Abstract:
|
A composite estimand is straightforward to construct with a binary outcome in a clinical trial - simply treat the intercurrent event (IE) as a failure and absorb this into your variable definition. The procedure is less clear in the context of continuous outcomes; however, using trimmed means provides one potential path forward. The approach sets the "missing values" as the worst observed outcome and then trims away a fraction of the distribution from each treatment arm before calculating differences in treatment efficacy. Because this method penalizes treatment arms with more IE, it provides a composite measure of the treatment’s effect on both the clinical outcome and the IE rate. The drawback of composite estimands lies in their interpretability; therefore, we prove under what assumptions this method can extend to other types of estimands. This paper elucidates, the often ignored, but important fact, that one statistical methodology can be aligned with different estimands depending on the veracity of the underlying assumptions.
|