Developing guidance for design and conduct of cluster randomised trials
Lead Research Organisation:
University of Birmingham
Department Name: Institute of Applied Health Research
Abstract
Why this research is needed:
Researchers constantly look for ways to improve patient's health. Before new treatments are introduced, they need to be tested. Researchers need to make sure they conduct these testing studies carefully. One particular type of study is called the cluster randomised trial. In this sort of study groups (called clusters) rather than individual patients are randomised to different treatments. These trials require large investments not only in terms of funding and resources but also the investment from the actual research participants. Whilst the research community has had many years to learn how to conduct patient randomised trials very well, they have a lot less experience in conducting cluster trials. Cluster trials are very different to conventional patient randomised trials. Unfortunately, many cluster trials are conducted in such a way that their results might be biased.
What this research aims to do:
One common source of bias occurs because recruitment of participants often happens after randomisation. For example, perhaps the experimental intervention is a new form of exercise class for patients who are over-weight. In those primary care centres randomised to the new classes, if potential research participants know they will be offered these classes many may be interested in participating. Whereas in those primary care practices where the new class is not offered, they may be no perceived incentive to participate in the research. At the end of the trial researchers would not be able to identify if any loss in weight was due to the new class, or the differences in the patients recruited. This sort of bias does not happen in patient randomised trials, because researchers understood that it was important to prevent this.
They are many other types of biases that have an impact on cluster trials. Other risks are more technical in nature and mostly relate to the mathematical models used to estimate treatment effects. These models only work well when there are a large number of clusters (>40) yet the average number of clusters in cluster trials is just 30 and many have a lot fewer than this. These technicalities mean that the estimates of how well the treatments work from these trials are likely to be poor reflections of the truth.
However, the understanding and ability to mitigate these risks of bias already exists. For example, to overcome the issues of recruitment differences between the two groups, the recruitment should be by someone who does not know what treatment the cluster has been allocated to. For example, in the exercise class trial, the patients shouldn't be told about whether they will get the exercise class until after they have decided if they want to participate in the trial. For the more technical bias, there are also solutions, more mathematical in nature. Unfortunately, this knowledge about these solutions is not widely known or available among those who implement these study designs.
How we will do the research
We propose to produce guidance so that cluster trials can be implemented to a much higher standard. We are a group of researchers who have learnt by experience, having helped to conduct many cluster trials, sometimes making mistakes but learning from those mistakes. We will also enlist the help of others - by ascertaining the views of other people who have practical experience in running cluster trials and also from scientists who understand the more technical issues. We will also review the literature to make sure the guidance is current.
What will the research mean?
Doing any study costs money, and if we know more about the study and the best ways to use it, the study will be more effective. Our guidelines and recommendations will make sure that researchers run cluster trials in the best way. Researchers will be more confident in the results, which will mean that people who make decisions on care will be more confident in using the results.
Researchers constantly look for ways to improve patient's health. Before new treatments are introduced, they need to be tested. Researchers need to make sure they conduct these testing studies carefully. One particular type of study is called the cluster randomised trial. In this sort of study groups (called clusters) rather than individual patients are randomised to different treatments. These trials require large investments not only in terms of funding and resources but also the investment from the actual research participants. Whilst the research community has had many years to learn how to conduct patient randomised trials very well, they have a lot less experience in conducting cluster trials. Cluster trials are very different to conventional patient randomised trials. Unfortunately, many cluster trials are conducted in such a way that their results might be biased.
What this research aims to do:
One common source of bias occurs because recruitment of participants often happens after randomisation. For example, perhaps the experimental intervention is a new form of exercise class for patients who are over-weight. In those primary care centres randomised to the new classes, if potential research participants know they will be offered these classes many may be interested in participating. Whereas in those primary care practices where the new class is not offered, they may be no perceived incentive to participate in the research. At the end of the trial researchers would not be able to identify if any loss in weight was due to the new class, or the differences in the patients recruited. This sort of bias does not happen in patient randomised trials, because researchers understood that it was important to prevent this.
They are many other types of biases that have an impact on cluster trials. Other risks are more technical in nature and mostly relate to the mathematical models used to estimate treatment effects. These models only work well when there are a large number of clusters (>40) yet the average number of clusters in cluster trials is just 30 and many have a lot fewer than this. These technicalities mean that the estimates of how well the treatments work from these trials are likely to be poor reflections of the truth.
However, the understanding and ability to mitigate these risks of bias already exists. For example, to overcome the issues of recruitment differences between the two groups, the recruitment should be by someone who does not know what treatment the cluster has been allocated to. For example, in the exercise class trial, the patients shouldn't be told about whether they will get the exercise class until after they have decided if they want to participate in the trial. For the more technical bias, there are also solutions, more mathematical in nature. Unfortunately, this knowledge about these solutions is not widely known or available among those who implement these study designs.
How we will do the research
We propose to produce guidance so that cluster trials can be implemented to a much higher standard. We are a group of researchers who have learnt by experience, having helped to conduct many cluster trials, sometimes making mistakes but learning from those mistakes. We will also enlist the help of others - by ascertaining the views of other people who have practical experience in running cluster trials and also from scientists who understand the more technical issues. We will also review the literature to make sure the guidance is current.
What will the research mean?
Doing any study costs money, and if we know more about the study and the best ways to use it, the study will be more effective. Our guidelines and recommendations will make sure that researchers run cluster trials in the best way. Researchers will be more confident in the results, which will mean that people who make decisions on care will be more confident in using the results.
Technical Summary
Cluster randomised trials are a firmly established alternative to individually randomised trials. It is well known that the CRT increases sample size, but it is less appreciated that the design is at greater risk of biases compared to the individually randomised design. There is a growing body of methodological work on how to mitigate these risks. We argue that there is no specific and comprehensive guidance available to help design and conduct CRTs so they are free or low risk of bias; and no guidance on when the risk of bias in a CRT would be so great that an individually randomised trial would be preferable.
The overarching aim of this proposal is to develop guidance on when CRTs are appropriately justified and to improve the conduct of CRTs to reduce their risk of bias. The broad aims of these linked guidelines are thus to enable researchers to determine:
When is a CRT an appropriate design choice and when is an individually randomised trial preferable?
What is the appropriate choice of cluster to maximise efficiency and minimise contamination?
How should a CRT be conducted to mitigate biases, especially due to identification and recruitment biases?
How should randomisation and analysis be performed in a CRT to ensure inferences are free from bias?
This guidance will be developed by conducting a review of the methodological literature (capitalising on novel methods) to inform the development of a consensus guidance document, developed iteratively using a Delphi process. The PI and three CIs are arguably world leaders in the methodology of cluster randomised trials, as well as having considerable applied experience. All have extensive international collaborations and through these collaborations the members of the steering committee are expected to consist of many of the world's leading cluster trialists. The consensus guidelines will be accessible to applied researchers and will include key guidance as well as exploration and elaboration.
The overarching aim of this proposal is to develop guidance on when CRTs are appropriately justified and to improve the conduct of CRTs to reduce their risk of bias. The broad aims of these linked guidelines are thus to enable researchers to determine:
When is a CRT an appropriate design choice and when is an individually randomised trial preferable?
What is the appropriate choice of cluster to maximise efficiency and minimise contamination?
How should a CRT be conducted to mitigate biases, especially due to identification and recruitment biases?
How should randomisation and analysis be performed in a CRT to ensure inferences are free from bias?
This guidance will be developed by conducting a review of the methodological literature (capitalising on novel methods) to inform the development of a consensus guidance document, developed iteratively using a Delphi process. The PI and three CIs are arguably world leaders in the methodology of cluster randomised trials, as well as having considerable applied experience. All have extensive international collaborations and through these collaborations the members of the steering committee are expected to consist of many of the world's leading cluster trialists. The consensus guidelines will be accessible to applied researchers and will include key guidance as well as exploration and elaboration.
Organisations
Publications
Blackwood B
(2022)
Co-ordinated multidisciplinary intervention to reduce time to successful extubation for children on mechanical ventilation: the SANDWICH cluster stepped-wedge RCT.
in Health technology assessment (Winchester, England)
Hemming K
(2023)
Commentary: Estimands in cluster trials: thinking carefully about the target of inferenceand the consequences for analysis choice.
in International journal of epidemiology