Issues related to the unit of analysis

More than two treatment groups

Many trials are designed to compare more than two treatments. However, there may be two or more experimental interventions, for example a drug at different doses, or variations on a counseling intervention. Alternatively, there may be more than one control group, perhaps an established treatment and a placebo.

The most common problem reviewers experience is trials with several experimental groups. If you are comparing each of the treatment groups with placebo in two separate meta-analyses (i.e. as two separate comparisons in RevMan), then the study can be treated as two separate trials (intervention 1 versus placebo and intervention 2 versus placebo). If however, you are putting all three arms of the study into the same meta-analysis it can be tempting to, for example, enter the data as if it were two trials, one comparing high dose with placebo and one comparing low dose with placebo. But when you then pool these results in a meta-analysis you will be counting the placebo patients twice. This approach should not be followed.

There are two main approaches to dealing with trials like this. The first is to break up the control group into several parts, so that the total numbers add up to the original size of the group. The second is to group together all the experimental groups and compare them collectively with the control group. There’s no single right answer, since both approaches have advantages and disadvantages.