Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Matching in cohort studies

    Hello
    I am trying to conduct a cohort study to compare image quality between two hospitals. Both hospitals are using the same imaging device . I have 500 patients from hospital A and another 500 from hospital B. Beside the location of the imaging ( Hospital A Vs Hospital B) , there are other confounders that may affect image quality like body weight , age, ..etc. I want to match patients from hospital A with those from hospital B in regards to these confounders. Then, I will compare outcomes in the matched cohort. How can I run matching in Stata? . There are many resources about matching in case-control studies, however, I could not find the commands for matching in cohort studies. Thank you in advance for your help.

  • #2
    There's a reason you don't find much in the way of matched cohort studies out there: they're not very practical. In a case-control study, you typically have a relatively small number of cases and a huge pool of controls from which to pick matches. Because of the huge pool of controls, the fact that some people may have unusual combinations of values for the matching variables is not so problematic.

    But in your situation the number of people in the exposed and unexposed groups are fairly comparable, 500. But that's a difficult number for matching: it's large enough that you will probably encounter unusual combinations of your matching variables. But it's also small enough that you probably will not find matches for all of them in the other group. If you have several matching variables, you are likely to end up with an unsatisfactorily small set of matched pairs. You are probably better off forgetting about matching and just do your analysis with adjustment for these variables in a multivariate analysis.

    That said, if you want to see if you can get an adequate number of matched pairs, just use the same programs that you would use to do the matching for a case-control study. The matching algorithms do not depend on the design being case-control: they will be just as correct when applied to cohort data.

    Comment


    • #3
      Dear Clyde,
      Your answer was really informative and helpful. Using your comments, I will discuss this further with our statistician who chose to use matched cohort instead of multivariate analysis. Thank you very much.

      Comment


      • #4
        Apart from the "adjustments" for confounders in the multivariate regression, maybe some steps could also turn out helpful: inclusion and exclusion criteria could select a well "matchable" sample; stratification in the statistical analysis would work well in many cases.
        Best regards,

        Marcos

        Comment

        Working...
        X