Gravar-mail: Differential dropout and bias in randomised controlled trials: when it matters and when it may not