A1314
Title: Estimands versus algorithms in studies with competing events and interest in treatment mechanism
Authors: Takuya Kawahara - Harvard Medical School and Harvard Pilgrim Health Care Institute (United States) [presenting]
Jessica Young - Harvard Medical School and Harvard Pilgrim Health Care Institute (United States)
Abstract: In the presence of competing events, investigators might be interested in a direct treatment effect on the event of interest that does not capture treatment effects on competing events. Classical survival analysis methods that treat competing events like censoring events, at best, converge to a controlled direct effect that captures the treatment effect under the complete elimination of competing events which is often difficult to imagine. Recently, separable direct effects were proposed, which are the effects of modified versions of the study treatment with mechanisms removed other than those directly affecting the event of interest. These alternative notions of direct effect may have more practical relevance. Examples of data-generating conditions will be illustrated under which controlled and separable direct effects are identified but may take different values to varying degrees, including possibly different signs. This provides insights into the degree to which using even an unbiased estimator for a controlled direct effect could misleadingly inform a separable direct effect when that is really of the underlying interest. Conditions are also considered under which neither effect is identified due to the presence of an unmeasured common cause of the event of interest and the competing event and the bias associated with consistent estimation algorithms for these different effects.