Risk set matching / sampling

[johanna3]

Hi,

I need to do risk set matching on my dataset.

The cases are women whom become pregnant at eg. day X after diagnosis. The controls are the entire population and a woman can be a control if not yet pregnant at day X. 

Does anyone know where to download syntax for a macro or anything that can be of use. Besides the time variable I have an already calculated PS score to match on. I would prefer to do it with replacement. 

[StatDave]

You might look into PROC PSMATCH to see if it can do what you need.

[johanna3]

Dear Dave,
PSMATCH does not work as it condition on the future, so I need to do a risk set matching instead.

[sbxkoenk]

Hello,

Have you tried to program it yourself with a data step (or inside a macro)?

If yes, you can post the code in this thread (even when it's entirely wrong). People (community members) will notice / discover what you try to do and will correct your statements in a quick and efficient way.

What you can also do: post a little program where you create an extract of the cases dataset and an extract of the controls data set (use a datastep with datalines). Indicate how a possible outcome dataset would look like and community members will help you to go from the dataset(s) you have to the dataset you want.

You can search the web (or this communities site) for SAS-papers on risk set matching but you will do a lot of reading as a result and nothing will exactly be what you want.

Kind regards,

Koen

[johanna3]

Hi Koen,

Thank you for your reply.

I have done a lot of reading and found some macros - but no where I can download some macro syntax.
This surprises me as this is a "common" problem/methodology, and I thought some procedure (like there is now procedures for PS matching of all kinds) or pre-existing code/macro/syntax existed.

Im afraid Im not good enough at coding to be able to figure out to code something like this. I have found an old paper (1995) explaining different matching procedures - and exactly this one I need to create a risk set - using a macro called %match - with an appendix to follow up. I was just curios if that kind of syntax was available anywhere to be adjusted to my data of course - but principally "all" risk set matching is the same - based on different variables of course.

This is the paper: https://www.mayo.edu/research/documents/biostat-56pdf/doc-10026923

[sbxkoenk]

Hello,

I haven't looked at the paper you have provided (yet), but here's another paper that might help:

Paper 152-30 (SAS Users Group International 30 in Philadelphia)
SAS® Programs to Select Controls for Matched Case-Control Studies
Robert Matthews and Ilene Brill
University of Alabama at Birmingham, Birmingham, AL

https://support.sas.com/resources/papers/proceedings/proceedings/sugi30/152-30.pdf

However if you are not good at coding it can become problematic to interpret and grasp all the code provided there.

Also, the code is sub-optimal (especially for big data sets). I suggested a quicker solution (and it was accepted as a solution) by means of hash-tables in this communities entry:

https://communities.sas.com/t5/SAS-Programming/Macro-is-taking-a-long-time-to-run-Risk-set-sampling/...

You will find even more communities entries on this topic if you supply 'risk set' as keywords to search on.

I agree 'risk set matching' / 'risk set sampling' is a common problem that needs to be 'solved' very often but a macro that covers all possible use cases / flavors would have to be very generic and therefore very lengthy. The data step is so very agile and can be so easily 'tweaked' to your personal requirements that your code can be very brief.

Good luck,

Koen