Do patients’ preferences matter?

Patients’ preferences are important. Where treatment choices have different and well understood outcomes, what matters most when deciding which treatment is best is the patient’s preference. The intriguing question is whether strong treatment preferences can also affect outcomes—that is, can the preferences themselves exert a therapeutic effect over and above the anticipated effects of compliance and attrition, and if so, by how much? We know very little about this area. In the linked study (doi:10.1136/bmj.a1864), the Preference Collaborative Review Group reviews the effects of patient preferences on attrition and outcomes in trials of patients’ preferences using a meta-analysis of individual patient data.1

We know that the placebo effect can be profound. Its most plausible mechanism of action involves belief in an effect, and this must relate strongly to preferences.2 But a double blind trial of placebo versus nothing is needed to identify a placebo effect. Some meta-analyses have shown a placebo effect,3 but more convincing evidence comes from observation rather than experiment.

Henry Beecher provided evidence of a placebo effect when he used saline to anaesthetise injured patients after the morphine had run out—patients experienced little operative pain and less shock than expected.4 However, we do not know whether a proportion of patients would be anaesthetised anyway without the saline. The coronary drug project compared clofibrate and other drugs against placebo in men at high risk of coronary events. Mortality from coronary heart disease was 35% lower in men who took their placebo than in those who did not take their placebo as prescribed.5 The benefit could be explained by a simultaneous effect on risk reducing behaviour, although statistical adjustment for 40 potential confounders did not change the result. In observational studies of hormone replacement therapy, mortality from coronary heart disease seemed to be 50% lower,6 but this effect was subsequently found to be the result of preference effects and confounding.7

Why do we know so little about the placebo effect yet so much about active treatment? The answer lies partly in our ability to randomise active treatment against previous treatments or placebo because this will tell us about their relative efficacies. However, interpreting information from a randomised trial comparing a placebo with no treatment is complicated by uncertainties about belief and its consequence—who believes what, by how much, and why they believe. Randomising between genuine preferences is obviously (and uniquely) impossible, and it may be unethical to randomise patients if their preferences are strong.8 Grant and colleagues demonstrate elegantly that strong preferences seem to depend on the prognosis in patients with chronic gastro-oesophageal reflux.9 Hence, the evidence on both preference and placebo effects is usually observational and likely to be both confounded and attenuated—a lethal combination in epidemiology. How close can we get to clear evidence about the extent of such preference effects? A clinical trial design that might reduce confounding has been described, but it is complex, needs large numbers of participants, and cannot completely overcome the problems outlined above.10 The next best option is to measure preferences in trial participants and use that information in the analysis as one possible determinant of outcome.

The Preference Collaborative Review Group has carried out a meta-analysis of such trials, which mostly looked at musculoskeletal treatments.1 The group obtained individual patient records and examined the extent to which patients who received their preferred treatment did better than those who were indifferent or not allocated their preference. It found that the effect size of treatments in patients who were randomised to their preferred treatment was significantly greater than in patients who were indifferent to the treatment assignment (effect size 0.162, 95% confidence interval 0.011 to 0.314; P=0.04) but that preferences had little effect on attrition. This is a major step in our understanding, and it shows that preferences can have an effect on outcomes. However, the effects could be explained by people with strong preferences refusing to be randomised, so that the people included in these trials would be those with weak preferences. Also, the effects are bound to be context specific—some illnesses might be more amenable to the effects of preferences than others.

Because it is generally difficult to observe preference effects reliably,11 12 a strong theory of how these preferences might happen is a prerequisite for research. If preferences or beliefs in treatment can be shown to have measurable and attributable effects on precursors of recovery or improvement, then a proper study would be justified. Such a study will be complex and difficult and without a strong theoretical basis would be hard to justify.

Meanwhile preference effects will remain the elephant in the room and widely held to be fanciful. Preference effects, especially where preferences concentrate on one treatment, can in theory seriously confound unblinded randomised studies where the observed advantage seems misleadingly to favour an inactive treatment.13 Randomised evidence is considered most rigorous, but maybe this idea is misplaced when we know so little about the nature and extent of preference effects.