Sensitivity to missing not at random dropout in clinical trials: Use and interpretation of the trimmed means estimator