Intended for healthcare professionals

Rapid response to:

Education And Debate

Complex interventions: how “out of control” can a randomised controlled trial be?

BMJ 2004; 328 doi: https://doi.org/10.1136/bmj.328.7455.1561 (Published 24 June 2004) Cite this as: BMJ 2004;328:1561

Rapid Response:

Complex interventions: more thought needed

Hawe et al (2004) have given an interesting solution to the problem
of evaluating complex interventions, by standardising interventions by
function rather than by form. Their Table 1 gives some idea of how
standardisation may be carried out either by form or by function. I agree
that to treat something like an educational intervention as if it were a
drug is a serious mistake. However I feel that their ideas still need some
refinement. Firstly, even apparently simple drug trials can be regarded as
complex, because issues such as adherence depend on a variety of
influences, including feedback as to whether symptoms improve or not. As I
see it, the problem with their suggestions is the generalisability of the
outcome. For example, a reader of a trial standardised by function might
like to educate his/her own patients about depression. Faced with the fact
that each site in the trial adopted a different form of intervention, how
does the reader choose a suitable one for his/her patients? One suggestion
is to supply an algorithm which will recommend an optimum intervention for
a given case mix. An important concept is that the algorithm is developed
in pilot or Phase II trials, so that the Phase III trial is testing the
algorithm, which has been developed a priori. Thus the interventions will
vary depending on the literacy, culture and leaning styles of the
patients, but any subsequent user of the intervention will know how to
apply it to his/her patients.

There is an analogy here with evaluation of complementary therapies
such as homoeopathy. Therapists sometime claim that randomised controlled
trials in homoeopathy are impossible since the treatment has to be
tailored to the individual. Trialists counter this by saying the
intervention is simply ‘referral to a homoeopathist’ – what the
homoeopathist actually does is a ‘black box’ that it is impossible to
disentangle. Thus referral to the homoeopathist is the ‘function’. One
might assume that homoeopathists have an intuitive algorithm that helps
them optimise care for their individual patients. The important design
aspects here are to have a control group (such as conventional medical
care), some form of random allocation and an agreed outcome measure that
is applicable to both groups. A challenge to the designers of trials here
is to separate the effect of therapy from the effect of the therapist.
Thus a trial with a single therapist would be a poor one, since a patient,
faced with a successful outcome from such a trial would have no idea
whether it would be worth going to a different therapist. Similarly,
trials of complex interventions are poor if they restrict entry to, say,
well educated, middle class patients.

Competing interests:
None declared

Competing interests: No competing interests

27 October 2004
Michael J Campbell
Professor of Medical Statistics
Institute of Primary Care, University of Sheffield S5 7AU