Propensity Score Matching for Case-Control Studies - help!

[Faye Wang]

Hi all,

I'm studying a rare outcome and would like to do a 1:4 case-control match using propensity scores as there are numerous covariates that I'd like to control for. I'd like to use propensity scores to create a cohort of cases with matched controls before embarking on any further analysis (similar to the SAS 1:N case-control matching macro).

Below are the steps of my thought process and my questions regarding them:

1) Create ps score
- should i use a logistic vs. pscore vs. teffects vs. psmatch2? and how does one go about it?

2) Matching Process
- how do I match the cases to controls in a 1:N fashion? Ive seen the ccmatch command but it doesn't appear that it'll work well in this situation

I apologize for the rudimentary questions and thank you in advance for your help!

- F.

[Mike Lacy]

I don't have time to go into the details here but:
1) You could use whatever binary response model you want to create a propensity score. I'd use -logistic-.
2) Matching for case-control studies is a very common question on StataList, with many past discussions, to some of which I have contributed. If you search the StataList archives using your favorite search engine with something like /site:statalist.org case control match/, you will find a lot of helpful past answers to questions similar to yours.
3) I would mildly question using propensity scoring for a matched analysis here. As you might know, matching in case-control studies does not work in the same way as it does in cohort studies, and I wonder about whether matching on all the variables entailed in a propensity scoring approach is actually going to work as you want. I presume there's a literature on best practices for propensity score matching for case-control studies, which I have not read, but which I would if I was considering what you have in mind.

Feel free to follow up as necessary.