topic How to create a control group based on multiple matching variables? in SAS Programming

How to create a control group based on multiple matching variables?

Denali — Wed, 22 Sep 2021 17:10:44 GMT

Dear All,

I am designing a matched case-control study. I have 25 patients diagnosed with COVID and cancer as the case group. I would like to select 50 subjects from a dataset containing 4000 cancer patients as a comparison/control group based on the following matching variables: age (+/-5 years), cance type (8 cancer types: lung, melanoma, sarcoma...etc), gender (male vs. female), year of chemotherapy, race (White vs. Black vs. Others) and survival > 6 months (Yes/No). The priority of the matching variables are listed as the order above.

Can you please advise me on the code? Please let me know if you need further clarifications.Thank you!

Re: How to create a control group based on multiple matching variables?

Reeza — Wed, 22 Sep 2021 17:21:42 GMT

Have you looked into PROC PSMATCH?

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.4/statug/statug_psmatch_examples04.htm

If you need code examples you will need to provide data examples, specifically both of your input data sets.

Re: How to create a control group based on multiple matching variables?

Denali — Wed, 22 Sep 2021 17:36:03 GMT

@Reeza Thanks for your suggestions. l am not specifically looking for propensity score matching but just the regular matching step. Is there a code for that instead? Do I need to run the Proc Freq first to ensure the frequency of each variable or it does not matter so much?

Re: How to create a control group based on multiple matching variables?

Reeza — Wed, 22 Sep 2021 17:42:08 GMT

Then you're just writing manual SQL code. There isn't a proc for direct matching, but the Mayo clinic used to have a bunch of SAS macros.
I personally thing PSMATCH with the REGION and CALIPERS can be set to do what you want but I've never directly tried it.

I would have assumed that you've ensured that your criteria makes sense already for the matching so not sure what the PROC FREQ will add to the process.