Complex interventions: how “out of control” can a randomised controlled trial be?BMJ 2004; 328 doi: http://dx.doi.org/10.1136/bmj.328.7455.1561 (Published 24 June 2004) Cite this as: BMJ 2004;328:1561
- 1Department of Community Health Sciences, University of Calgary, 3330 Hospital Drive NW, Calgary T2N 4N1, Alberta, Canada
- Correspondence to: P Hawe
- Accepted 24 March 2004
Complex interventions are more than the sum of their parts, and interventions need to be better theorised to reflect this
Many people think that standardisation and randomised controlled trials go hand in hand. Having an intervention look the same as possible in different places is thought to be paramount. But this may be why some community interventions have had weak effects. We propose a radical departure from the way large scale interventions are typically conceptualised. This could liberate interventions to be responsive to local context and potentially more effective while still allowing meaningful evaluation in controlled designs. The key lies in looking past the simple elements of a system to embrace complex system functions and processes.
The suitability of cluster randomised trials for evaluating interventions directed at whole communities or organisations remains vexed.1 It need not be.2 Some health promotion advocates (including the WHO European working group on health promotion evaluation) believe randomised controlled trials are inappropriate because of the perceived requirement for interventions in different sites to be standardised or look the same.1 3 4 They have abandoned randomised trials because they think context level adaptation, which is essential for interventions to work, is precluded by trial designs. An example of context level adaptation might be adjusting educational materials to suit various local learning styles and literacy levels.
Lead thinkers in complex interventions, such as the UK's Medical Research Council, also think that trials of complex interventions must “consistently provide as close to the same intervention as possible” by “standardising the content and delivery of the intervention.”5 By contrast, however, they do not see this as a reason to reject randomised controlled trials.
These divergent views have led to problems on two fronts. Firstly, the field of health promotion is being turned away from randomised controlled trials.1 3 4 This could have heavy consequences for the future accumulation of high quality evidence about prevention. Secondly, when trials with organisations and whole communities do go ahead, the story is consistently becoming one of expensive failure—that is, weak or non-significant findings at huge cost.6–8 Could one of the reasons for the interventions not working be that the components have been overly standardised?
Something has to change. The current view about standardisation is at odds with the notion of complex systems. We believe that an alternative way to view standardisation could allow state of the art interventions (and ones that might look different in different sites) to be more effective and to be meaningfully evaluated in a randomised controlled trial. First, however, we have to re-examine our understanding of the term complex intervention.
What is a complex intervention?
The MRC document A Framework for the Development and Evaluation of Randomised Controlled Trials for Complex Interventions argues that “the greater the difficulty in defining precisely what exactly are the ‘active ingredients' of an intervention and how they relate to each other, the greater the likelihood that you are dealing with a complex intervention.”5 The document gives examples of complex interventions from the setting up of new healthcare teams, to interventions to get treatment guidelines adopted, to whole community education interventions. Setting aside the problem that this definition is also consistent with a poorly thought through intervention, we believe that the field could benefit by delving further into complexity science.
Complexity is defined as “a scientific theory which asserts that some systems display behavioral phenomena that are completely inexplicable by any conventional analysis of the systems' constituent parts.”9 Reducing a complex system to its component parts amounts to “irretrievable loss of what makes it a system.” 9 Those of us who have decomposed interventions into components for process evaluation might feel uncomfortable at this point. Yes, we may have been able to describe an intervention, say, simply in terms of the percentage of general practitioners who attend the training workshops and the percentage of patients who report having read the leaflets. Thinking about process evaluation in this way is the norm.10 11 But by doing so, have we really captured the essence of the intervention? We have, if all we think our intervention to be is the sum of the parts. But that is not, by definition, a complex intervention. It remains a simple one.
Standardising complex interventions
So, could a controlled trial design (which requires something to be replicable and recognisable as the intervention in each site) ever be appropriate to evaluate a (truly) complex intervention? The answer is yes. The crucial point lies in “what” is standardised. Rather than defining the components of the intervention as standard—for example, the information kit, the counselling intervention, the workshops—what should be defined as standard are the steps in the change process that the elements are purporting to facilitate or the key functions that they are meant to have. For example, “workshops for general practitioners” are better regarded as mechanisms to engage general practitioners in organisational change or train them in a particular skill. These mechanisms could then take on different forms according to local context, while achieving the same objective. 12 (table).
Defining integrity of interventions
With most (simple) interventions, integrity is defined as having the “dose” delivered at an optimal level and in the same way in each site.10 Complex intervention thinking defines integrity of interventions differently. The issue is to allow the form to be adapted while standardising the process and function. Some precedents exist here. For example, Mullen and colleagues conducted a meta-analysis of 500 patient education trials and showed that interventions were more likely to be effective if they met particular criteria fitting with behavioural change theory—for example, being tailored to the patient's individual learning needs or being set up to provide feedback about a patient's progress.17 The indicators of quality were driven by theory and concerned the functions provided by the key elements of the intervention rather than the elements themselves (such as a video).
Context level adaptation does not have to mean that the integrity of what is being evaluated across multiple sites is lost. Integrity defined functionally, rather than compositionally, is the key.
Real world contexts
We are not the first to think this way. In school health, Durlak discussed non-standard interventions that “cannot be compartmentalised into a predetermined number and sequence of activities.”18 This sounds like complex interventions. Characterised by activities like capacity building and organisational change, these interventions have specific, theory driven principles that ensure that non-standard interventions (different forms in different contexts) conform to standard processes. They are still evaluable by randomised controlled trials. Indeed, a randomised controlled trial of such an intervention (which is “out of control” to some ways of thinking) might be exactly what is required to provide more convincing evidence that community development interventions are effective.
More studies of this type would help to reverse the current evidence imbalance when policy makers weigh up “best buys” in health promotion. At present they often have to compare traditional areas like asthma education (which usually come with randomised controlled trial evidence) with community development (which is usually supported only with case study evidence).19 The more conservative, patient targeted interventions backed by randomised controlled trials generally win hands down.19
Rethinking ways to use the intervention-context interaction to maximum effect may make complex interventions stronger. The MRC document on complex intervention trials calls for standardisation but also recognises the need in the exploratory phase to “describe the constant and variable components of a replicable intervention.”5 But it does not say how to make this distinction.
An alternative way of thinking about standardisation may help. The fixed aspects of the intervention are the essential functions. The variable aspect is their form in different contexts. In this way an intervention evaluated in a pragmatic, effectiveness, or real world trial would not be defined haphazardly, as it sometimes is now,20 as the default option for whenever researchers were not able to accomplish the standardised components that they idealised. Instead, with lateral thinking, theorising about the real world context would become the ideal,21 22reversing current custom.23 That is, instead of mimicking trial phases which assume that the “best” or the “ideal” comes from the laboratory and gets progressively compromised in real world applications, community trial design would start by trying to understand communities themselves as complex systems and how the health problem or phenomena of interest is recurrently produced by that system.
Standardisation has been taken to mean that all the components of an intervention are the same in different sites
This definition treats a potentially complex intervention as a simple one
In complex interventions, the function and process of the intervention should be standardised not the components themselves
This allows the form to be tailored to local conditions and could improve effectiveness
Intervention integrity would be defined as evidence of fit with the theory or principles of the hypothesised change process
The shackles of simple intervention thinking may prove hard to throw off. Although an intervention may be described as complex, the signs of simple intervention thinking will be apparent in how the intervention is described and whether integrity is tied to the extent to which certain standardised forms are present. Investigators should justify the approach they take with interventions—that is, whether interventions are theorised as simple or complex. Complex systems rhetoric should not become an excuse to mean “anything goes.” More critical interrogation of intervention logic may build stronger, more effective interventions.
Contributors and sources All authors were collaborators in a cluster randomised intervention trial in maternal health promotion.14 All are participating in a newly funded international collaboration on complex interventions funded by the Canadian Institutes of Health Research. PH drafted the original idea for the paper based on experience and conversations with TR and AS. All contributed to developing the idea and writing the paper.
Funding PH and AS are senior scholars of the Alberta Heritage Foundation for Medical Research. PH is also supported by an endowment as Markin Chair in Health and Society at the University of Calgary.
Competing interests None declared.