In clinical trials, patients sometimes discontinue study treatments prematurely due to reasons such as adverse events. Treatment discontinuation occurs after the randomisation as an intercurrent event, making causal inference more challenging. The Intention-To-Treat (ITT) analysis provides valid causal estimates of the effect of treatment assignment; still, it does not take into account whether or not patients had to discontinue the treatment prematurely. We propose to deal with the problem of treatment discontinuation using principal stratification, recognised in the ICH E9(R1) addendum as a strategy for handling intercurrent events. Under this approach, we can decompose the overall ITT effect into principal causal effects for groups of patients defined by their potential discontinuation behaviour in continuous time. In this framework, we must consider that discontinuation happening in continuous time generates an infinite number of principal strata and that discontinuation time is not defined for patients who would never discontinue. An additional complication is that discontinuation time and time-to-event outcomes are subject to administrative censoring. We employ a flexible model-based Bayesian approach to deal with such complications. We apply the Bayesian principal stratification framework to analyse synthetic data based on a recent RCT in Oncology, aiming to assess the causal effects of a new investigational drug combined with standard of care vs. standard of care alone on progression-free survival. We simulate data under different assumptions that reflect real situations where patients' behaviour depends on critical baseline covariates. Finally, we highlight how such an approach makes it straightforward to characterise patients' discontinuation behaviour with respect to the available covariates with the help of a simulation study.

翻译：在临床试验中，患者有时会因不良事件等原因提前中止研究治疗。治疗中止作为随机化后发生的并发事件，使得因果推断更具挑战性。意向性治疗（ITT）分析可提供治疗分配效应的有效因果估计，但未考虑患者是否必须提前中止治疗。我们提议采用主分层方法处理治疗中止问题，该方法在ICH E9(R1)附录中被认可为处理并发事件的策略之一。在此框架下，我们可将整体ITT效应分解为针对不同潜在连续时间中止行为患者组的主因果效应。分析中需考虑连续时间上的中止行为会产生无限多个主分层，且对于永不中止的患者，中止时间无法定义。另一个复杂因素是中止时间和时间至事件结局均受行政删失影响。我们采用灵活的基于模型的贝叶斯方法应对这些复杂性。通过应用贝叶斯主分层框架，我们基于近期肿瘤学RCT的合成数据进行分析，旨在评估试验性新药联合标准治疗与单独标准治疗对无进展生存期的因果效应。我们根据不同假设模拟数据，这些假设反映患者行为依赖关键基线协变量的真实情境。最后，通过模拟研究强调该方法如何利用可用协变量直接刻画患者的中止行为特征。