Estimands and Sensitivity Analyses

An estimand is a quantity used to define a treatment effect in a clinical trial. In many cases, clinical trial planners skipped the step of defining the estimand in their rush to pick a test statistic and calculate planned sample size(s). This would sometimes lead to ambiguity on how results of a trial were to be interpreted. In this chapter we describe estimands in detail and explain the importance of defining estimands when planning randomized trials and doing this before picking a test statistic to use in evaluating trial outcomes. The estimand is key to defining the scientific question the trial needs to address. When patients drop out or fail to follow a planned regime within a randomized clinical trial and stakeholders disagree on how this impacts the analysis of the trial, interpretability of this trial can be called into question. A clear definition of treatment effect ought to capture how dropouts and protocol violators will be handled. In this chapter sensitivity analyses are tied to the definition of the estimand in a trial. In practice, sensitivity analyses are often ad hoc and only addressed after a study is completed. Considering both estimands and sensitivity analyses in planning will improve the interpretation of results from completed randomized trials. While regulators (e.g., the US Food and Drug Administration) have been particularly interested in advancing these ideas, utilization of these ideas ought to improve the interpretability of randomized trials more generally.