The National Academy of Sciences has recommended radical changes in handling missing data in confirmatory trials. Subjects should be followed regardless of discontinuation of treatment, and values after discontinuation should be used in the analysis. Endpoints should exist and be meaningful for almost all subjects. Studies should be designed and conducted so as to minimize the number of missing observations.

All single imputation methods (e.g., LOCF) are deprecated. MAR methods may also be unrealistic. Serious efforts must be made to develop and justify plausible models for missingness and outcome, and they are unlikely to be simple.

Implementation would entail big changes in the way trials are designed, conducted, and analyzed by sponsors and reviewers, but I think the recommendations are sound