Analysis of duration of response in oncology trials

The fraction of patients who respond to treatment and the duration of response in the subset of responding patients are commonly evaluated in oncology trials of cytotoxic compounds. While formal, comparative analysis of the fraction of patients responding to treatment is straightforward in a randomised trial, analyses that attempt to compare treatments in terms of the duration of response in responding patients are likely to be biased since the groups being compared are defined by the post-treatment outcome of response rather than by randomisation. Subsets of responding patients may not be comparable with respect to baseline prognostic factors and, consequently, formal comparative analysis is discouraged by the European Medicines Evaluation Agency. In an attempt to combine both the fraction of patients responding to treatment and the duration of response in responding patients, Temkin considered the probability of being in response function (PBRF) as a description of the treatment difference. Begg and Larson subsequently developed a parametric version of the PBRF under the exponential assumption. This paper briefly considers the PBRF as a means of estimating the expected duration of response across all randomised patients, thereby allowing a formal and unbiased comparison of treatments for duration of response. Building on earlier work, a more general and flexible approach to estimating the expected duration of response is offered to generalize beyond the exponential distribution.