The process of randomization in a clinical tri al involves creating two populations with com parable prognoses1. This implies a balanced distribution of prognostic factors, which can either be pre-known, such as cardiovascular risk fac tors in the case of cardiovascular outcomes, or unknown. Unknown prognostic factors refer to variables that may or may not influence the oc currence of the measured outcome and are not known prior to the trial.
Difference in outcomes
To confidently attribute the difference in out comes between two populations to the evalu ated intervention rather than an imbalanced distribution of other prognostic variables, the following criteria must be met: 1) allocation concealment, 2) blinding, 3) analysis according to randomization (previously known as inten tion-to-treat analysis), 4) complete follow-up, 5) no early termination of the study due to dem onstrated benefit in the intervention. Failure to meet even one of these criteria could lead to an overestimation of the intervention effect by up to 30%2-4.
Placebos
Placebos are defined as inert substances with no pharmacological activity. Properly using them as comparators in randomized studies is essential for fulfilling the first two criteria5.
Allocation concealment refers to procedures designed to prevent both the person assigning the treatment and the patient from knowing the group to which the participant is being assigned. Adequate blinding refers to procedures intended to prevent patients, caregivers, event adjudicators, event recorders, and trial proces sors from knowing which intervention each arm of the research received. These procedures often involve centralized randomization, sequential numbering of medication, and ensuring identi cal appearance between the intervention and the placebo.
Essential conditions
Based on the above, we can deduce that a pla cebo in a randomized study must fulfill two es sential conditions: 1) have a neutral effect on the measured outcomes, and 2) be indistinguishable from the intervention.
To fulfill both conditions, one might expect that the placebo should faithfully replicate the composition of the reference product (a formu lation that is always well known), except for the active ingredient. This ensures the fulfillment of our first condition.
But what would happen if the absence of the active ingredient resulted in a change in appear ance or another characteristic of the placebo that made it easily distinguishable from the intervention? In that case, in order to fulfill ad equate blinding while reducing the risk of bias, we should add to the placebo some substance that imitates the perceptible characteristics of the active ingredient.
Disturbing question
This raises an even more disturbing ques tion: What if adding components to the pla cebo that mimic the active ingredient causes it to no longer have a neutral effect on the out comes? Could we easily detect it? Let’s see an example.
In 2019, the REDUCE-IT study was published6. This randomized study, with a placebo compar ator, included patients with established cardio vascular disease or diabetes who were receiving statin treatment and had fasting triglyceride levels of 135 to 499 mg/dL. The intervention arm received 2 grams of ethyl eicosapentaenoic acid (EPA). In the control arm, the placebo used con tained mineral oil to mimic the color and con sistency of EPA. The study evaluated the inci dence of a combined outcome of cardiovascular events, demonstrating a difference of almost 5% in absolute terms in favor of the intervention arm after 4.9 years of follow-up (17.2% vs. 22.0%. HR 0.75; 95% CI, 0.68 to 0.83; p <0.001).
These results were surprising to the research ers, both due to their inconsistency with previ ous studies with other omega-3s and the pres ence of a greater-than-expected benefit based on the observed triglyceride level changes7.
To address this question, the results of the STRENGTH study, published in 2020, were await ed8. This study tested a combination of two ome ga-3s (75% EPA and 25% DHA) against a placebo with a composition different from that used in REDUCE-IT (corn oil instead of mineral oil), and no significant differences in the combined car diovascular events were observed after 3.5 years of follow-up.
One year later, Takahito Doi and colleagues published an analysis based on a cohort study that imitated the designs of REDUCE-IT and STRENGTH9. By combining the changes in tri glyceride levels, LDL, and C-reactive protein ob served in the active oil and respective placebos of the original studies, they estimated hazard ratios for the combined cardiovascular events for all study arms. They concluded that the in consistency between the two studies could be partly explained by the different effects of the comparators (mineral oil vs. corn oil). These findings were specifically attributed to an ef fect of the mineral oil used in REDUCE-IT on the intestinal absorption of statins, reflected in a 10.9% increase in LDL in the control arm during the study.
Mechanism of exaggerating the effect of an intervention
This example serves to illustrate a poorly de scribed mechanism of exaggerating the effect of an intervention and the difficulty involved in its systematic detection.
For authors conducting systematic reviews, it would be useful to acknowledge this phenom enon as a potential cause of inconsistency be tween clinical trial results, and we suggest in corporating it as “inadvertently active placebo risk” in the assessment of bias risk.
Highlight
To highlight this phenomenon, we propose that those assessing the risk of bias in a trial with a placebo comparator consider three ques tions:
1. Is there potentially any difference between the composition of the placebo and the inter vention, other than the active ingredient?
2. Is the observed effect consistent with our previous knowledge?
3. Is the observed effect consistent with that demonstrated in other similar studies?
Final commentary
In a recent systematic review by Cohcrane, Laursen et al. did not find any differences be tween the use of active placebos vs. standard placebos10. However, they defined an active pla cebo as any intervention designed to imitate the perceptible non-therapeutic effects of an experi mental intervention (e.g., anticholinergic effects of tricyclic antidepressants), while a standard placebo was considered to be any intervention designed to mimic only the external properties of the experimental intervention. This analysis does not address our concern, as the issue of an inadvertently active placebo is that it is de fined as a standard placebo by the authors of the study in question.
So, future other reviews should elucidate whether the described phenomenon is excep tional and of little significance, or if it is a fre quent and underestimated occurrence.