Background/AimsPhase II trials with a safety lead-in (SLI), allowing one or two dose de-escalations based on toxicities from the first several patients at a dose, are increasingly common in oncology. Despite increased demand, formal evaluation of SLI designs remains limited to dose selection without evaluating the performance of the phase II efficacy recommendation. This study proposes a novel framework to evaluate phase II SLI trials with one to three dose levels and examines their operating characteristics compared to more rigorous dose-finding and monitoring alternatives.MethodsInformed by a review of registered and published phase II SLI trials, we simulated trials comparing the commonly used Simon two-stage design with an SLI of six-patient cohorts to an extension adding toxicity monitoring at the end of each stage. For two and three dose levels, we further compared against phase I dose-finding using the Continual Reassessment Method (CRM), Bayesian Optimal Interval (BOIN), or 3 + 3, followed by Simon with toxicity monitoring, where patients treated at the selected dose in phase I are rolled over to phase II efficacy evaluation. Methods were evaluated by simulation on overall performance (based on joint correctness of dose selection and phase II recommendation) and average sample size. Sensitivity analyses assessed alternative specifications, including phase II design parameters, omission of rollover, and toxicity-monitoring frequency.ResultsFor one dose level, adding toxicity monitoring to Simon with SLI decreased recommendations of toxic but efficacious doses from 24% to 7%, with little difference for safe doses. For two or three dose levels with at least one acceptable dose, SLI approaches had decreased overall success compared to using CRM or BOIN first for dose selection (up to 19% and 23% lower, respectively). Algorithm-based approaches (SLI and 3 + 3) more often terminated trials incorrectly for toxicity, even with acceptable doses, but had up to 36% higher success when all doses exceeded the toxicity threshold. Without phase II toxicity monitoring, unsafe recommendations exceeded 30% for SLI Simon, decreasing to 6% when toxicity monitoring was incorporated. Rollover modestly reduced sample size without substantial loss of overall success.ConclusionsThe Simon with SLI often results in unsafe or inefficient trial recommendations. Incorporating toxicity monitoring throughout phase II is necessary to reduce such recommendations, while using CRM or BOIN for initial dose selection can improve overall success when candidate doses are expected to be well-tolerated. Given that trials differ in prior dose-toxicity information, candidate dose levels, and design priorities, we provide a framework to conduct study-specific evaluations and quantify the trade-offs of using SLI compared to more formal dose selection and toxicity-monitoring alternatives.

Equipoise is usually considered a requirement for the ethical conduct of phase III randomised controlled trials. Notwithstanding theoretical critiques since the concept's initial proposal, equipoise has remained an important moral consideration. In oncology, many research questions are now pursued via clinical trial designs that diverge in various ways from conventional, fixed-design, two-arm randomised controlled trials. In this article, we outline the challenges that these new designs pose for implementing equipoise. Specifically, we consider challenges posed by trials incorporating seamless designs, 'immortal' trials, and trials incorporating adaptive techniques. While we note that equipoise retains a central role in the ethical evaluation of conventional randomised controlled trials, we contend that its utility as a moral construct is challenged by more recent, 'novel randomised controlled trial' designs. Consequently, other ethical tools become important in considering the different types of risk that trial participants are exposed to in modern clinical trial designs.

This essay responds to Heymemann et al.'s analysis of the applicability of clinical equipoise to emerging trial methods in cancer. Briefly, I argue that equipoise presses study designers to conduct trials at the point of maximum community uncertainty. Accordingly, equipoise encourages maximum information gain per patient enrolled (or inversel, minimal patient burden per gain in information). As such, equipoise is as much a condition for the ethical conduct of new study designs as it is for old ones. I close by suggesting that, whereas Heymemann et al. suggest new trial designs present a challenge to the operationalization of equipoise, I argue that equipoise presents a challenge for those wishing to operationalize new trial designs.