Assessing the feasibility of complex and innovative trial designs

最近提出了各种各样的复杂和创新的设计（CID），这些设计超越了标准的随机对照试验。这些设计有可能使临床试验更快，更有效，从而提高药物开发的速度，更快地为患者提供更好的治疗。然而，这些设计的新奇导致了它们在实践中运行的可行性的巨大不确定性。当使用复杂的设计时，可能会进一步放大临床试验中招募、依从性和随访不良的长期担忧，而CID特有的特征，如中期分析、多次随机化、生物标志物指导的分层以及治疗组的添加或删除，会导致进一步的不确定性。对这些复杂的设计进行建模并评估其可行性的定量方法可以帮助避免昂贵和耗时的不可行试验，同时确定值得自信投资的有希望的情况。该项目将首先着手了解CID的关键属性与其可行性，并与试验管理人员，方法学家和统计学家进行协商。然后，将对模拟试验过程和结局的适当方法进行审查，并将其扩展到CID病例，然后应用于肿瘤学CID的两个真实的试验病例研究。然后将解决三个方法论问题：a）我们应该如何使用CID的模型来通知他们的设计，（例如停止/前进）和复杂（例如试验过程优化）试验设计决策; B）我们应如何使用早期、中试或历史试验的数据来校准CID试验模型，从而改进其预测;以及c）我们如何分析CID模型，以更好地了解早期/试点工作应针对的不确定性的关键领域。尽管考虑到对预测和决策的关注，贝叶斯框架可能特别合适，但也可以探索频率论方法。CID通常需要模拟来确定其统计特性，这一事实表明，所开发的方法在计算上要求很高，因此，该项目将以实施和传播高效、免费提供和记录良好的软件包为重点，这些软件包将促进其在实践中的应用。

A wide variety of complex and innovative designs (CIDs), which move beyond the standard randomised controlled trial, have been proposed recently. These designs have the potential to make clinical trials faster and more efficient, thereby increasing the speed of drug development and delivering better treatments to patients sooner. The novelty of these designs, however, has led to substantial uncertainty about how feasible they are to run in practice. Perennial concerns about poor recruitment, adherence and follow-up in clinical trials may be further amplified when using complex designs, while features specific to CIDs such as interim analyses, multiple randomisations, biomarker guided stratification, and the adding or removal of treatment arms, contribute further uncertainty. A quantitative approach to modelling these complex designs and assessing their feasibility could help avoid expensive and time-consuming infeasible trials whilst identifying promising situations which warrant confident investment. This project will initially set out to understand the key attributes of a CID in relation to its feasibility, in consultation with trial managers, methodologists, and statisticians. Appropriate methods for modelling trial processes and outcomes will then be reviewed and extended to the case of CIDs, before being applied to two real trial case studies of CIDs in oncology. Three methodological questions will then be addressed: a) how should we use models of CIDs to inform their design, in terms of both simple (e.g. stop/go) and complex (e.g. trial process optimisation) trial design decisions; b) how should we use data from early phase, pilot, or historical trials to calibrate CID trial models and thereby improve their predictions; and c) how can we analyse CID models to better understand the key areas of uncertainty which should be targeted in early-phase / pilot work. Although a Bayesian framework may be particularly appropriate given the focus on prediction and decision making, a frequentist approach may also be explored. The fact that CIDs typically require simulation to determine their statistical properties suggests that the developed methods will be computationally demanding, and so the project will be underpinned by a focus on the implementation and dissemination of efficient, freely-available and well-documented software packages which will facilitate their application in practice.