Loss to follow-up is often hard to avoid in randomised trials. This article suggests a framework for intention to treat analysis that depends on making plausible assumptions about the missing data and including all participants in sensitivity analyses