A treatment policy strategy if often used to handle intercurrent events such as treatment withdrawal in clinical trials. Such an approach seeks to estimate the effect of a treatment, regardless of whether patients withdraw from the treatment schedule early. This requires the collection of outcome data following treatment withdrawal, however data is often missing after treatment withdrawal complicating the analysis.

In this setting, retrieved dropout multiple imputation has been proposed as a useful method for estimation. This approach imputes off-treatment data based only on observed off-treatment data. But this may be impractical with limited observed data post-treatment withdrawal. Alternatively, reference-based multiple imputation can be used which assumes treatment withdrawals behave like those observed in a specified reference group. But this makes strong assumptions and disregards observed off-treatment outcomes.

This presentation will review these two different methods of imputation followed by an introduction to a novel approach, referred to as retrieved dropout reference-base centred multiple imputation, that draws its influences from the two aforementioned methods. The expected bias and root mean square error (RMSE) for this new method will be analytically explored, followed by application to an anti-depression trial.