Related Experiment Videos
Statistical methods in epidemiology. IV. Confounding and the matched pairs odds ratio
1Division of Child Health, Sheffield Children's Hospital, University of Sheffield, Western Bank, UK. a.s.rigby@sheffield.ac.uk
Purpose:
This paper introduces readers to the problem of confounding in epidemiology, how it differs from effect modification and how to deal with it statistically. There are several options for dealing with confounding. These include techniques based on matching, on stratification and on regression. This paper reviews the first method, matching and the matched pairs odds ratio.
Method:
For 1:1 pair matching (1 case and 1 matched control), the matched pairs odds ratio is introduced as the ratio of discordant pairs. A method for calculating confidence intervals based on the normal approximation is described.
Results:
Some properties of the matched pair design are illustrated by taking examples from the authors' own teaching experiences.
Conclusion:
Matching remains a difficult design option in epidemiology. Its 'best' use is for special types of studies such as for those on twin pairs.