|
Abstract:
|
Treatment effects on recurrent events are of primary clinical interest in many diseases such as asthma or chronic heart failure. When death makes it impossible to experience further events, defining an appropriate measure of the treatment effect, the estimand, is challenging. For example in chronic heart failure, patients may experience repeated hospitalizations, but are also at an increased risk of death. For a test treatment which reduces mortality compared to a control treatment, one may observe more hospitalizations under the test treatment simply because patients with a high risk of hospitalizations may die earlier under the control treatment. Many statistical analysis procedures have been proposed for such data, however it is often unclear which estimands these imply. As emphasized in the causal inference literature and the ICH E9(R1) guideline, the definition of the estimand of interest should precede any statistical analysis. We focus here on estimands for recurrent events terminated by death that have a causal interpretation. We also discuss whether common statistical analyses imply causal estimands of interest. Chronic heart failure is used as a motivating example.
|
