In 1948 the UK Medical Research Council’s streptomycin trial established the principles of the modern clinical trial, and since then randomised controlled trials have offered a gold standard for the evaluation of new medical treatments. The parallel groups design used in that first trial has proved to be a remarkably enduring and powerful scheme, but variations and innovations in trial design continue to be explored. Innovation may be needed to tackle practical constraints on trial conduct, but it can also allow us to improve the efficiency of a trial – that is to reduce the number of participants or other resources needed to gather the evidence we need. PCTU has a flourishing methodological research programme centred on innovative trial design.
Specific topics of our work include:
Stepped wedge cluster randomised trials
Stepped wedge trials are cluster randomised trials in which the clusters are sampled repeatedly over time, with different clusters crossing over from the control to the experimental intervention at different times. They can have practical advantages over parallel group trials, but they can also be more efficient in some circumstances. A much fuller understanding has emerged in the last couple of years of the theoretically optimal scheme for a stepped wedge design, but there are still many challenges in the implementation, analysis and reporting of stepped wedge trials that we are keen to investigate. We are also interested in applying what we learn from stepped wedge trials to the design and analysis of observational studies that involve multiple baselines and repeated assessments.
If individual participants are recruited at a cost (either a financial or an ethical one) then it may not be beneficial to keep recruiting for the full duration at every cluster. Instead an incomplete design – one with gaps in the recruitment schedule – might be preferred. We have been studying the properties of a novel incomplete design called the dog-leg. Though initially proposed for individually randomised trials it is likely to have wider application in cluster randomised trials: among two-stage “crossforward” designs (where clusters are only allowed to cross between interventions in one direction, from the control to the experimental intervention) the dog-leg design will often require the fewest participants. Incomplete designs for cluster randomised trials with more than two stages merit further investigation.
Individually randomised stepped wedge designs
In a cluster randomised trial a stepped wedge design can be more efficient than a parallel groups design. Could a stepped wedge also improve the efficiency of an individually randomised trial? We have identified situations where this is the case, and we are continuing to study how the lessons learned from longitudinal cluster randomised trial design can be taken back to individually randomised trials.
Re-randomisation designs are individually randomised trials in which participants can be re-enrolled and re-randomised each time they experience a new treatment episode. For example, in a trial comparing two different interventions during pregnancy, participants could be re-randomised for each new pregnancy. This design can facilitate a more rapid recruitment rate, and could reduce trial costs.
A versatile approach to sample size calculation using simulation
Ensuring the number of participants in a clinical trial is no larger than it needs to be is an important ethical principle. But new developments and increasing complexity in trial design can mean that accurate formulae for calculating sample size are not available (indeed with some adaptive trial designs analytical solutions to sample size calculation problems are simply unattainable). Monte Carlo simulation can help in this case. Simulation has a rough-and-ready reputation, and some statisticians may favour the elegance of an equation even when only approximate, but simulation offers a simple, precise, and versatile approach to sample size calculation that is easily validated by others such as statistical reviewers. We have developed a sample size calculation software add-on for Stata called SimSam that can calculate the sample size needed to achieve given statistical power for any analysis under any data generating model that can be programmed in the host package.
References to our published papers include:
- Hooper R, Bourke L. Cluster randomised trials with repeated cross-sections: alternatives to parallel group designs. BMJ 2015;350:h2935.
- Hooper R. Versatile sample size calculation using simulation.Stata Journal 2013;13(1):21-38
- Hooper R & Bourke L. The dog-leg: an alternative to a cross-over design for pragmatic clinical trials in relatively stable populations. Int J Epidemiol 2014;43(3):930-936.
- Kahan BC et al. A re-randomisation design for clinical trials. BMC medical research methodology. 2015; 15:96