Statistical methods for interrupted clinical trials

Large-scale randomised controlled clinical trials are an essential part of the evaluation of healthcare treatments including new drugs, surgical techniques and behavioural interventions. Like many other parts of life, ongoing clinical trials have been affected by the global coronavirus pandemic. The cancellation of non-essential medical procedures, restrictions on face-to-face assessments and outpatient non-attendance due to lockdown restrictions, illness or reluctance to visit hospitals or healthcare centres have led to recruitment and data collection being suspended for many ongoing clinical trials.

As restrictions start to be relaxed, researchers have the opportunity to restart clinical trials that were interrupted. The questions of whether or not this is worth doing, or of the best way to analyse the data either in a restarted trial or in one that is not restarted, may raise some challenges, however. This project will research statistical tools to help address these questions. These methods will also be of value in other settings when trials are interrupted due to challenges in recruitment or funding, or due to the influence of new results from other research.

If a trial is restarted, depending on the clinical area in which the trial is being conducted, there may be differences between the pre-pandemic and post-pandemic periods in the type of patients who enrol in the trial, the exact way in which measurements are taken, or even in the intervention to be assessed, for example for a psychological intervention for which delivery may have changed to being wholly or partially online. These differences, or heterogeneity, need to be accounted for in the statistical analysis, and may mean that a larger number of patients than initially anticipated need to be included in the trial in order to obtain a reliable result. We will identify methods for this analysis and evaluate these in the setting of interrupted trials. As it is important that analysis methods proposed are accepted by all stakeholders, we will organise workshops for clinical trialists, clinical trial statisticians and representatives of regulators, funders, science publishers and patients to discuss and hopefully lead to consensus on the most appropriate methodology.

If a trial is not restarted, the number of patients included will be smaller than initially planned. In many cases, particularly those in which patients are followed up in the clinical trial for a long period before the effect of the treatment is finally assessed, some early data may be available for patients recruited shortly before the start of the pandemic. This data may give additional information that can be included in the final analysis. We will explore statistical approaches to best utilise the information available in these data, extending existing methods where this is necessary.

In addition to developing and recommending methods for the analysis of trials that are or are not restarted, we will develop methods to help decide which of these is the best option depending on the amount of information already available and the degree of heterogeneity between pre-pandemic and post-pandemic periods that is anticipated. We will also develop methods that allow an analysis of the data already collected but also allow the option of restarting the trial if the results of the trial are not sufficiently clear. Specialist statistical methods are required for this analysis in order to ensure that the risk of an erroneous false positive trial result is not increased.

The research team includes experts in clinical trial statistics along with trialists and representatives of trial funders from a range of clinical areas to ensure that the research is applicable in a wide range of clinical trial settings.

The SARS-CoV-2 pandemic has led to the disruption of thousands of ongoing clinical trials. For most of these trials a decision needs to be made between restarting the trial, continuing to recruit additional patients, and stopping the trial, analysing the data already observed. This decision and the resulting statistical analysis in either case, raise statistical challenges. These will be addressed by this project, providing methods for trials interrupted by the pandemic and by other causes in the future.

If an interrupted trial restarts, changes in population, assessment or intervention might lead to treatment effect heterogeneity. We will explore data analysis methods allowing for this heterogeneity, for example based on meta-analysis approaches. We will extend Bayesian borrowing methods to allow down-weighting of earlier data depending on the extent to which they are discordant with later data. Simulation studies will assess and compare methods with particular focus on power and type I error rate control. To ensure that methods proposed are accepted as widely as possible we will hold consensus workshops with key stakeholders. We will also explore approaches based on group sequential methods to allow the trial to be restarted only if results from data already collected are unclear.

If a trial is stopped it is desirable for the analysis to make best use of all available data. This may include baseline data, or in trials with long term follow up, early outcome data, for patients for whom the final endpoint is not observed. Methods exist to use these data to improve the precision of the analysis. We will assess these methods, extending them where necessary to allow use of all early data and time to event outcomes.

We will illustrate methods proposed or developed using real case studies and determine the range of settings in which stopping and analysing the trial might be preferable to restarting leading to practical guidance for funders and trialists.