Matching on variables to create study cohort

[Matthew Milano]

I want to perform a cohort study in a large database (~8m observations) comparing drug A users (exposed) to non-drug A users (unexposed). First I need to create my cohort and I want to match the exposed and unexposed on two variables (year of birth and GP practice).

I attempted doing this by first creating a variable based on those two variables using: -egen yob_GP = group(yob GP)-. This gave me a new variable (yob_GP) with a unique code for each possibility.


My question: How do I match (1:5 in my large dataset) the exposed and unexposed based on this unique code? My plan after matching on yob and GP practice is to assign the same index_date in the exposed to non-exposed so the start of follow up starts at the same time. Then perform propensity score matching (1:3) on around 20 variables.

Sample below:

* Example generated by -dataex-. For more info, type help dataex
clear
input float ID str5 GP float yob int index_date byte Drug_A float yob_GP
 1 "a9914" 1960 17157 1 2
 2 "a9914" 1960     . 0 2
 3 "a9957" 1962     . 0 3
 4 "a9957" 1962     . 0 3
 5 "b6642" 1954 20009 1 1
 6 "b6642" 1954     . 0 1
 7 "c9884" 1970     . 0 5
 8 "c9884" 1970 19697 1 5
 9 "c9884" 1970     . 0 5
10 "c9914" 1969     . 0 4
11 "c9914" 1969 17490 1 4
12 "c9914" 1969     . 0 4
13 "c9914" 1969     . 0 4
14 "c9995" 1972     . 0 6
15 "c9995" 1972     . 0 6
16 "c9995" 1972     . 0 6
17 "c9995" 1972     . 0 6
18 "c9995" 1972     . 0 6
19 "c9995" 1972 17441 1 6
20 "c9995" 1972     . 0 6
end
format %td index_date

[Molly Jeffery]

What version of Stata do you have? Have you looked at teffects nnmatch and specifying ematch(yob GP)?

[Øyvind Snilsberg]

perhaps something like this,

preserve
keep if !Drug_A
rename (ID-Drug_A) =0
tempfile unexposed
save `unexposed'
restore

keep if Drug_A
joinby yob_GP using `unexposed'

bys yob_GP: sample 5, count