CONSORT 2010 extension: setting standards for reporting cluster randomised crossover trials

Over the past two decades my research has centred on systematic reviews. Developing and evaluating statistical and research methodology for reviews has been a focus, but I have also collaborated on many reviews. These collaborations inevitably reveal the importance of complete, clear, and accurate reporting of primary studies. At best, incomplete and inaccurate reporting wastes research investment—often substantial in the case of large randomised trials—and at worst, it may lead to incorrect conclusions with far reaching consequences.

Primary studies that are reported well contribute maximally to the evidence base and can be fairly assessed. The opposite is true when reporting is unclear or incomplete. I have spent countless hours scouring trial reports for information that should be present but is not; trying to decipher unclear text; or trying to decide what to do when the information reported is inconsistent within or across reports of the same study. This time adds up and takes away opportunities for other research. When multiplied across the thousands of systematic reviews conducted each year, this is an unconscionable waste of research effort.

More importantly, incomplete and inaccurate reporting inevitably leads to studies not contributing to an evidence base in the way they should. These studies might be excluded from a systematic review because researchers cannot determine whether the study meets the eligibility criteria, excluded from meta-analyses because the required statistics are not reported, or inaccurately judged to be at a particular risk of bias, and so on.

Our research investigating the reporting quality of a particular design—the cluster randomised crossover trial—found that incomplete reporting was common,12 motivating the development of an extension to the CONSORT (consolidated standards of reporting trials) 2010 statement.34 Reporting guidelines such as CONSORT are a way to circumvent incomplete reporting by providing recommendations of what should be reported so that users can fully interpret and reuse the findings in derivative products such as systematic reviews.

Cluster randomised crossover (CRXO) trials, as the name suggests, share features of the parallel group cluster and individual crossover designs. In a CRXO trial, groups of individuals (known as “clusters”) are randomised to sequences of treatments. For example, in a trial that evaluated whether head positioning of patients after acute stroke affected disability at 90 days, hospitals were randomised to implement either the lying-flat position in the first period and the sitting-up position in the second period (sequence 1), or the sitting-up position in the first period and the lying-flat position in the second period (sequence 2).5

The CRXO design can largely overcome the loss in power due to clustering, which means it can be used to investigate interventions where effects are expected to be small but important.678 Relatedly, using the CRXO design may make a randomised trial feasible in circumstances where a parallel group design would not be feasible because there are too few clusters available to detect the target difference, such as when there are a limited number of intensive care units in a country.89 Despite the advantages of the CRXO design, there are methodological complexities that can introduce bias. Careful reporting of the design, informed by tailored reporting guidance, is therefore imperative to ensure the risk of bias can be accurately assessed.

In our BMJ article,4 we provide reporting recommendations for CRXO trials developed through a consensus process, including a survey and consensus meeting with 55 people. The consensus process was strengthened by the involvement of statisticians with extensive knowledge of the CRXO design. The process resulted in consolidated reporting recommendations across the CONSORT 2010 statement and relevant extensions, modified items, and new items. New items on data sharing and patient and public involvement reflect the evolving expectations and requirements of the scientific and wider community.101112 In response to growing awareness of the need for greater clarity about the intervention effects being estimated,1314 we have also incorporated estimands into the statistical methods item.

While setting standards for reporting this design is an important step forward, there is still much to be done. Imminent publication of the CONSORT 2025 statement15 means we will need to align this extension with the updates in the main statement. Such alignment might ultimately be streamlined by adopting collaborative approaches to the coordination and harmonisation of closely related reporting guidelines, such as that being implemented for the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) statement and its extensions.16

Importantly, we also need to consider how we can support researchers in implementing the guidelines. In the short term this includes creating a bank of examples of complete reporting to accompany the items and guidance to access online. We encourage researchers to contribute to the bank of items by submitting examples to us. In the longer term, artificial intelligence could play an important role in helping researchers to assess whether they have completely reported before submission, or for journal editors and peer-reviewers to undertake such checks before publication.17