Matching

views updated May 17 2018

MATCHING

Matching is a method used to ensure that two study groups are similar with regards to "nuisance" factors that might distort or confound a relationship that is being studied. Consider a study designed to explore the possible effect of asbestos exposure on the risk of lung cancer. It is known that smoking increases the risk of lung cancer and that asbestos workers are more likely to be smokers than the general public. Smoking, therefore, confuses or confounds the asbestos-lung cancer relationship. In order to eliminate the confounding effect of smoking in a case-control study, one could ensure that the diseased and control groups had equal proportions of smokers.

Matching can be implemented using two main approaches: pair (individual) matching and frequency matching. Pair matching links each member of the case group to a member of the control group with similar characteristics. This type of matching is most commonly seen in studies using twins or paired body parts (e.g., a study of ocular disease in diabetes). Analysis in pair matching often uses the McNemar chi-square method. Frequency matching is more commonly used. The control subjects are chosen to ensure that the frequency of the matching factors is the same as found in the case group. Frequency matching can be implemented in various ways, including category matching, caliper matching, stratified random sampling, or a variant of pair matching.

Matching is most commonly used in case-control studies, although some researchers also use it in cohort studies. Theoretical analysis has shown that matching in cohort studies completely controls for any potential confounding by the matching factors without requiring any special statistical methods. There can be a loss of statistical power, however. On the other hand, in case-control studies matching does not completely control for confounding, thus requiring the use of statistical methods such as the Mantel Hanzel approach, standardization, or logistic regression. There can be substantial loss of power if a case-control study matches on a factor which is not actually a confounder.

The role of matching in epidemiological research is controversial. Many epidemiologists routinely match on age and sex, even when they are not confounders. This practice is to be discouraged. Since a matched case-control study still requires a complex statistical analysis and might reduce statistical power, an argument can be made that matching in case-control studies is not desirable (unless the distribution of the matching factor in the case group is extreme).

As an example of matching, consider a case-control study of the role of dietary factors in the etiology of lung cancer. Given the strong effect of smoking on lung cancer risk, and the relationship of smoking with low intake of fruit and vegetables, one might design a case-control study that matched on smoking status. Each new case would be classified into a smoking category (e.g., never smoked, past smoker, mild, moderate, or heavy smoker). Then, a control subject would be recruited who had a similar smoking history. Although this method appears to be a form of pair matching, it would be analyzed as a frequency matched study.

George Wells

(see also: Case-Control Study; Cohort Study; Epidemiology; Statistics for Public Health )

Bibliography

Kleinbaum, D. G.; Kupper, L. L.; and Morgenstern, H. (1982). Epidemiological Research: Principles and Quantitative Methods. Belmont, CA: Lifetime Learning Publications.

Rothman, K. J., and Greenland, S. (1998). Modern Epidemiology, 2nd edition. Philadelphia, PA: Lippincott-Raven.

matching

views updated May 29 2018

matching of a graph. See perfect matching.