Hi Henry,
As has been suggested before, I'm not sure that matching in the
setting of pre-collected data is worth the trouble unless you intend
to go back and collect more data on the sampled subset.
The way I see it, matching, when employed during the design phase,
permits control for important confounders while maximizing study
power. Another use for matching is as a way for adjusting for
confounders that are difficult to characterize and quantify (eg
confouding by genetic makeup or neighbourhood). Neither scenarios seem
to apply here but I could be missing something.
Best wishes,
salah mahmud
On 6/22/08, Henry <[email protected]> wrote:
> Many thanks for your suggestions, which will help me a great deal.
> Svend, I am using a longitudinal primary care data extracted by Dec 2002.
> A case-control study design has been used to identify an association
> between an outcome (event) and an exposure. I would like to control
> for potential confounders (age and sex) and thus matching on these
> variables.
> A 2x2 table is my interest to obtain the Odds ratios. This should give
> me the concordant and discordant pairs once the matching is done.
> Conditional logistic regression will definitely be the way forward for
> our analysis.
> My problem can be solved by what listers suggested of using dummy
> variables and then matching on age-groups excluding already matched
> cases.
> I guess this can be possible even when we extend to the 1:n matching
> (though still not sure till I try it out)
> Another important variable to introduce will be pair-ID for matched pairs.
> I had looked at what Maarten suggested but wasn't sure how to
> implement the packages and was also reluctant on the –sttocc- given
> nature of my data but with the suggestions; I can have a second look.
>
> *************************************************************************************************************************************************
> Dear Henry,
> Guess I would generate a new dummy variable (for both data sets) for
> the case where you want to merge by age-groups and then merge by this
> new age-group.
> Kind regards,
> Andrea
> *************************************************************
> There are now quite a lot of packages available in this area, see:
> -findit match treatment- (and try some other searches with -findit-)
> -- Maarten
> *************************************************************
> I get the impression that data have already been collected, and that
> the purpose of matching is to facilitate analysis (at the cost of
> dropping some of the control observations). Actually, matching
> complicates rather than facilitates analysis in case-control studies;
> at least you need to use conditional logistic regression (or -mcc-) to
> analyse correctly. So, if my impression is right, the recommendation
> is to analyse with -logistic- (or -cc-) including the potential
> confounders of interest, but avoiding to match and to remove any of
> the control observations. A variable like age could be grouped, e.g.,
> in five-year groups.
> Anyway, if you want or need to match, the usual way is to categorize
> a variable in, e.g., five year groups: 30-34, 35-39, etc. This is
> more handy, and it also facilitates reporting the results (you can
> stratify by age group).
> Hope this helps
> Svend
> **************************************************************
> On Fri, Jun 20, 2008 at 3:48 PM, Salah Mahmud <[email protected]> wrote:
> > For completeness, also see [ST] sttocc -- Convert survival-time
> > data to case-control data
> >
> > sttocc automates the process of sampling matched controls. It is
> > intended to generate nested case-control data from a cohort data but
> > it should not be difficult to "fool" it into sampling from a
> > cross-sectional data.
> >
> > You still need to create the grouped age variable as per above posts.
> >
> > In my experience, you will require several rounds of matching with
> > increasingly permissive age grouping to find matches to all your cases
> > unless you have lots of data and only 1 or 2 matching variables. This
> > could be implemented within a for loop where each successive loop
> > drops and then creates an age grouping variable that is slightly
> > cruder than its predecessor.
> >
> > For instance,
> > round age group variable
> > 1 agegroup = age ("exact" matching)
> > 2 agegroup = age collapsed into 2 yrs intervals
> > 3 agegroup = age collapsed into 3 yrs
> > intervals and so,
> >
> > Of course, you will need to exclude any matched cases (and perhaps
> > controls) before merging the ummatched cases to the remaining
> > controls.
> >
> >
> >
> >
> >
> >
> > On Fri, Jun 20, 2008 at 6:25 AM, Svend Juul <[email protected]> wrote:
> >>
> >> Henry wrote:
> >>
> >> I would like to carry out some matching for a case-control study using
> >> STATA but its proving to be a bit challenging to me. I have checked
> >> from achieves but a query close to mine on statlist was not answered
> >> in 2004. Could there be a way of matching cases to controls within a
> >> range of values say for age, a 40yr old case-patient can be matched to
> >> either a 38 or 39 or 40 or 41 or 42yr old control-patient? I have used
> >> the -merge- command to merge two datasets by sex and age of patients
> >> but it only works for 40yr old case matching a 40yr old control. For
> >> this case am still interested in a 1-1 matching but what if I extend
> >> this to a 1:n match? I want to have something of this sort:
> >>
> >> case-patient case-age sex control-patient control-age
> >> 00b7 35 1 00YP 35
> >> 00b7 35 1 0XC1 33
> >> 00b7 35 1 0001 36
> >>
> >> ==================================================================
> >>
> >> I get the impression that data have already been collected, and that
> >> the purpose of matching is to facilitate analysis (at the cost of
> >> dropping some of the control observations). Actually, matching
> >> complicates rather than facilitates analysis in case-control studies;
> >> at least you need to use conditional logistic regression (or -mcc-) to
> >> analyse correctly. So, if my impression is right, the recommendation
> >> is to analyse with -logistic- (or -cc-) including the potential
> >> confounders of interest, but avoiding to match and to remove any of
> >> the control observations. A variable like age could be grouped, e.g.,
> >> in five-year groups.
> >>
> >> Anyway, if you want or need to match, the usual way is to categorize
> >> a variable in, e.g., five year groups: 30-34, 35-39, etc. This is
> >> more handy, and it also facilitates reporting the results (you can
> >> stratify by age group).
> >>
> >> Hope this helps
> >> Svend
> >>
> >>
> >> __________________________________________
> >>
> >> Svend Juul
> >> Institut for Folkesundhed, Afdeling for Epidemiologi
> >> (Institute of Public Health, Department of Epidemiology)
> >> Vennelyst Boulevard 6
> >> DK-8000 Aarhus C, Denmark
> >> Phone: +45 8942 6090
> >> Home: +45 8693 7796
> >> Email: [email protected]
> >> __________________________________________
> >>
> >> *
> >> * 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/
> >>
> > *
> > * 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/
> >
>
> *
> * 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/
>
*
* 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/
