All Times EDT
Keywords: estimands, clinical trials, recurrent and terminal events, interpretability, robustness
Clinical trials are the gold standard for assessing the efficacy of therapeutical interventions and treatments in health research. The ICH E9 addendum on estimands has raised awareness that analysis and interpretation of trials involving complex disease processes are often not straightforward. Specification of estimands for treatment comparisons is complicated when patients are at risk of several potentially competing events, or when events can occur which interfere with the process of interest. In a trial in breast cancer patients with bone metastases, for example, interest may lie in the effect of bisphosphonates in preventing recurrent skeletal complications, but death may occur before the end of study. In cardiovascular trials, the risk of recurrent myocardial infarction and stroke is terminated by death. This is further complicated when so-called intercurrent events such as treatment switches or premature dropout can occur during follow-up. Some guiding principles for defining estimands in randomized trials with complex disease processes will be presented and discussed with a focus on recurrent and terminal events. Semi-parametric rate-based methods by Ghosh and Lin (2002) and Mao and Lin (2016) are among the most commonly used for assessing treatment effects based on recurrent and terminal events. We investigate factors influencing the limiting values of the Ghosh-Lin and Mao-Lin estimators using large sample theory, and report on numerical studies which illustrate their use and also robustness to violations of model assumptions. Our numerical investigation shows that single summary measures of treatment effects in complex disease processes should be accompanied by comprehensive intensity-based analyses to provide a clear understanding of how treatments impact disease. Data from a metastatic cancer trial and a stroke prevention trial are used to illustrate our points, and to provide guidance on strategies for dealing with intercurrent events.