1. Author's Reply

    Dear Editor

    Dr Parienti and colleagues have raised concerns [1] about a section of the commentary for which I am responsible. Their letter highlights the lack of a standard typology for trial design in clinical epidemiology. At the heart of this issue is not only what to call the different types of clinical trial, but how to analyse the results of each trial. Was the trial by Dr Parienti and colleagues a cluster randomised controlled trial (RCT) as I described it? In the trial each service (or study centre) was randomised to an alternating sequence, commencing with either the alcohol handrub first, or handscrubbing first. The trial could have been described as a cross-over design, although in cross-over designs, it is usually an individual that is randomised to an order of treatment.[2] Where the unit of randomisation is an intact social unit, such as a community, family, or service, the trial is usually called a cluster RCT.[3] Dr Parienti quite rightly points out that in cluster RCTs the study centre is normally randomised to the treatment or control, whereas he and his colleagues randomised each service to a sequence of interventions. Therefore, perhaps the trial is most accurately described as a cluster cross-over RCT.

    The substantive issue in Dr Parienti's letter is how their trial should have been analysed. It is my view that the trial should have been analysed as though it were a cluster design, as services were randomised to a sequence of interventions. Cluster designs can balance out the impact of extraneous variables, but less efficiently than trials that randomise by individual. Cluster designs normally adjust for this by inflating the number of clusters and participants that need to be enrolled in the trial.[4] Dr Parienti and colleagues do not seem to have done this, and there were only six services (or clusters) in their study. As Dr Parienti states, each service would have acted as its own control, and the 15 sequential cross-over periods certainly adds to the study's validity. I was perhaps remiss in not pointing this out in the commentary.

    Even though the cross-over periods controlled for between-cluster variance, the investigators did not analyse for the impact of time periods. Each service has a flow of patients and staff through it that will vary in some way - for instance there will be seasonal variation, which even if small, could influence the outcome. Therefore the services will not be the same over time. If the patients or the cross-over periods had been randomised, then the possible influence of this effect would have been balanced out. Randomising the service to an sequence of intervention would not have been able to control for this effect; and thus the trial's standard error could actually be greater, confidence intervals wider, and the p values larger than the reported result.[5] If a multi-level analysis (patient, time period and service) had been conducted then the reader could be reasonably assured of the lack of difference between the interventions. It is possible that even had Dr Parienti and colleagues used this method of analysis, they would have found little difference from the result they published. However, Dr Parienti and colleagues did not follow the dictum "as you randomise, so you shall analyse" and the knowledgeable reader is left uncertain as to whether handrubbing with alcohol as compared with handscrubbing is as equivalent as the study reports.

    I am grateful to Dr David Torgerson, University of York, and Mark Jones, University of Auckland, for their helpful review.


    (1) Jean-Jacques Parienti, for the ACM study group. Only cluster design lead to cluster effect [electronic response to Moralejo D and Jull D Handrubbing with an aqueous alcohol solution was as effective as handscrubbing with antiseptic soap for preventing surgical site infections] 2003

    (2) Jadad AR. Randomised controlled trials. London: BMJ Books,1998.

    (3) Donner A, Klar N. Design and analysis of cluster randomization trials in health research. London: Arnold, 2000.

    (4) Kerry SM, Bland JM. Analysis of trials randomised in clusters. BMJ 1998;316:54.

    (5) Bland JM, Kerry SM. Trials randomised in clusters. BMJ 1997;315:600.

    Submit response
  2. Only cluster design lead to cluster effect

    Dear Editor

    On behalf our Study group, I would like to thank EBN’s Editors for their interest in our work.

    In their review, major concerns regarding the design and analysis of our study were raised by Drs Moralejo and Jull, mainly because antiseptic protocols were randomised by services (called "clusters") but outcome was analysed by patients. The authors concluded that while our trial is intriguing, "whether hand rubs and handwashing are truly equivalent remains unclear". In my opinion, the fact that Drs Moralejo and Jull misunderstood the design of our trial does not justify that they unvalidate our conclusion.

    The reason why we did not account for cluster randomisation is simple: the design of our study was not a "cluster randomised trial", as they incorrectly suggested. To be the case, each of the 6 services should have been randomly assigned to one of the 2 protocols for the complete 16-month period of the study. In this case only, characteristics of each service, such as teaching versus non-teaching hospitals or surgical team experience, would have differentially affected outcome, because of intracluster dependence between patients within each service.

    In fact, each service alternated protocols monthly, so that they equally contributed to include patients in both protocols, as shown in Table 1 of our article. Initial randomisation of the services rather than continuous randomisation of patients was not an error but a choice. Its rational was clearly discussed in our article.

    Finally, we must stand on our conclusion that handrubbing and handscrubbing were equivalent in preventing surgical site infection.

    Submit response
« Parent article

Free Sample

This recent issue is free to all users to allow everyone the opportunity to see the full scope and typical content of EBN.
View free sample issue >>

EBN Journal Chat

The EBN Journal Chat offers readers the opportunity to participate in discussion about research articles and commentaries from Evidence Based Nursing (EBN).

How to participate >>

Don't forget to sign up for content alerts so you keep up to date with all the articles as they are published.