Consort 2010 statement: extension to cluster randomised trialsBMJ 2012; 345 doi: https://doi.org/10.1136/bmj.e5661 (Published 04 September 2012) Cite this as: BMJ 2012;345:e5661
All rapid responses
I would like to emphasise using two examples that application of intention-to-treat analysis (ITT) is quite tricky in the context of a cluster randomised trials (CRT) and therefore should be more discussed in the literature. It can prevent bias but can also produce something worse: results that do not correspond with the objectives of the study.
First: A CRT in which randomisation units also include individuals who are not from the target population (the population about which inferences are to be made). For example, the study reported by Kirkwood et al 1 aimed to assess the effect of vitamin A supplementation on mortality rate in women aged 15-45 years. Randomisation units were compounds with individuals of different ages and both sexes, but only women aged 15-45 years were assigned to receive or not receive the intervention. Rightly, the authors performed ITT, comparing mortality rate restricted to the target population regardless of compliance with vitamin A. But if the authors had decided to perform ITT in the strict sense including all randomised individuals, they would compare mortality rates in the entire population of the compounds, regardless of age and sex. However this analysis would not estimate the effect of vitamin A in the target population.
Second: A CRT in which randomisation units in which all individuals are from the target population. The ongoing CRT on influenza vaccine in children, which takes households as randomisation units and aims to assess the direct and indirect effects of vaccination in children aged 6 months to 10 years2. Only these children will be assigned to receive vaccine but all the other people in the household are also in the target population. The indirect effect will be assessed by comparing the occurrence of disease in all older individuals in households with vaccinated children to those in households with non-vaccinated children, following a strict ITT. It is not described, but it is possible that an ITT “once randomized, always analyzed” in this case would be justifiable, comparing rates between households with children assigned to be vaccinated and households with children assigned to control group, including both children and adults. This would correspond to the overall effect of vaccination in the study population when only a fraction is vaccinated3.
In conclusion, the formula “once randomized, always analyzed” is too simplistic and the decision on ITT should be made carefully. More attention should be paid to the target population, otherwise the results of an ITT would not make sense.
Sergio S. Cunha
Universidade Federal de Pernambuco, Brazil
1. Kirkwood BR, Hurt L, Amenga-Etego S, Tawiah C, Zandoh C, Danso S, et al. Effect of vitamin A supplementation in women of reproductive age on maternal survival in Ghana (ObaapaVitA): a cluster-randomised, placebo-controlled trial. Lancet 2010;375(9726):1640-9.
2. Sullender W, Fowler K, Krishnan A, Gupta V, Moulton LH, Lafond K, et al. Design and initiation of a study to assess the direct and indirect effects of influenza vaccine given to children in rural India. Vaccine 2012;30(35):5235-9.
3. Hanquet G, Valenciano M, Simondon F, Moren A. Vaccine effects and impact of vaccination programmes in post-licensure studies. Vaccine 2013;31(48):5634-42.
Competing interests: No competing interests