Impact of active placebo controls on estimated drug effects in randomized trials: a meta-epidemiological study

Our objectives were, first, to assess the impact of active placebo vs standard placebo control interventions on estimated benefits of experimental interventions in pharmacological randomized clinical trials and to explore causes for heterogeneity and, second, to assess the impact of placebo control type on estimated harm, attrition, and co-intervention use.

This was a meta-epidemiological study. We searched PubMed, Ovid Embase, the Cochrane Library, Google Scholar, PROSPERO, citations, and personal files until 2024 and included meta-analyses containing both active placebo-controlled and standard placebo-controlled randomized clinical drug trials from reviews of any topic. We used 2-stage meta-epidemiological models to investigate the impact of placebo control type. The primary analysis was the pooled ratio of odds ratios (RORs) of outcomes of benefit, where ROR < 1 indicated overestimation of benefits in standard placebo-controlled vs active placebo-controlled trials. We investigated robustness in sensitivity analyses and explored heterogeneity. The secondary outcomes were pooled RORs of harm, attrition, and co-intervention use.

We included 67 reviews with 123 meta-analyses of 1698 trials. The pooled ROR of benefit was 0.98 (95% CI 0.77-1.24, I² = 44%, τ² = 0.30, 95% prediction interval 0.33-2.92, 64 meta-analyses, 727 trials). The result did not change much in sensitivity analyses, and heterogeneity analyses were inconclusive. The results of secondary analyses were imprecise.

We did not detect a statistically significant average impact of active placebo vs standard placebo controls on estimated drug effects. However, the results were uncertain as the CIs included important impacts in both directions, and the prediction interval for benefit indicated considerable heterogeneity, which was not reliably explained in the analyses.

In many clinical drug trials, patients and health care providers are not told who are allocated to the experimental drug and who are allocated to a similarly looking placebo control. This "blinding" procedure aims to protect against bias by ensuring an impartial assessment. Some patients may, however, experience noticeable side effects from the drug, eg, dry mouth or sedation, which can lead them to guess their allocated intervention. This is called loss of blinding or "unblinding." Active placebo controls are different from regular placebo controls because they contain ingredients that also imitate the side effects of the drug. We do not know whether this protects against bias from loss of blinding. Therefore, we wanted to assess if trials with active placebo controls yielded different results than similar trials with regular placebo controls. We found such trials by using meta-analyses of comparable trials with both types of placebo controls for various conditions, eg, pain and depression. With more than 100 meta-analyses, including approximately 1700 trials, this is the largest investigation of active placebo controls. We found that in trials using regular placebo controls, the experimental drug effect was slightly larger compared to similar trials with active placebo controls, on average (2%), but there was considerable variation between meta-analyses. The result could be a chance finding and, due to random variation, was compatible with both a 23% exaggeration and 24% underestimation of effect (when measured as odds ratio). In addition, the average may conceal relevant context-specific circumstances under which active placebo controls are more important or less important. We explored several potential circumstances without finding an explanation for the variation between meta-analyses. We conclude that it remains unclear if, and under which circumstances, active placebo controls are more reliable than standard placebo controls.