Attrition bias: what the dropouts would have told you

Attrition bias arises when people who leave a study before it ends differ, in some way tied to the outcome, from the people who stay in it. Some dropout is normal in almost any study that follows people over time. The bias appears when dropout is not random, and especially when it happens at different rates, or for different reasons, in the different groups being compared.

A weight-loss trial illustrates the pattern well. People who find a diet unpleasant or ineffective are more likely to quit early, while people who are doing well tend to stick around to see the result. If the treatment group loses more of its struggling participants than the comparison group does, the people remaining in the treatment group at the final measurement are no longer a fair sample of everyone who started, and the reported result flatters the treatment.

Two traps follow. First, a trial can report an impressive completion rate while still hiding a meaningful, and biased, pattern in who exactly did not finish. Second, simply comparing the people who finished each group, rather than everyone originally assigned to it, can turn a fair comparison into an unfair one, since it discards the very participants whose absence carries information about the treatment.

When a study reports its result, check how many participants dropped out of each group, whether the dropout rates differed between groups, and whether the analysis included everyone as originally assigned, an approach usually called intention to treat. A study that reports a striking benefit alongside heavy or uneven dropout deserves a second look at what happened to the people who left. Pair any completion figure with a comparison of dropout reasons across the groups.