Trial methodology and patient characteristics did not influence the size of placebo effects on pain

Abstract

Objective

To determine whether trial-design, patient-type, or placebo-type factors influence the size of the placebo analgesic effect in clinical trials.

Study Design and Setting

Trials that measured pain outcomes in Hróbjartsson and Gøtzsche's meta-analysis were retrieved and coded for eight factors potentially predictive of placebo effect size. Random effects meta-regression was used to explore the predictive power of each factor on placebo effect size. The factors investigated aspects of trial design (nonstandardized co-analgesia, co-intervention), patients (pain type, patient group, residual pain score), and placebo (placebo type, indistinguishability, structural equivalence). The meta-analysis undertaken in the original study was also repeated to confirm the results.

Results

The pooled effect of placebo was 3.2 points on a 100-point scale (95% confidence interval [CI] = 1.6–4.7). None of the selected factors influenced the size of placebo effect: the effect of all factors was close to zero, all CIs spanned 0, and P-values ranged from 0.13 to 0.90.

Conclusion

This study confirms the findings of previous researchers that, at present, the evidence for large placebo analgesic effects in clinical trials is lacking. Importantly, this analysis also establishes that larger placebo effects are not associated with particular aspects of the trial methodology, patient, or placebo type.

Introduction

In the 1950s, a review by Beecher was responsible for persuading the scientific community that placebo effects were powerful enough to produce satisfactory relief in a large proportion of patients with various medical conditions [1]. Recently, Hróbjartsson and Gøtzsche argued that Beecher had overestimated the size of the placebo effect because his estimates were based on within-group improvements reported in placebo groups [2], [3]. Within-group improvements reflect not only the placebo effect but also factors such as the natural course of the condition and regression to the mean. Hróbjartsson and Gøtzsche pointed out that the size of placebo effects is more accurately estimated by subtracting improvements seen in a no-treatment group from the improvement seen in a placebo group.

This reasoning underpinned Hróbjartsson and Gøtzsche's reviews of the effects of placebo [2], [3]. They systematically reviewed randomized trials that included both placebo and no-treatment conditions. Meta-analysis of this clinically heterogeneous set of trials revealed statistically significant effects on pain of approximately 6 points on a 100-point scale. We suspect, few clinicians would consider effects of this magnitude to be of clinical significance. Additionally, there was some indication that the effect may have been due to a small study effect, such as reporting bias [3].

Despite the striking evidence provided by the Hróbjartsson and Gøtzsche reviews that placebos have only modest effects, their findings have often been ignored in recent research [4], [5], [6], [7]. This may be in part due to criticisms directed at the methods used in the reviews, such as pooling studies on a heterogeneous group of medical conditions [6], [7], [8] and including trials in which co-interventions were allowed inthe no-treatment group [9]. In line with these criticisms, we were concerned that placebo effects were potentially masked by some aspect of the trial methodology or patient type. Our aim was to define some of these potentially confounding factors and test whether they influenced placebo effect size. For the purposes of this article, we have considered placebo effects to be those therapeutic effects resulting from the treatment ritual. We acknowledge that this definition does not take into account the totality of the effect of interactions between patients and providers [10].