Abstract
The duration of response (DoR) is defined as the time from the onset of response to treatment up to progression of disease or death due to any reason, whichever occurs earlier. The expected DoR could be a suitable estimand to measure the efficacy of a treatment but is in practice difficult to estimate, since patients' follow-up times are often right-censored. Instead, the restricted mean duration of response (RMDoR) is often used. The RMDoR in a time τ is equal to the expected DoR restricted to the interval (0,τ) . In this paper, we consider the behaviour of the RMDoR as a function of τ and its suitability as a measure to quantify the efficacy of a treatment. Besides, we focus on the estimation of the RMDoR. In oncology, the events response to treatment and progression of disease are typically detected through time-scheduled scans and are therefore interval-censored. We describe multiple estimators for the RMDoR that deal with the interval censoring in different ways and study the performance of these estimators in single arm trials and randomised controlled trials.