How do missing outcomes affect interpretation of a randomized trial?

Why

First describe how much outcome information is missing in each group and at which visits. Then examine whether missingness relates to adverse effects, poor response, withdrawal or study operations. Missing completely at random, missing at random conditional on observed variables and missing not at random represent different assumptions, not labels established by a normality test.

For instance, patients who discontinue because symptoms worsen may be less likely to report later scores. A complete-case analysis could make a poorly tolerated treatment look better. Prevention through follow-up and outcome collection is better than any statistical repair.

During design, align data collection and planned analysis with the estimand, define handling of intercurrent events and describe assumptions. During analysis, choose a defensible method such as a specified mixed model or multiple imputation where appropriate and conduct sensitivity analyses, including plausible departures from missing-at-random assumptions. Never claim that imputation reconstructs the true missing outcomes.