If you have frequency matched in a case-control study, you generally do
not need to use a matched-pair analysis (e.g., conditional logistic
regression). However, you still need to control for the matching
variables in the analysis (e.g., unconditional logistic regression with
matching variables as covariates). This is because matching results in
the controls being more similar to the cases than they are in the source
population, biasing the OR towards 1 (see for example Rothman KJ and
Greenland S, Modern Epidemiology 2nd ed, Lippincott-Raven Publishers,
1998, pp150-152).
The only situation where there is no need to control for the matching
variable in the analysis for a case-control study is if the matching
variable is not associated with the exposure, in which case there would be
no need for matching since that variable could not be a confounder.
Variables matched on in a cohort study or randomized trial do not need to
be adjusted for in the analysis.
Peter
----------------------------
Peter W. Callas, PhD
Medical Biostatistics
University of Vermont
Burlington, VT 05405
USA
email: [email protected]
----------------------------
On Sun, 2 Apr 2006, Marino, Jennifer wrote:
> Raoul-
> If your frequency matching was successful the distribution of whatever
> variable you matched on should be the same in your cases and controls.
> (You can look at whether the matching was successful, of course, with a
> simple tabulation by case status.) You don't need to use conditional
> logistic regression, which uses the matched set as the base unit for
> analysis and takes the loss of independence between observations imposed
> by 1:n matching into effect - if you do use it, incorrectly, it will
> decrease the power to detect associations and provide inappropriately
> broad confidence intervals for your coefficient estimates.
>
> If your frequency matching was incomplete you can include the match
> variable in your analyses to control for it. Controlling for age by
> frequency matching might still be inadequate if the variable
> distribution within a subgroup is patterned in a fashion related to your
> outcome and exposure, in which case you can control for it to some
> extent by including it in your multivariate analysis. (For example, for
> some women's health outcomes, even after frequency matching on age we
> might still need to take age into account in the analysis because age
> has a different effect in women before and after menopause and our
> matching didn't correct for that.)
>
> Hope that helps,
> Jen Marino
> University of Washington
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Raoul C
> Reulen
> Sent: Sunday, April 02, 2006 6:37 AM
> To: [email protected]
> Subject: st: frequency matching in case-control study
>
>
> Hi,
>
> If I would conduct a case-control study and would want to frequency
> match cases and controls, what STATA command would I have to use? Can I
> simply include a variable that presents the group id (1=males age 50,
> 2=males age 60, 3=female age 50 etc.) and use .logit? Or should I use
> .clogit? My understanding is that the command .clogit is used for
> individual matched case-control data. Thanks.
>
> Raoul
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
