Problems with stset on long data

Sigrid Vikjord

Join Date: Mar 2018
Posts: 69

Problems with stset on long data

02 Nov 2019, 09:28

Hi,

I'm trying to stset survival data in long format. I used reshape to convert from wide to long:

Code:

. reshape long PartDat PartAg HADSAnxi HADSDepr, i(id) j(_j)
(note: j = 2 3)

Data                               wide   ->   long
-----------------------------------------------------------------------------
Number of obs.                    78962   ->  157924
Number of variables                 712   ->     709
j variable (2 values)                     ->   _j
xij variables:
                      PartDat2 PartDat3   ->   PartDat
                        PartAg2 PartAg3   ->   PartAg
                    HADSAnxi2 HADSAnxi3   ->   HADSAnxi
                    HADSDepr2 HADSDepr3   ->   HADSDepr
-----------------------------------------------------------------------------

But when I'm trying to stset using id, I get an error due to "multiple records at same instant":

Code:

. stset enddate, id(id) failure(RegisStat==5) origin(time PartDat) 

                id:  id
     failure event:  RegisStat == 5
obs. time interval:  (enddate[_n-1], enddate]
 exit on or before:  failure
    t for analysis:  (time-origin)
            origin:  time PartDat

------------------------------------------------------------------------------
    157,924  total observations
    157,924  multiple records at same instant                   PROBABLE ERROR
             (enddate[_n-1]==enddate)
------------------------------------------------------------------------------
          0  observations remaining, representing
          0  subjects
          0  failures in single-failure-per-subject data
          0  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         .
                                          last observed exit t =         .

When I look at the failure variable RegisStat (død=failure, bosatt=still alive), it looks that there might be an error in my reshape; for instance is individual 1 marked as dead (død) at both observation intervals:

Code:

. list id enddate RegisStat PartDat _t0 _t in 1/20

     +----------------------------------------------------+
     | id     enddate   Regi~tat     PartDat   _t0     _t |
     |----------------------------------------------------|
  1. |  1   15nov2015        Død   22 Oct 96     0   6963 |
  2. |  1   15nov2015        Død   05 Oct 06     0   3328 |
  3. |  2   15may2009        Død   27 Nov 95     0   4918 |
  4. |  2   15may2009        Død   07 May 07     0    739 |
  5. |  3   15aug2010        Død   15 Apr 97     0   4870 |
     |----------------------------------------------------|
  6. |  3   15aug2010        Død   23 May 08     0    814 |
  7. |  4   15jan2019     Bosatt   14 Sep 95     0   8524 |
  8. |  4   15jan2019     Bosatt           .     .      . |
  9. |  5   15jan2019     Bosatt   23 Jan 97     0   8027 |
 10. |  5   15jan2019     Bosatt           .     .      . |
     |----------------------------------------------------|
 11. |  6   15jun2003        Død   09 Oct 95     0   2806 |
 12. |  6   15jun2003        Død           .     .      . |
 13. |  7   15jan2019     Bosatt   31 May 96     0   8264 |
 14. |  7   15jan2019     Bosatt   03 May 07     0   4275 |
 15. |  8   15jan2019     Bosatt   27 May 97     0   7903 |
     |----------------------------------------------------|
 16. |  8   15jan2019     Bosatt   18 Apr 08     0   3924 |
 17. |  9   15jan2019     Bosatt   08 Jan 97     0   8042 |
 18. |  9   15jan2019     Bosatt   07 Jan 08     0   4026 |
 19. | 10   15jan2019     Bosatt   29 Nov 95     0   8448 |
 20. | 10   15jan2019     Bosatt   25 Jun 07     0   4222 |
     +----------------------------------------------------+

I tried to include RegisStat in the reshape command, returning the following error message:

Code:

. reshape long PartDat PartAg HADSAnxi HADSDepr RegisStat, i(id) j(_j)
variable _j contains all missing values
r(498);

When I remove id from stset, I do not get the same error, but the observation time is too short (last observed exit should be around 22 years):

Code:

. stset enddate, failure(RegisStat==5) origin(time PartDat) 

     failure event:  RegisStat == 5
obs. time interval:  (origin, enddate]
 exit on or before:  failure
    t for analysis:  (time-origin)
            origin:  time PartDat

------------------------------------------------------------------------------
    157,924  total observations
     41,877  ignored because never entered
          6  observations end on or before enter()
------------------------------------------------------------------------------
    116,041  observations remaining, representing
     24,924  failures in single-record/single-failure data
  662179727  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         0
                                          last observed exit t =     8,554

Enddate is created this way:

Code:

-----------------------------------------------------------------------------------
RegisStat                                                            Registerstatus
-----------------------------------------------------------------------------------

                  type:  numeric (double)
                 label:  RegisStat

                 range:  [1,5]                        units:  1
         unique values:  3                        missing .:  0/157,924

            tabulation:  Freq.   Numeric  Label
                       117,678         1  Bosatt
                           730         3  Utvandret
                        39,516         5  Død

Code:

gen enddate = mdy(02,01,2019)
format enddate %td
label variable enddate "End of follow-up date"
replace enddate = RegisStatDat if RegisStat==5

Both reshape and stset have been working perfectly with wide data, using the same syntax.

Including enddate:

Code:

(note: enddate2 not found)
variable enddate already defined
r(110);

What do I do wrong? Could is be that I've misspecified the enddate variable? It is created this way:

Tags: None

Bjarte Aagnes

Join Date: Apr 2014
Posts: 783

04 Nov 2019, 12:38

I recommend "Joint modelling of longitudinal and survival data in Stata" by Michael J Crowther. It seems the following example from slide 17/18 show a stset command on data with similar structure:

Code:

clear

* https://www.mjcrowther.co.uk/pdf/JM_course_lectures.pdf slide 17/18

input id logb str20 trt time stime died
4 .5877866     D-penicil     0           5.27051 1
4 .4700036     D-penicil     .51473      5.27051 1
4 .5306283     D-penicil     1.01851     5.27051 1
4 1.163151     D-penicil     1.99595     5.27051 1
4 1.308333     D-penicil     3.43336     5.27051 1
4 1.386294     D-penicil     4.00285     5.27051 1
4 1.667707     D-penicil     4.99398     5.27051 1
end

list , clean

bysort id: gen start = time
bysort  id: gen stop = start[_n+1]

gen event = 0
bysort id: replace stop = stime if _n==_N
bysort id: replace event = died if _n==_N
 
list id logb trt start stop event if id==4, table noobs sepby(id)

stset stop, enter(start) failure(event=1) id(id)

* END https://www.mjcrowther.co.uk/pdf/JM_course_lectures.pdf slide 17/18

stdescribe, noshow
stvary, noshow

. Also, you might consider using attained age as the time-scale.

Last edited by Bjarte Aagnes; 04 Nov 2019, 12:40.

Comment

Sigrid Vikjord

Join Date: Mar 2018

Posts: 69
#3

05 Nov 2019, 03:17

Thank you and tusen takk, Bjarte Aagnes! The slides you shared was exactly what I needed.
Comment

Announcement

Problems with stset on long data

Comment

Comment