What came first – the analysis or the estimand?
Analysis of Clinical Trials Theme
Different statistical analyses of the same data commonly lead to different results. This is hardly surprising: they are probably making different assumptions about the model or about the missing data. But sometimes they are estimating different quantities, or estimands. For example, missing data in the experimental treatment arm of a randomised trial might be imputed on an implicit assumption either that patients remained on experimental treatment, or that they received standard care: and these lead to estimation of different estimands.
It is very tempting to define analyses and then comment on what they estimate, but the pharmaceutical industry is now moving towards defining an estimand and then choosing suitable analyses. I will discuss these changes, together with other situations where an understanding of estimands is important, including covariate adjustment and data that are missing due to death.
On arrival, external attendees should sign in at reception.