Intended for healthcare professionals

Analysis

Graphical method for depicting randomised trials of complex interventions

BMJ 2007; 334 doi: https://doi.org/10.1136/bmj.39045.396817.68 (Published 18 January 2007) Cite this as: BMJ 2007;334:127
  1. Rafael Perera, senior research fellow in statistics1,
  2. Carl Heneghan, deputy director1,
  3. Patricia Yudkin, reader in medical statistics2
  1. 1Centre for Evidence Based Medicine, Department of Primary Health Care, University of Oxford, Oxford OX3 7LF
  2. 2Department of Primary Health Care, University of Oxford
  1. Correspondence to: pat.yudkin{at}dphpc.ox.ac.uk
  • Accepted 15 November 2006

Making the what, when, and who of non-drug treatments easier to understand would benefit researchers and readers

Complex interventions consist of several separate components combined to produce a desired outcome.1 Evaluation of such interventions in randomised trials will generally lead to complex comparisons between trial groups.2 Moreover, text descriptions in journal articles may obscure aspects of the interventions in the trial and hinder comparison between them. To counter these problems we have produced a single image that presents the components of all interventions in the trial and compares different treatment arms. The aim is to clarify the structure of the contrasted interventions and thus aid interpretation of the trial results.

The need for clear comparisons

We studied 169 randomised trials of non-drug interventions in primary care published between 1999 and 2003. We searched Medline, PSIQInfo, Bioabstracts, and Embase using the free text search terms “randomised controlled trials” and “primary care” and their synonyms, and excluding the term “placebo” appearing in the title or abstract; we also hand searched reference lists of retrieved papers. In many of these papers the interventions were incompletely described. We identified three principal problems: identifying the different components of the intervention, establishing the time at which components were delivered, and defining the differences between intervention arms.

To clarify these aspects we suggest that it would be helpful to depict the experimental and control interventions graphically. The proposed graph is similar to a flowchart, with each treatment arm represented in a specific column, and with all the intervention components presented within that column. The time scale of the trial runs from top to bottom on the left hand side, with the times of randomisation and outcome measurement (or measurements) clearly marked. Each component of an intervention is depicted separately. Components delivered concurrently are displayed side by side, while those delivered consecutively are shown one beneath the other.

We regard components either as objects or activities. Objects are represented by squares (to reflect their fixed nature) and activities by circles (to reflect their flexibility). Different components are labelled with different letters. Below the diagram, a legend gives a brief description of each component, including its form, content, functions, and details of who delivers it. If necessary, additional material can be given in the text. This approach will convey as much information as a written description, will clarify the basic structure of the experimental intervention, and elucidate the differences between treatment arms, as the examples below show.

Intervention intensity and repeated components

Lewin et al reported a parallel arm trial of self management in patients with newly diagnosed angina.3 The experimental intervention introduces the “angina plan,” the objective of which is to allow patients to manage their condition using cognitive behavioural techniques. The intervention is enhanced by nurse support in the form of interviews and telephone calls. The original paper describes the intervention in about 590 words. With the aid of the diagram (fig 1) we can easily recognise that the experimental arm has a much more intensive intervention than the control, with repeated nurse telephone contacts at 1, 4, 8, and 12 weeks after the last contact in the control group. (An alternative way of depicting these repeated telephone contacts would be to use a single circle, and to label the time line with the four times of delivery). The diagram also clearly shows that no component is common to both interventions.

Figure1

Fig 1 Graphical depiction of interventions in a trial of self management in patients with newly diagnosed angina3

Flexible interventions

Jarman et al reported a parallel arm trial of community based nurses specialising in Parkinson's disease.4 The experimental intervention consists of training nurses to specialise in Parkinson's disease and, after clearly specifying their areas of responsibility, requesting them to support patients for the two year trial. The diagram (fig 2) shows the resources needed for the community nurse intervention and also highlights the possible variation in the interventions in both experimental and control arms. In the experimental arm, the nurses are trained before randomisation; after randomisation (baseline) the nurses are given a car, a mobile phone, and a clear description of their areas of responsibility. Moreover, the timing of the intervention is not static; in both the experimental and control arms patient care is given at any point (and potentially at several times) in the two years that the trial lasts.

Figure2

Fig 2 Graphical depiction of interventions in a trial of community based nurses specialising in Parkinson's disease4

Multiple arms for multiple comparisons

Aveyard et al examined the effect on smoking cessation of the pro-change course.5 The trial tested three experimental interventions using the pro-change course with increasing levels of contact (none, telephone call from lay person, and appointment with nurse). The control group received standard support material. The written description of the interventions had 715 words. Our diagram (fig 3) shows immediately the cumulative nature of the experimental interventions. Although each intervention is complex, the comparison between successive interventions is relatively simple, each differing from the last by a single component. The year long interval between the control intervention and trial outcome also stands out, in contrast to the six month interval in the experimental arms.

Figure3

Fig 3 Graphical depiction of interventions in a trial of an expert system and self help manual to aid smoking cessation5

Advantages of using graphs

Graphical depiction of an entire intervention allows its structure to be quickly understood. With the experimental and control interventions placed side by side on the diagram, differences between them—such as in the time elapsing between their delivery and the trial outcome—become obvious.

We believe that the discipline of constructing a diagram will help at the design stage of a trial. By focusing attention on the components of the intervention, it prompts researchers to think through the structure, timing, and contents of each component in detail and to describe the components adequately. The exercise should help to ensure that the control intervention has been adequately considered and described and that the difference between the experimental arm and the control arm is appropriate for measuring the effect of the intervention.2

For the reader of the trial a graph will allow the details of an intervention to be quickly and easily grasped. Aspects that may be missed in a long verbal description stand out clearly, thus the differences between experimental and control interventions become obvious. The CONSORT trial flowchart has improved transparency and accurate reporting of the numbers of participants at different stages of a study. We suggest that our proposed graphical method would similarly increase the clarity of reporting of complex intervention trials.

Summary points

  • Complex interventions often require long explanations that are difficult to follow

  • Graphical representation could help make descriptions clearer

  • The graph would prompt researchers to focus on the structure and timing and ensure appropriate comparisons

  • Readers would be able to see the differences between comparison groups immediately

Footnotes

  • We thank Paul Glasziou for helpful comments.

  • Contributors and sources: RP and PY had the idea of a graphical depiction. RP created the graphical display, built the trial database, and drafted the article. CH helped in the creation of the graphical display and contributed to the writing. PY originated the investigation into complex interventions, helped in the creation of the graphical display, and contributed to the writing.

  • Funding: CH is funded by a Department of Health Research Development Award.

  • Competing interests: None declared.

References

View Abstract