stset with multiple failure per subject data

[Thomas Pasvol]

Hi everyone,

I would like to ask for some advice on stset-ing my dataset with multiple failures per subject.

The event is 'admission to hospital'. So each subject can be admitted again after they are discharged.

I have set up my data as recommended by the Cleves tutorial for conditional risk set modelling (3.2.4) but when I stset, it only recognises the first line per id' & I get an output for 'single-failure-per-subject-data', with blank cells for the subsequent rows.

. stset end, fail(admit) enter(start) exit(end) id(id)

id: id
failure event: admit != 0 & admit < .
obs. time interval: (end[_n-1], end]
enter on or after: time start
exit on or before: time end

------------------------------------------------------------------------------
298 total observations
5 observations end on or before enter()
63 observations begin on or after exit
------------------------------------------------------------------------------
230 observations remaining, representing
230 subjects
31 failures in single-failure-per-subject data
188,425 total analysis time at risk and under observation
at risk from t = 0
earliest observed entry t = 20,698
last observed exit t = 21,792


Any help much appreciated.

thanks,

Tommy.

[Paul Dickman]

Providing a data example with dataex (as suggested in the FAQ) would be useful. You can make up the data if privacy is a concern.

You may need to specify exit(end .) instead of exit(end)

[Thomas Pasvol]

Dear Paul,

exit(end .) instead does not seem to work. Here are the first ten study participants as per dataex (I have removed the other variables as they are sensitive information). 'ID' is participant ID number, start is 'study' entry time, 'end' is study exit time, 'admit' represents an event and 'time' is analysis time.

Would be incredibly grateful for any suggestions. Thanks!

input int id float(start end) byte admit float time
1 20774 21187 1 413
1 21188 21792 0 604
2 20742 21792 0 1050
3 20781 21792 0 1011
4 21134 21792 0 658
5 21236 21792 0 556
6 20854 21304 1 450
6 21308 21792 0 484
7 21068 21792 0 724
8 20713 21792 0 1079
9 20837 21792 0 955
10 20836 21383 1 547
10 21384 21699 1 315
10 21705 21792 0 87

[Thomas Pasvol]

Copying CODE lines from dataex

. dataex id start end admit time, varlabel

----------------------- copy starting from the next line -----------------------
[CODE]
* Example generated by -dataex-. To install: ssc install dataex
clear
input int id float(start end) byte admit float time
1 20774 21187 1 413
1 21188 21792 0 604
2 20742 21792 0 1050
3 20781 21792 0 1011
4 21134 21792 0 658
5 21236 21792 0 556
6 20854 21304 1 450
6 21308 21792 0 484
7 21068 21792 0 724
8 20713 21792 0 1079
9 20837 21792 0 955
10 20836 21383 1 547
10 21384 21699 1 315
10 21705 21792 0 87

[Paul Dickman]

Sorry. It's exit(time .).

Following is come code that I hope will take you closer to the solution, but there are some issues I can't resolve.

There are two variations of the conditional risk set model; one where time is measured from study entry and one where time is measured from the previous event (i.e., the clock is set to zero after each failure). This code is for when time is measured from study entry. If you want to use the clock-reset variant then use your time variable rather than start/end times.

The result of this -stset- doesn't exactly mirror your data. For ID=10, for example, you can see that there was an event at time=21699 and you have the next observation starting at time=21705. The result of -stset-, however, is for that observation to start at time=21699. I don't know how to resolve that, maybe someone with more insight into multiple-failure data can provide additional help.

. input int id float(start end) byte admit float time

           id      start        end     admit       time
  1. 1 20774 21187 1 413
  2. 1 21188 21792 0 604
  3. 2 20742 21792 0 1050
  4. 3 20781 21792 0 1011
  5. 4 21134 21792 0 658
  6. 5 21236 21792 0 556
  7. 6 20854 21304 1 450
  8. 6 21308 21792 0 484
  9. 7 21068 21792 0 724
 10. 8 20713 21792 0 1079
 11. 9 20837 21792 0 955
 12. 10 20836 21383 1 547
 13. 10 21384 21699 1 315
 14. 10 21705 21792 0 87
 15. end

. 
. stset end, fail(admit) enter(start) exit(time .) id(id)

                id:  id
     failure event:  admit != 0 & admit < .
obs. time interval:  (end[_n-1], end]
 enter on or after:  time start
 exit on or before:  time .

------------------------------------------------------------------------------
         14  total observations
          0  exclusions
------------------------------------------------------------------------------
         14  observations remaining, representing
         10  subjects
          4  failures in multiple-failure-per-subject data
      8,945  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =    20,713
                                          last observed exit t =    21,792

. 
. list id start end admit _t0 _t _d _st

     +-------------------------------------------------------+
     | id   start     end   admit     _t0      _t   _d   _st |
     |-------------------------------------------------------|
  1. |  1   20774   21187       1   20774   21187    1     1 |
  2. |  1   21188   21792       0   21187   21792    0     1 |
  3. |  2   20742   21792       0   20742   21792    0     1 |
  4. |  3   20781   21792       0   20781   21792    0     1 |
  5. |  4   21134   21792       0   21134   21792    0     1 |
     |-------------------------------------------------------|
  6. |  5   21236   21792       0   21236   21792    0     1 |
  7. |  6   20854   21304       1   20854   21304    1     1 |
  8. |  6   21308   21792       0   21304   21792    0     1 |
  9. |  7   21068   21792       0   21068   21792    0     1 |
 10. |  8   20713   21792       0   20713   21792    0     1 |
     |-------------------------------------------------------|
 11. |  9   20837   21792       0   20837   21792    0     1 |
 12. | 10   20836   21383       1   20836   21383    1     1 |
 13. | 10   21384   21699       1   21383   21699    1     1 |
 14. | 10   21705   21792       0   21699   21792    0     1 |
     +-------------------------------------------------------+

[Thomas Pasvol]

This is brilliant Paul. Very helpful.

The reason the end dates are like that is because each event is an admission to hospital. Then while the patient is in hospital, they are not at risk of a further event until they are discharged. So the clock shouldn't really reset until the date of discharge. I will scratch my head and ask around. Thanks again!

T

[Wossenseged Jemberie]

Dear Statalist members,
I had post a question on proportionality assumption for multiple failures on this link (Statalist). It is, in short if the proportionality assumption should hold true for multiple failure data and if the Stata commands (such as stphtest) can be used with robust variance-covariance matrix (multiple failure) to check proportionality assumption.
thank you for your guidance