Following the guidance documents on handling missing data in clinical trials by US FDA and EMA in 2010, sensitivity analyses are recommended in general to assess the robustness of the analysis results and evaluate the impact of deviation from the missing data assumptions. Recently many new approaches for sensitivity analysis have been proposed for analysis of longitudinal clinical trials with missing data. However, there are still a lot of ambiguity and challenge on the definition of estimand, or the parameter of treatment effect to be addressed by the clinical trials. Different approaches used in the primary and sensitivity analysis models sometimes may address different estimands. Without a clearly defined estimand, it can be complicated or sometimes comparing apples to oranges. In this roundtable luncheon discussion, we will use some examples to illustrate the problems, and share experiences and perspectives on these issues from regulatory and pharmaceutical industry statisticians. Estimands associate with mixed model for repeated measures, control-based imputation, delta-adjustment-based tipping point, and some other sensitivity analysis methods will be discussed.