Online Program

Return to main conference page

All Times EDT

Thursday, September 24
Thu, Sep 24, 12:00 PM - 1:15 PM
Virtual
Roundtables

TL18-ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials (301106)

*Sharon E McDermott, Covance 
Swarna Reddy, Covance 

Keywords: Sensitivity analysis, ICH E9(R1) addendum

ICH E9(R1) finalized 20November 2019 states “The protocol and the analysis plan should pre-specify the main estimator that is aligned with the primary estimand and leads to the primary analysis, together with a suitable sensitivity analysis to explore the robustness under deviations from its assumptions”. The addendum to this guidance includes additional detail on use of sensitivity analyses. Sensitivity analyses include known tests of assumptions, such as for proportional hazards or normality. These also include examining actual data to see how different imputations impact analysis. If the examination is done one variable at a time, it can be a long process and must be interpreted as a whole. If it is done with several variables at once, it is not possible to identify the variables that are sensitive to the outcome. Missing data are the data points most often examined, but other key assumptions in the analyses should also be examined. Depending on the endpoint, key assumptions can be that ratio of the hazards for any two subjects is constant over time or a distribution assumption such as normal, exponential. The Addendum also mentions discussing whether assumptions are testable or untestable.

Questions for the table to discuss: Q1. Is a discussion of sensitivity analysis in each of your analysis plans? Are there displays for the study report or other discussion in the study report? Q2. Is a sensitivity analysis needed for futility analyses and adaptive design decisions in addition to the main statistical analysis? Q3: For untestable assumptions, is there value in examining how outliers impact the results? Q4: What happens if the sensitivity analysis shows the main analysis is not robust?