# Group Randomized Study

A group-randomized study is a

randomized? trial in which the

intervention? is assigned to a group of people but the

outcome? is assessed separately for each person. For example, we might assign 10 primary care practices to intensive management of Type II diabetes and 10 practices to usual care, and measure diabetes

control? in each patient as the outcome. Group randomized studies require special care in planning and in analysis.

## Planning A Group Randomized Study

Group randomized studies require special planning to allow for similarities in outcome between people in the same group. You need to know how similar the people are and you need an estimate of the

intra-class correlation? (

ICC). Typically, you will need a larger number of participants than comparable ungrouped studies. The

sample size? will depend on the

ICC and the

group size, as well as the

significance level? and

power? desired.

You have to know the intra-class correlation (

ICC) to start. Then you need some idea about the possible number of people per group. (You might have a range of possible values.)

As with any study design, you set the desired Type I error value, alpha, and the desired power. (LINKS)

You can then figure out how many groups you will need to randomize. You can look at different numbers of people per group to see what is most efficient! (LINK TO DOWNLOADABLE MACROS)

## Efficient Group Randomized Studies

Note that the larger the

ICC, the less efficient it is to add more people to a group.

**One guideline to keep in mind** is to try to keep the

VIF? at 2 or below. Furthermore, you have to balance the expenses, and it may be a lot more expensive to add a group than to add more people. But this won't help if you do not have enough power. If you have a very small number of groups, you also have to adjust comparisons of means for small degrees of freedom.

## Analysis of Group Randomized Studies:Special Considerations

- Analysis should use clustered-data methods?
- If the number of groups? or cluster?s is small, you will need to use t test?, not normal approximation
- Special care is needed for reporting group-randomized studies (LINK to CONSORT statement for group-randomized trials).

## Related Topics

## Related Articles

## References

*

Topic revision: r5 - 14 Dec 2010 - 19:38:03 -

MaryBanach