Generate Random Dates Following a Distribution

[Anindit Chhibber]

Hi Eevryone,

I am trying to do a case control analysis for patients that are on drug 'Statin' matched with pateints that are not using the same drug and looking at outcome 'Y'.

Here in example below, I have MRN (unique identification number for each patient), StatinUser (0/1 for nonuser vs user) and OrderDate (Date on which statin was ordered). I would like to generate random order dates for each non statin user following the distribution of dates in 'orderdate' variable.

For instance, if i have 100 Statin users and 10 of them have ordered a statin in year 2010 and 20 of them have ordered statin in year 2015, I would like 10% of non statin users to have an order date in year 2010 and 20% of non statin users should have an order date in the year 2015.

* Example generated by -dataex-. To    install: ssc    install    dataex
clear
input long MRN float StatinUser int    OrderDate
802 1 16714
3897 1 17476
4077 0     .
16733 0     .
32086 1 16147
33100 1 19428
35303 0     .
35402 0     .
44859 0     .
46631 1 18004
47191 1 17755
51607 0     .
52936 1 16611
54114 0     .
55186 1 20467
63594 0     .
67389 1 20489
70508 0     .
94664 0     .
101758 1 17583
103275 0     .
106922 1 20956
114538 0     .
117457 1 19739
119313 1 18295
119750 1 18507
122721 0     .
122911 1 20555
124115 0     .
140665 0     .
141747 1 16236
141812 1 19166
144469 1 18777
149492 0     .
153932 1 16540
156828 0     .
157974 1 17680
162925 0     .
165324 0     .
171231 0     .
173534 0     .
174284 1 15956
174466 0     .
181628 0     .
188730 0     .
191346 1 21102
194977 0     .
197376 0     .
197426 0     .
207639 1 19527
220996 0     .
225169 1 18973
226852 1 19012
226878 0     .
228536 0     .
232462 1 20370
233668 0     .
235622 0     .
242065 1 18644
248450 1 19648
250316 0     .
267021 0     .
275867 1 18857
278226 1 20422
278770 1 19884
279968 0     .
286021 0     .
301515 0     .
307009 0     .
309781 0     .
319129 0     .
323121 0     .
329961 0     .
330845 0     .
331025 1 15105
333765 1 19599
335653 1 21606
336354 0     .
336388 0     .
337352 0     .
348391 1 16643
354142 0     .
356733 0     .
367078 0     .
368811 0     .
369744 1 19204
374363 0     .
382770 1 21165
385724 0     .
386219 1 21053
401612 0     .
424622 0     .
432898 0     .
433003 0     .
434175 1 17787
443721 1 16828
446468 1 19317
454843 0     .
469080 1 19837
469924 0     .
end
format %td OrderDate

Thanks!
Andy

[Mike Lacy]

You describe your study as "case-control," yet some of your description would seem to go against that. This may just be a terminological issue, or it could reflect a study design issue that is deeper than the question you ask. "Cases/controls," in common usage, are defined with respect to an outcome, but you describe cases and controls defined with respect to what I'd presume to be an explanatory variable, namely statin use. I might misunderstand you here.

In any case, I don't think the matching and case-control features of your study matter for your question. I gather that you want to *impute* a random OrderDate for each statin non-user so that the distribution of the year is the same among users and non-users. It is in general not possible to do that in a finite sample with different numbers of users and nonusers, as you have provided. Instead, we can impute years so that the *expected* distribution of year is the same for users and non-users. That is what I show below. Or, if you *do* have equal numbers of users and non-users, what I offer below can be modified to produce exactly the same year distribution among the user/nonuser populations, so post again if that's your situation.

Finally, you do not say what you want for the distribution of OrderDates within year, only that you want the *year* distribution to be the same. I'd presume (?) that you'd like to pick, with uniform probability, a day within each imputed year, and use that for the OrderDate. So, for each non-user, a remaining task beyond the code below is to assign a day number from 1, ..., 365(6) within each year and convert that to a date. I don't work with dates very much, so I'll leave that part to you or to someone else here who works with dates routinely. You could, alternatively, pick the actual OrderDate the same way I have picked the year below, but I'm presuming that's not what you want.

* Example generated by -dataex-. To    install: ssc    install    dataex
clear
input long MRN float StatinUser int    OrderDate
802 1 16714
3897 1 17476
4077 0     .
16733 0     .
32086 1 16147
33100 1 19428
35303 0     .
35402 0     .
44859 0     .
46631 1 18004
47191 1 17755
51607 0     .
52936 1 16611
54114 0     .
55186 1 20467
63594 0     .
67389 1 20489
70508 0     .
94664 0     .
101758 1 17583
103275 0     .
106922 1 20956
114538 0     .
117457 1 19739
119313 1 18295
119750 1 18507
122721 0     .
122911 1 20555
124115 0     .
140665 0     .
141747 1 16236
141812 1 19166
144469 1 18777
149492 0     .
153932 1 16540
156828 0     .
157974 1 17680
162925 0     .
165324 0     .
171231 0     .
173534 0     .
174284 1 15956
174466 0     .
181628 0     .
188730 0     .
191346 1 21102
194977 0     .
197376 0     .
197426 0     .
207639 1 19527
220996 0     .
225169 1 18973
226852 1 19012
226878 0     .
228536 0     .
232462 1 20370
233668 0     .
235622 0     .
242065 1 18644
248450 1 19648
250316 0     .
267021 0     .
275867 1 18857
278226 1 20422
278770 1 19884
279968 0     .
286021 0     .
301515 0     .
307009 0     .
309781 0     .
319129 0     .
323121 0     .
329961 0     .
330845 0     .
331025 1 15105
333765 1 19599
335653 1 21606
336354 0     .
336388 0     .
337352 0     .
348391 1 16643
354142 0     .
356733 0     .
367078 0     .
368811 0     .
369744 1 19204
374363 0     .
382770 1 21165
385724 0     .
386219 1 21053
401612 0     .
424622 0     .
432898 0     .
433003 0     .
434175 1 17787
443721 1 16828
446468 1 19317
454843 0     .
469080 1 19837
469924 0     .
end
format %td OrderDate
//
set seed 3486534 // make reproducible
gen year = year(OrderDate)
//
// First, a reality check:
if (StatinUser == 0) & !missing(year) {
  display as error "Some non user has an order date; error"
  exit
}
//
// Put statin users (non-missing on year) at top of file and count them.
sort year 
count if !missing(year)
local last = r(N)  // highest observation # with nonmissing year
//
// Each nonuser gets a year sampled with replacement from nonmissing observations.
gen int pos = ceil(runiform() * `last')
replace year = year[pos] if (StatinUser == 0)
// Check the result
tab year StatinUser, col chi2
di "Chi2 shows that year distributions do not differ beyond chance."