Subgroup analyses

Be cautious of results for a selected group of people within a study.

Comparisons of health actions often report results for selected groups of participants, called “subgroups”. This is because people want to know whether the effect is different for different types of people. However, subgroup analyses are often poorly planned and reported and their results may be unreliable.

Explanation

Effects measured in groups of people do not apply to everyone in the groups. Comparisons of health actions often report results for selected participants to assess whether the effect of a health action is different for different types of people (e.g., men and women or different age groups). This is called subgroup analysis. Unfortunately, planned subgroup analyses are often not reported and many randomized trials report subgroup analyses that were not planned. In addition, reasons for doing subgroup analyses are rarely provided in trial protocols and reports.

Most differences in effects suggested by subgroup analyses are likely to be due to the play of chance and are unlikely to reflect true differences in the effect of a health action between different subgroups of people.

 

 

Example

In 1983, the authors of a paper that presented 146 subgroup analyses of a large comparison of drugs called beta blockers in people who have recently had a heart attack (the Beta Blocker Heart Attack trial), found that roughly 2.5% of the subgroup analyses had results that statistically were “significantly worse and 2.5% had results that statistically were “significantly” better. This is exactly what would be found if any difference in effect were due to the play of chance. The differences of the effect in the subgroups could have just happened by chance.

Remember: In a comparison, or a review, findings based on results for subgroups of people within treatment comparisons may be misleading.

Read more about Key Concepts

  Small studies

P-values 

Back to Top