Despite the measures that are taken to minimize the risk of it occurring, missing data remains an issue in the analysis of clinical trials. The amount of missing data and the underlying mechanism can limit the ability to draw definite conclusions and can even lead to incorrect inferences being drawn. Whilst imputation approaches have been used at length to handle this issue, there is no definite agreement on the best imputation method, yet choosing the wrong one could have serious ramifications.

In the context of binary data, one simple method thought to be conservative in its assessment of the effect of the active treatment is non-responder imputation. With this method, participants with missing data are assumed to be non-responders, regardless of the reason for the missing data or whether or not the subject was responding to treatment. In superiority trials, this method is understood to result in more cautious estimates of drug effect on outcome measures.

In equivalence trials, on the other hand, the focus lies on whether one treatment is equivalent to another in terms of e.g. clinical efficacy. In this context, the alternative and null hypothesis from a superiority trial are inverted and non-responder imputation may bias results towards equivalence. The size of this effect depends on many factors, including the amount of missing data. Results from a simulation study with a wide range of scenarios will be shown to discuss to what extent this imputation approach can lead to biased results, and potential alternatives will be discussed.