Missing data in clinical trials: a different point of view.
Background and aim: Despite of the high frequency of missing data observed in most clinical trials, the problem continues to be overlooked. While single imputation techniques consistently underestimate the variance, multiple imputation approaches yield more accurate estimators. Even so, the unverifiable assumption of missing at random renders these strategies to be unreliable in many instances. In this short review, a clinical perspective is proposed to revise the main concepts related to missing data in the context of clinical trials.
Conclusion: Considering that in modern clinical research advanced statistical softwares are readily available, sophisticated methods including regression imputation, multiple imputation and maximum likelihood should be the preferred techniques to deal with missing data. However, a definitive solution does not exist, and prevention is by far the best treatment, since the uncertainty of the missing outcomes is always difficult to address.