

    I have done these analyses for 1st and 2nd generation of immigrants from different country of origin. This gives the difference in incidence between these groups and the nation as a whole. I know want to analyse and describe the difference between the 1st and 2nd generation immigrants. I could do it direvtly without going through the sn´tandardisation, but would prefer using the ratio of the standardised SIR for clarity.

    I have 11.000.000 posts in the file so these analyses takes a long time. I think I could use collapse the data but do not know how.

    can someone guide me:
    1. how to collapse the data,
    2. How to do the poissonregression to obtain the ratio of the two SIRs? And how can I describe what this ratio is?

    Roland


  • #2
    If generation has codes 1 and 2, _d is the event and E is the estimated probability of the event from the native population experience, the IRR for 2.generation will be the ratio between SIR for the two generations (generation 2 / generation 1):
    Code:
    poisson _d i.generation , exposure(E) irr
    Since you use the old merge syntax, you may have a pre-11 version of Stata and thus no access to factor notation. In that case, code generation as 0/1 and:
    Code:
    poisson _d generation , exposure(E) irr
    poisson also works with collapsed data, and this should work (not tested):
    Code:
    collapse (sum) _d E , by(generation)

    Comment


    • #3
      Thank you Svend. It works - of course.

      Comment

      Working...
      X