Dropout can lead to bias even when the same proportion of people drop out across the arms of a trial. This happens when the people who drop out differ in some relevant aspect across trial arms. Likewise, it is possible for dropout rates to be different across trial arms and for this not to bias results. This happens when the factor causing the dropout is not associated with the outcome and/or capacity of treatment to influence that outcome. The risk of bias can be reduced by avoiding simple imputations, such as last value carried forward. A recent study  uses simulations to explore bias due to dropout. The paper goes on to explain how different methods of imputation may mitigate bias. But the CLAHRC WM Director takes issue with the authors on one point – showing that bias can arise both when dropouts are equal and unequal across trial arms is not tantamount to saying the risk of bias is the same across these scenarios. So readers should not infer that a measure of reassurance cannot be drawn when dropouts are the same across arms of a study.
— Richard Lilford, CLAHRC WM Director
- Bell ML, Kneward MG, Fairclough DL, Horton NJ. Differential dropout and bias in randomised controlled trials: when it matters and when it may not. BMJ. 2013. 346:e8668.