Expectational confounds in randomised double-blind clinical trials

Michael D Kirk-Smith, The School of Health Sciences, The University of Ulster, Newtonabbey, Northern Ireland, BT37 0QB.

David D Stretch, The Greenwood Institute of Child Health, The University of Leicester Medical School, Westcote House,
Westcotes, Westcotes Drive, Leicester, LE3 0QU.

A recent study (Linde et al., 1997) of 186 placebo-controlled homeopathy trials found 26 of them to be of "good quality" and
worthy of inclusion in a meta-analysis. This analysis concluded that there was an overall positive effect of homeopathic
treatment.

No known process in current physical science can account for the effect of such infinitesimally dilute solutions. It occurred to us
that before concluding that the conventional laws of physics are incomplete or in error, it is useful to have considered and
eliminated alternative explanations for the results. One alternative explanation concerns the implementation of randomised
double blind clinical trials (RDBCTs).

We present here an analysis of psychological processes operating in RDBCTs which concludes that current interpretations of
RDBCTs are potentially flawed. This may partly explain the positive results observed in the meta-analysis. It provides a good
candidate for one of the "unknown and unidentifiable sources of bias" said to exist in RDBCTs (2). Thus, it has consequences for
medical research methods.

The problem can be simply stated. The RDBCT appears to be a rational method of inferring causality, and appears to control for
experimenter and patient expectations. Their presence was a motivating factor in devising double-blind methodology. So their
plausibility has been well-accepted. An RDBCT may be double-blind at its beginning. However, it is possible that the blinding
may not continue to hold as the study proceeds, given the inevitable construction of beliefs and inferences by those involved as
they observe and learn about the trial.

If those in contact with the patients (the experimenters) begin to suspect that a patient under a certain regime is improving (and it
does not matter if it is for a placebo or treatment), then from this point on there is a possibility that these beliefs or expectations
will affect the experimenters' behaviour. They could then give subtle and unconscious cues to the patient. All this may result in
the unconscious biasing of any ratings about changes in clinical symptoms made by either experimenters or patients. Such effects
could begin to happen at any time during the trial. It then becomes uncertain how much of any observed effect is due to
such situationally-generated expectation processes.

The risk would be increased if experimenters manage more than one patient in the trial, because they could then make
discriminations amongst patients. This could tend to polarize their beliefs and expectations. Such categorisation would imply that
patients' data become non-independent.

If the possibility of these situationally-generated expectations is not carefully monitored and accounted for from the start of the
trial to its end, then it is not formally possible to exclude them, leading to a plausible alternative explanation for the observed
results. Any RDBCT which does not have such monitoring as part of its design is therefore at risk of being confounded. The
conclusions drawn from it have validity only if the researchers accept their responsibilities; i.e., they must demonstrate that this
alternative explanation does not apply.

Since such monitoring has not routinely been included in the published RDBCTs that we have seen (we have looked at the forty
most recent RDBCTs in JAMA, BMJ, NEJM and The Lancet), we conclude that these risks may apply, and that this confound
may be present. We recommend that methods of monitoring and accounting for psychological processes of expectations and
beliefs be included in RDBCTs.

References

Linde, K., Clausius, N., Ramirez, G., Melchart, D., Eitel, F., Hedges, L.V. & Jonas, W.B. (1997) Are the clinical effects of
homeopathy placebo effects? A meta-analysis of placebo-controlled trials. Lancet, 350, 834-843.

Vandenbroucke, J.P. (1997) Homeopathy trials: Going nowhere. Lancet, 350, 824-825.