Intention to treat: why dropouts still count in the result

Intention to treat is a rule for analyzing a randomized trial: every participant is counted in the group they were originally assigned to at random, no matter what happened afterward, including if they stopped the treatment, switched groups, or never finished the study. The alternative, comparing only the people who actually completed their assigned treatment, is called a per-protocol analysis, and it can look far more flattering for a treatment while being far less trustworthy.

The reason intention to treat matters becomes clear once dropout is uneven. If people quit a trial's drug arm because of side effects, or because the drug was not working, and a per-protocol analysis drops them from the comparison, the drug arm left standing is a healthier, more compliant group than the one that was originally randomized, no longer a fair match for the comparison group. Intention to treat keeps that original random assignment, and the fairness it bought, intact.

Two traps follow. First, a trial can describe its analysis as intention to treat without actually including every randomized participant, a gap serious enough that reviewers of major trials have found reports claiming the label while quietly excluding patients who did not comply. Second, intention to treat tends to understate a treatment's effect among people who actually take it as directed, since it is measuring the effect of being assigned the treatment, not the effect of receiving it, which is a different and sometimes more useful question.

When a trial reports its results, check explicitly whether the analysis included every randomized participant in their original group or only those who completed the protocol as assigned. A large gap between the number randomized and the number analyzed is worth noticing on its own. Pair an intention-to-treat result with a look at any per-protocol figures the trial also reports, since a wide difference between the two often points to meaningful dropout.