In randomised clinical trials, patients typically undergo some screening stages before randomisation. These stages complicate design as patients may be randomised after admissions to screening have been halted. Stopping the admission of new patients too late may lead to over-enrolment, as more patients than planned are randomised, increasing costs, unnecessary patient exposure and the time required to complete recruitment. Stopping too early may result in failure to achieve the planned sample size. Predictions of the stopping time should therefore account for patients currently in screening and the number of the patients to be randomised. We present a novel statistical framework for joint modelling screening and randomisation and determining an adaptive stopping time to stop admitting new patients to screening before the planned randomisation target is reached. The aim is to reduce over-enrolment while maintaining a high probability of achieving the target. We represent the screening and randomisation pathway as a multistage process with stage-specific durations and transition probabilities estimated from trial data. The model yields the distribution of the number of patients who will be randomised from the current screening pipeline and the probability of reaching the planned randomisation target. If this probability exceeds a specified threshold, such as 95% or 99%, admission of new patients to screening should be stopped. Otherwise, the model is updated as additional data accrue and is used to forecast future pipeline and the stopping time. The framework can incorporate scheduled site initiations and future closures using evolving stochastic processes, and supporting forecasts at global, regional and country levels. Illustrative simulations show that the approach can reduce over-enrolment substantially while preserving a high probability of reaching the target sample size.

翻译：暂无翻译