Rapid Responses to:

LETTERS: Don Nutbeam Oakley's case for using randomised controlled trials is misleading BMJ 1999; 318: 944b [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

Think randomisation in community trials -despite the problems

Judith Lumley   (30 April 1999)

Think randomisation in community trials -despite the problems 30 April 1999
Judith Lumley Send response to journal: Re: Think randomisation in community trials -despite the problems	Dear Editor Professor Nutbeam, in his letter of April 3rd, makes important points about contemporary health promotion programmes. They often involve whole populations or population subgroups as the units of interventions. The interventions themselves have multiple components, some of which are difficult to measure. They are likely to take time to affect health and social outcomes. The question remains: do these factors really preclude randomisation? We are currently in the throes of a community intervention programme (PRISM: Program of Resources, Information and Support for Mothers). The population of interest is women in the year after birth - the aim is to reduce their physical and emotional health problems, identified in recent years as substantial1-3 - and the unit of intervention is whole local communities (local municipalities). The strategies are multiple: there are components with the two major groups of primary care providers; specific programme elements for recent mothers and the opportunity for ownership, development and adaptation through a local community development officer and local steering committee. The process evaluation is equally complex and sophisticated with the use of a variety of qualitative and quantitative methods. There is a lengthy lead time for the programme to be implemented and shown to be in place before health outcomes and changes in health services can be expected and measured. PRISM is a cluster-randomised trial. There were 33 communities, metropolitan and rural, eligible (by size criteria) for participation in the State of Victoria. All were given written information and offered a briefing about the project. After this process 21/33 signed a Memorandum of Understanding with the University expressing their willingness to take part in PRISM including agreement to be randomised to intervention or comparison. Contiguous boundaries were taken into account in defining sets of possible pairs before randomisation. Explicit discussion of the evaluation of health and social programmes - including aspects of trial design and the implications of randomisation - with the potential partners in local government was an essential part of the project's development. What the briefings also drew to our attention was the very real possibility of selection bias into intervention or comparison status if randomisation had not been used. Some communities were easier to reach, or had more resources, or seemed more friendly, or were more keen to be involved and it would have been difficult to remain uninfluenced by all those factors without an independent process of allocation within all the possible sets of pairs, stratified for geographic size, number of births and grading of recent and current community activity. Though the complexities of community interventions and their appraisal are definitely daunting those issues apply equally in randomised and non-randomised designs. We would encourage other groups to consider cluster designs and explicit discussion of randomisation and its implications with community collaborators and partners. Yours sincerely Judith Lumley Professor/Director References 1 MacArthur C, Knox EG. Health after childbirth; an investigation of long term health problems beginning after childbirth in 11,701 women. London: HMSO, 1991. 2 Glazener C, Abdella M, Stroud P, Naji S, Templeton A, Russell I. Postnatal maternal morbidity: extent, causes, prevention and treatment. Br J Obstet Gynaecol 1995; 102: 282-7. 3 Brown S, Lumley J. Maternal health after childbirth: results of an Australian population-based survey. Br J Obstet Gynaecol 1998: 105; 156- 161.