subgroups analysis between propensity score matched variables

[Reinout Heijboer]

Hi,

After I've done a nearest neigbor propensity score matching, I have made an comparisons between the 2 treatment subgroups based on their matched variables. With the exception of the difference in gender, the differences in all matched variables (15 variables) were successfully reduced (no significant P-value between both treatment groups).

Now I want to do an subanalysis of only male patients by difference in treatment to see if they are associated with higher rates my outcomes.

Would you reckon this is a good solution to include gender in my study? Do I need to do a Bonferroni adjustment because of multiple comparisons?

Thank you in advance,

Reinout

[Marcos Almeida]

You may wish to estimate it anew, this time with the option ematch(gender).

[Reinout Heijboer]

I've used the psmatch2 command. You recon I should use teffect nnmatch (outcome_variable varlist) (treatment_variable), ematch(gender)
command?

[Marcos Almeida]

Actually, you may wish to try and see what happens.

In fact, I assumed you were using standard command in Stata, therefore, the - teffects - command, since there is a FAQ recommendation to specify user-written programs otherwise.

[Reinout Heijboer]

Thanks Marcos for taking the time to look at my problem!
The teffects command didn't really work for me. Would you know a method that I can use to work with my current matching. Maby an extra analysis of my outcomes if I specify for male only?
Thanks,

Reinout

[Marcos Almeida]

Hello Reinout,

When you commented it didn't "work" for you, perhaps sharing command and output - as recommended in the FAQ - would entice helpful advice.

That said, "nearest neighbor" is just one of the strategies (besides, it can be "honed" so as to "fit" well towards your data) when performing a propensity score. You may "match" under other method, delve with calipers or you may perform other sort of propensity score analysis, without having to match, but still "balancing" on the variables as well.

If you haven't done so far, the Stata Manual is a great way to start.

[Reinout Heijboer]

I think I've had a bit of an overkill on information concerning PSM. My matching is correct, but I should not have done a T-test on my before/after table to assess if matching was done correctly.
An accepted method to assess equal distribution of matched variables is by using standardized differences definded as the mean difference between the groups divided by the SD of the treatment group (Austin, Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples; Stat Med 2009). Can I express this as a number and state that matching was done proper if this number is <0.25?

Or Should I use the -pstest- command for standardized % bias before and after matching; or -pbalchk- for standardized mean differences, not expressed as a percentage?
The standardized % bias should be below 10, and the standardized mean differences should be <0.05 if I'm correct?