Matching

Advantages of Matching

There are benefits to matching in addition to the intuitive appeal of comparing groups that appear similar to one another. Matching can facilitate the selection of a referent group without requiring identification of the entire base population. For example, it may be fairly easy to select as a matched control the “next” patient at a hospital or clinic where cases are identified. On the other hand, it may be much more difficult to enumerate and then enroll a random sample of all patients from the hospital or all potential patients from the surrounding geographic area. Matching can also be an efficient way to identify controls when controlling for factors such as neighborhood or sibship is of importance. Because there would be very few existing appropriate control subjects (people from the same neighborhood or sibship as cases) in the overall base population, choosing a random sample of this population is unlikely to yield a suitable control group. Finally, matching may result in a gain in precision of the estimate of association. This will be most apparent when the matching variable is a strong confounder.

Disadvantages of Matching

Matching also has disadvantages that should be carefully considered. Matching on many variables may make it difficult to locate matched controls, and information on the matching factors needs to be collected for “extra” controls that will not actually end up matching to cases. These factors may decrease cost efficiency of the study. Also, matching variables cannot be considered as independent risk factors themselves. This is because they have been set by design to be distributed similarly in cases and controls. (It is still possible to assess effect modification by the matching variables.) Finally, overmatching may result [Page 636]in reduced statistical precision or a biased estimate of association. Matching on strong correlates of exposure, variables associated with exposure but not disease, or factors that are affected by the outcome or exposure of interest should be avoided.

...