What can be done to control the placebo response in clinical trials? A narrative review

Abstract

The desire to reduce high placebo response rates in clinical trials is a popular concept. However, few studies have rigorously examined the effectiveness of methods to control for placebo responses that are relevant to randomized controlled trials. The primary objective of this review was to evaluate the effect of experimental placebo manipulations in randomized controlled trials (RCTs). We critically reviewed studies designed to manipulate placebo responses including positive expectations regarding the effectiveness of the placebo treatment, manipulating the time spent with subjects, and training study staff and subjects to accurately report symptom severity. These efforts have generally resulted in reduced placebo response and improved discrimination between drug and placebo. Interventions that neutralize staff and subject expectations and improve the ability of subjects to accurately report symptom severity have shown the most promise. Reduction of the placebo response has the potential to accelerate the development of new therapeutics.

Introduction

In the realm of clinical trials, placebos are necessary tools to provide evidence that a new treatment is more effective than an inert treatment. In this context, the placebo response, which refers to the degree of clinical improvement reported by patients assigned to the placebo group, is a major factor that can distort the measurement of net treatment effect in a randomized controlled trial (RCT).¹ The study outcome, determined by the observed difference in outcomes between the drug and placebo groups, is driven by the degree of improvement in the placebo group, which varies by study and by site.² Several factors contribute to the placebo response, including the natural history of the disease, regression to the mean, baseline score inflation, patients and researchers’ biases, co-interventions, the Hawthorne effects, and the placebo effect itself (Table 1).³

Table 1:

Key terms and concepts

Term	Definition
Placebo response	The placebo response refers to the degree of clinical improvement reported by patients assigned to the placebo group. The placebo response can be driven by several factors including the natural history of the disease, regression to the mean, baseline score inflation, patients’ and researchers’ biases, co-interventions, the Hawthorne effects, and the placebo effect itself.3
Placebo effect	The placebo effect refers to that component of the placebo response which is attributable to the act of receiving the inert treatment, including the neurobiological and psychological mechanisms of expectancies.14,17
Placebo manipulation	The act of implementing a condition aimed at altering the placebo response (example: altered drug labeling or increasing expectations verbally).
Expectation	An expectation refers to an anticipation of an event or outcome that can be derived from multiple sources including past experiences; written and verbal information; or anticipation of an event from the research or clinical staff that are intentionally measured via scales.1,6,9
Expectancy	Anticipated future outcomes or events that may be implicit and are not intentionally measured via scales.1,5
Interoceptive accuracy	Interoceptive accuracy is the extent to which an individual directs their attention internally, as opposed to externally such as focusing on external cues. 9
Assay Sensitivity	The ability to discriminate an effective treatment from a control.18

Note: While alternative definitions have been described¹⁹, these definitions have been found to be most commonly used and useful in the context of clinical trials and laboratory research.^1,17

The placebo effect is an improvement in outcome due in large part to the expectancies of positive treatment effects that trigger in turn, a chain of neurobiological changes.¹ While understanding of the mechanisms underlying the placebo effect continues to evolve, important drivers are anticipation of therapeutic benefit, and observational and social learning mechanisms.^1,4,5 Patients can derive their expectancies from written and verbal information; social media; discussions with other patients; perceived probability of receiving active treatment and effectiveness of that treatment; as well as expectancies of the provider and research staff, which are transmitted to the patient consciously and unconsciously, verbally and non-verbally.^6–8 Studies have also shown that some patients are more influenced by the cues that drive expectancy than others patients; these patients are often those whose attention is externally rather than internally directed.^9,10

In the context of RCTs, it has been demonstrated that as the placebo response increases the difference between the placebo and active arm decreases, reducing the likelihood that the trial will meet statistical significance of the primary endpoint.^11,12 Placebo responses are especially problematic in studies that rely on subjective or effort-dependent patient-reported outcomes.^3,9 The increasing failure of late phase clinical trials to show superiority of active treatment over placebo has been observed in depression, neuropathic pain, cancer pain, multiple sclerosis, Parkinson disease, and more. ⁴ Further, evidence has shown that placebo responses has been increasing over time in some indications^4,11–13 without a commensurate increase in response to active treatment; thus, the necessity to learn the mechanism behind the placebo response and how to mitigate it in the context of RCTs has become increasingly vital.

Several methods to control the placebo response without undermining the observed response to active treatment have been attempted in clinical trials, such as exclusion of placebo responders during placebo lead-in periods¹⁴, alternative study designs such as sequential parallel comparative designs^14,15, and various methods for measuring² or “neutralizing” ^5,16, i.e. managing, subject’s expectations. However, few studies have rigorously examined the effectiveness of these methods in the context of RCTs.³

The objective of this review is to summarize the methodology and results of published studies that evaluate interventions for modifying placebo responses in RCTs to illustrate which interventions are supported by evidence.

Methods:

Our primary objective was to evaluate the effect of a placebo manipulation in the context of RCTs. We chose to focus on treatments that are regarded as effective so that the interpretation of the differential action of placebo manipulation on drug vs. placebo arms would be evident. Some articles included in this narrative review were already known to the authors, however, a literature search was also conducted for clinical trials that randomize subjects to receive the placebo manipulation condition or a standard condition and then randomize subjects to receive active or placebo treatment. We limited our search to studies with both randomized placebo manipulation conditions and randomized treatments (placebo vs. active) so that causality may be inferred if a difference in placebo response was found. The search included articles published from inception to 2020 in PubMed. Search terms included a combination and variations of “placebo”, “placebo response”, “placebo manipulation on”, “clinical trial”, “factorial design”, and “double-blind”. The selection criteria were: 1. Random assignment to either an active treatment or a placebo, and 2. random assignment to one or more placebo manipulation conditions (interventions intended to influence the placebo response). The treatment vs. placebo arms were to be randomized and double-blinded, and the placebo manipulation condition randomized, but not necessarily masked. We did not require the placebo manipulation condition to be masked because most manipulation conditions require study staff to understand and perform these different manipulation conditions on the study participants. We also limited our investigation to studies where participants were patients with a symptomatic disease being treated (as opposed to healthy volunteers) because clinical trials on treatment efficacy, in which a high placebo response can be detrimental, are not typically done on healthy volunteers. We did not restrict our search to any one indication because high placebo responses are found across many indications, mechanisms of the placebo response appear to be similar across indications, and the pool of studies examining how to manipulate placebo responses in an RCT is small.

Studies meeting the inclusion criteria were presented in a table highlighting the main findings of each intervention method. Select studies were chosen to be expanded upon if they were considered successful based on our criteria: (1) the study treatment was deemed efficacious—defined as a statistically significant main effect (or near significant for small samples) of study treatment vs. placebo across all subjects or within a subset of subjects, and (2) the placebo manipulation was considered effective (defined as a statistically significant reduction in the response to placebo among patients in the placebo manipulation condition compared to control). All studies were reviewed for the discussion and conclusion.

Results:

An initial search of the literature identified only 7 studies that fulfilled our criteria, which were in the areas of asthma, migraine, knee osteoarthritis, and chronic pain (Table 2). Of these, 3 failed to demonstrate efficacy (i.e. statistically significant treatment effect) of the treatment vs. placebo in the primary analysis cohort, and 1 failed to demonstrate an effect of the placebo manipulation condition on the placebo response vs. control. The other 3 studies demonstrated successful placebo manipulation in a successful treatment vs. placebo RCT, which is most relevant to the planning for RCTs. These three identified studies were primarily investigator-initiated studies conducted at academic research centers as opposed to industry-sponsored RCTs conducted in support of a new drug application.

Table 2.

Studies Manipulating Placebo Responses in Randomized Controlled Clinical Trials

Reference	Indication	N	Study Designa	Intervention(s) Testedb	Clinical Outcome Measures	Results
Studies that failed to demonstrate efficacy of analgesia or placebo manipulation
De Craen et al. 200120	Chronic pain for ≥ 6 months	111 subjects	2 × 2 factorial randomized double-blind placebo-controlled trial. Intervention: Positive or neutral messaging about the treatment. Treatment: Tramadol or placebo.	Positive or neutral information about the expected effect of the treatment verbally expressed by the physician.	1°: Sum of pain intensity differences (SPID) between baseline and 1-hour post-treatment.	•  No significant effect of expectancy on the analgesic effects of tramadol versus placebo. • No significant analgesic effect of tramadol relative to placebo in the combined cohort.
Kemeny et al. 200721	Mild intermittent and persistent asthma	55 subjects	2 × 2 factorial randomized double-blind placebo-controlled crossover trial Intervention: Positive or neutral expectations. Treatment: Salmeterol or placebo prior to methacholine challenge.	Enhanced or efficient physician interactions that communicated positive or neutral expectations regarding treatment effects, respectively.	1°: Airway hyperreactivity (measured by FEV1). 2°: Psychological measures (to predict placebo responders) including depression symptoms, hostility, anxiety and ratings of the treatment and physician.	• Placebo bronchodilator significantly reduced objective airway hyperreactivity measures relative to baseline, but to a lesser extent than did salmeterol. • Positive physician encounters had no effects on objective measures of lung function in either treatment group or on the number of placebo responders.
Suarez-Almazor et al. 201022	Pain due to osteoarthritis of the knee	527 subjects	2-stage, 2 × 2 factorial randomized double-blind placebo-controlled trial with wait list control group. Intervention: High or neutral expectations Treatment: traditional Chinese or sham acupuncture	High expectations of treatment effectiveness delivered verbally by acupuncturist and by brochure, or neutral expectations of treatment effectiveness delivered verbally by acupuncturist and by brochure	1°: Joint-Specific Multidimensional Assessment of Pain (J-MAP), Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), and satisfaction with knee procedure (SKIP) scores	• Traditional Chinese acupuncture was not superior to sham acupuncture • Patients receiving either treatment showed improvement for most outcome measures relative to wait list control group. • Patients in the high expectations group had significant improvements in pain scores and satisfaction relative to the neutral expectations group independent of treatment group.
Dossett et al. 20158	GERD	24 subjects	2 × 2 factorial randomized double-blind placebo-controlled trial. Intervention: Standard or enhanced physician visit. Treatment: Acidil or placebo.	The standard physician visit consisted of the physician asking routine questions about GERD history, symptoms, etc. The enhanced physician visit consisted of more time spent with the patient asking additional questions.	1°: Percentage of patients with ≥ 50% improvement in GERD symptom severity compared to baseline	• No significant improvements in GERD or dyspepsia were noted between subjects receiving Acidil or placebo. • Subjects in enhanced visit group were more likely to have a ≥ 50% reduction in GERD symptoms, greater improvements in dyspepsia symptoms, and were more likely to decrease the number of antacid tablets used compared to those in the standard visit group.
Studies that demonstrated efficacy of both analgesia and placebo manipulation
Wise et al. 200916	Poorly controlled asthma	595 subjects	2 × 2 factorial randomized double-blind placebo-controlled trial with a usual care control group. Intervention: Enhanced or neutral messaging. Treatment: Montelukast or placebo.	Enhanced (optimistic) messages to increase the expectation of benefit or neutral messages to neutralize expectations. Each message presentation included 3 components: 1) a scripted message read by staff; 2) a computer presentation; and, 3) altered appearance of the capsules and brand name use.	1°: Mean change in daily peak expiratory volume over 4 weeks 2°: Patient-reported lung function and asthma symptom control	• Enhanced messaging had no effect on lung function compared to neutral messaging in the montelukast or placebo groups. • No differences in lung function outcomes between the neutral placebo and usual care groups. • Patients in the enhanced placebo group but not in the enhanced montelukast group had improved self-reported asthma control compared to their respective unenhanced group. • Enhanced messages led to no significant difference in asthma control outcomes between the placebo and montelukast treatment groups. • Patients in the neutral placebo group had better self-reported asthma control than those in the usual care group.
Kam-Hansen et al. 201423	Episodic migraine	66 subjects	2 × 3 factorial within-subjects, repeated-measures, randomized double-blind placebo-controlled trial with an expanded “balanced placebo” Intervention: Negative, neutral, or positive information about the treatment. Treatment: Maxalt or placebo.	3 information conditions were tested: negative (study drug or placebo labeled as “placebo”), neutral (study drug or placebo labeled as “Maxalt or placebo”), and positive (study drug or placebo labeled as “Maxalt”); labeling was true for 4 attacks and false for 2 attacks. Each patient was treated for 7 attacks with the conditions presented in counterbalanced order	1°: Percentage change in headache pain scores between baseline (recorded 30 min after migraine onset) and 2 h later	• Overall, Maxalt resulted in greater headache severity relief than placebo. • Uncertain labeling (i.e., “Maxalt or placebo”) resulted in greater pain relief in both the placebo and active treatment conditions than when labeled as “placebo.” • Placebo labeled as “Maxalt” was not significantly different in effectiveness than Maxalt labeled “placebo,” and placebo labeled as “placebo” was more effective than no treatment.
Treister et al. 201818	Painful diabetic neuropathy	61 subjects	2-stage, 2 × 2 factorial randomized double-blind placebo-controlled crossover trial. Intervention (training stage): Psychophysical accurate pain reporting training or control. Treatment (evaluation stage): Pregabalin or placebo.	Subjects completed training in 2–4 sessions across a 4-week period. Training included evaluating the intensities of noxious stimuli mechanically applied to the thumbnail bed with a MAST device using the 1–10 NRS. Subjects received feedback on reporting accuracy without accuracy feedback from researchers. The control group received no training.	Training stage 1°: R2, calculated as an indicator of accuracy in pain reporting; Evaluation stage 1°: Mean change from baseline in 24-hour recall average pain intensity on 0–10 NRS	• Training stage: Subjects’ pain reporting skills improved during training (mean R2 across the 1st, 2nd, 3rd and 4th sessions: 0.52, 0.62, 0.64, and 0.71, respectively). • Evaluation stage: No statistically significant differences in 24-h average pain intensity on the 0–10 NRS between pregabalin and placebo groups. • Training group had a positive SES (0.31) with a trend towards statistical superiority (p = 0.108) of pregabalin over placebo. • Untrained control group had a negative SES (−0.21) and a significantly larger (p = 0.018) placebo response compared to the training group.

FEV₁, forced expiratory volume in 1 s; GERD, gastroesophageal reflux disease; MAST, multimodal automated sensory testing; SES, standardized effect size; N, Number of subjects included in the analysis

^aIn some of the studies, the randomized placebo effect intervention was single-blind, and the randomized treatment was double-blind.

^bRefers only to interventions or methods intended to manipulate the placebo response.

Enhanced messaging manipulates placebo response in a study on asthma¹⁶

The objective of this study was to investigate whether information intended to increase expectations of therapeutic benefit could affect asthma outcomes in patients randomized to receive montelukast (a marketed leukotriene inhibitor) or placebo. There were 601 subjects divided into 5 treatment groups: (1) placebo with enhanced messaging, (2) placebo with neutral messaging, (3) montelukast with enhanced messaging, (4) montelukast with neutral messaging, and (5) usual care. Patients were subjected to either an enhanced messaging presentation designed to increase expectations of therapeutic benefit or a neutral messaging presentation designed to elicit neutral expectations of therapeutic benefit. There were 3 components to each messaging condition: (1) a scripted message read by study staff, (2) a computer presentation, and (3) altered capsule appearance and brand name use. Patients assigned to the usual care group were not subjected to any presentation.

Examples of enhanced expectation messaging included: “Singulair is a new medication that can be prescribed by your doctor to prevent asthma symptoms and make you feel better”. Computer presentations consisted of 16–21 screens of narrated educational information about asthma. The enhanced computer presentation was introduced by research staff using positive messaging about the benefits and safety of the medicine being tested, and the presentation emphasized the value and potency of treatment and included a commercial for montelukast (Singulair) featuring attractive young adults leading active lifestyles unhindered by asthma. The neutral computer presentation was introduced by staff using neutral messaging without inferences about treatment benefits, and the interactive presentation showed the same basic information about asthma, but without positive messages about the expected benefits of montelukast or the television commercial. The appearance of the capsules was controlled to account for expectations that might arise due to the use of particular colors. The enhanced group received capsules that were 2-tone blue and the neutral group received off-white capsules. Additionally, treatment in the enhanced group was referred to by brand name, Singulair, while treatment in the neutral group was referred to by the generic name, montelukast.

There were significant main effects of drug on all physiological outcome measures, indicating superiority of the active treatment. In contrast, there were no significant main effects of message alone on physiological outcome measures; enhanced messaging had no effect on lung function compared to neutral messaging in either treatment group, and there were no differences between the neutral placebo and usual care groups. However, placebo-treated patients who received enhanced messaging had improved scores on subjective self-report questionnaires including the Asthma Control Questionnaire (ACQ) and the Asthma Symptom Utility Index (ASUI) compared to montelukast-treated patients (Figure 1). Additionally, there was a significant (p < 0.01) drug × intervention interaction for ACQ scores, and a significant (p = 0.03) main effect of intervention on ASUI scores.

Figure 1. Effect of treatment and placebo intervention on patient-reported outcomes assessing asthma control and symptoms – the Asthma Control Questionnaire (ACQ) and Asthma Symptom Utility Index (ASUI).

Self-reported outcomes using the Asthma Control Questionnaire (possible score range 0–6, lower score is better) and Asthma Symptom Utility Index (possible score range 0–100, higher score is better). ACQ: P_{drug × message} = 0.01; ASUI: P_message = 0.05, P_{drug × message} = 0.07; error bars = 95% confidence intervals. Data extracted from Wise et al.¹⁶

In summary, neutral messaging reduced the response to placebo, but not to drug, compared to enhanced messaging. The net effect was restoration of the expected difference between drug and placebo, whereas in the enhanced messaging condition the expected separation between drug and placebo disappeared. This effect was observed with subjective but not objective outcome measures.

Altered placebo and drug labeling changes affect outcomes after episodic migraine attacks²³

This study hypothesized that clinical outcomes for migraine would improve progressively as the supplied information varied from negative (0% chance of receiving active treatment) to uncertain (50% chance of receiving active treatment) to positive (100% chance of receiving active treatment). Patients were given 1 of 2 treatments (placebo or 10 mg rizatriptan [Maxalt]) labeled in 1 of 3 ways (“Maxalt,” “placebo,” or “Maxalt or placebo”) for 6 migraine attacks. During enrollment, all 66 subjects were provided basic information about the purpose of the study and told that Maxalt is an anti-migraine medication that works best if taken 30 minutes after the onset of headache. Patients were also told that after an initial untreated attack, they would receive a brown study drug envelope labeled as “Maxalt,” “placebo,” or “Maxalt or placebo” for the following 6 attacks. Labeling was true for 4 attacks and false for 2 attacks (Maxalt labeled as placebo and placebo labeled as Maxalt).

Headache relief scores were lowest when pills were labeled “placebo” (whether true or not) and highest when pills were labeled as “Maxalt” (whether true or not). Uncertain labeling (i.e., “Maxalt or placebo”) was associated with larger reductions in pain scores than “placebo” labeling in both treatment conditions. Placebo pills labeled as “placebo” were more effective than no treatment. Surprisingly, placebo labeled as “Maxalt” was not significantly different in effectiveness than Maxalt labeled “placebo”.

In summary, information provided by way of printed labeling significantly contributed to active drug and placebo responses. Both uncertain information (“Maxalt or placebo” labeling) and positive information (“Maxalt” labeling) resulted in larger therapeutic effects than negative information (“placebo” labeling), regardless of the actual treatment.

Accurate pain reporting training diminishes the placebo response: Results from a randomized, double-blind, crossover trial (Treister)¹⁸

The aim of this study was to test whether training subjects with painful diabetic neuropathy to report experimental pain accurately can reduce variability in clinical pain reporting, or impact the responses to placebo or drug. In this 2-stage study, 61 subjects were first randomized to the Accurate Pain Reporting Training (APRT, WCG Analgesic Solutions, Wayland, MA) group or the control group (no training) in the training stage, and then subsequently randomized to receive pregabalin or placebo for 10–13 days each in the evaluation stage. Subjects randomized to the APRT group completed 2–4 in-clinic training sessions over a 4-week period. Each training session consisted of 4 discrete applications of 6 different stimulus intensities (pressure) on each thumbnail bed and were instructed to rate the pain from each stimulus on a 0–10 numerical rating scale (NRS). At the end of testing, subjects reviewed a scatterplot of their own ratings versus actual stimulus intensities, and researchers used scripts to provide standardized feedback on the relationship between their pain ratings and the actual stimulus intensities. Researchers were instructed not to correct subjects’ ratings or to insinuate that they were wrong, but instead to emphasize the importance of accurate reporting. The control group received no training. The primary outcome measure in the training stage was the correlation between stimulus intensities and pain ratings—quantified using the R² value where 0 is no correlation and 1 is complete correlation— as an indicator of pain reporting accuracy. The primary outcome measure in the evaluation stage was mean change from baseline in 24-h recall average pain intensity on a 0–10 NRS.

Results from the training stage showed that subjects’ pain reporting skills improved across the 4 training sessions, from average R² values of 0.52 to 0.62 to 0.64 to 0.71, respectively. Results from the evaluation stage showed that there was no significant difference in the primary outcome measure (24-h recall average pain intensity) between the pregabalin and placebo groups in the whole population; however, subjects in the APRT group showed a positive standardized effect size (SES=0.31) with a trend towards statistical superiority (p = 0.108) of pregabalin over placebo. Conversely, subjects in the untrained control group had a negative SES (−0.21) indicating greater responses to placebo than to pregabalin. There was a significantly larger (p = 0.018) placebo response in the untrained group than in the APRT group, which contributed to the difference in SES between groups. Variability of clinical pain scores was lower in the trained vs. the untrained group.

In summary, subjects were successfully trained to report experimental pain more accurately, and better pain reporters were more likely to discriminate active treatment from placebo, due to a lower placebo response. The theory behind these findings is that subjects who are able to look internally to determine their accurate pain levels are less likely to be influenced by external cues that can increase placebo responses.¹⁰ These findings support previous studies indicating that high variability in pain reporting is associated with larger placebo responses, but not larger responses to drug,^24,25 and extend these findings by demonstrating that a training intervention can reverse these effects.

Discussion

Large and increasing placebo responses pose a serious obstacle to the success of drug discovery and clinical development programs across therapeutic areas. There is conflicting evidence as to which geographic region the placebo response is most pronounced, but it is clear that the issue of high placebo responses are a global phenomenon.^12–14,26 Placebo responses are especially problematic in studies that rely on subjective patient-reported outcomes, and can have severe detrimental effects on assay sensitivity, contributing to the failure of trials of efficacious therapies^3,9.

A few lessons can be learned from the three studies that successfully implemented a placebo manipulation condition in RCTs. First, the findings in the studies by Wise and collegues¹⁶ and Kam-Hanson and collegues²³ suggest messages or information that increase patient expectancies, mask the demonstration of active treatment benefits. Thus, research staff and sponsors should minimize the dissemination of information that may enhance expectations (verbal and nonverbal) and carefully frame any given information in a neutral manner, without exaggerating the expected benefits of the treatment under study. It is important to note, however, that neither study measured subjects’ levels of expectation following the manipulation condition; therefore, it remains unconfirmed whether the effectiveness of the intervention was mediated through expectation. Second, other factors such as the use of brand names, treatment labeling, and the coloring of capsules or pills should be taken into consideration when attempting to neutralize study subject expectancies.

As shown by the studies summarized in Table 2, the results of manipulating placebo responses are not always consistent; however, a few key points should be noted. First, in the studies in which the intervention did not work as expected the active treatment was either administered at a dose that was possibly too low to have a significant effect, such as a single low dose of tramadol in the study by de Craen and colleagues,²⁰ or the treatment may have been inherently ineffective, such as the homeopathic treatment for GERD in the study by Dossett and colleagues.⁸ In effect, these studies lacked an effective treatment, so could not answer the question about whether the placebo response manipulation acted differentially upon an effective treatment vs. a placebo. Second, the studies that successfully manipulated the placebo response in our analysis generally utilized tightly controlled and well-scripted placebo manipulation conditions that were delivered by well-trained staff. For example, the successful study by Wise and colleagues prepared scripts for study staff to read and used computer-based presentations to deliver messages about treatment expectancies¹⁶, while the less successful intervention the study by de Craen and colleagues used positive non-scripted interactions with physicians to communicate treatment expectancies.²⁰ Third, intervention intensity had a clear effect on trial success. For example, the successful study by Wise and colleagues combined multiple interventions designed to enhance the placebo¹⁶, while the unsuccessful studies by de Craen and colleagues and Dossett and colleagues failed to show statistically significant effects of single-component interventions.^8,20 Robust and comprehensive strategies appear to be required for effective manipulation of placebo responses in RCTs. It is also important to point out that the studies had varying sample sizes and that smaller samples may not have had the power necessary to see a significant effect.

The emerging science of placebo effects has uncovered multiple potential methods for reducing placebo responses in RCTs. In general, these strategies aim to neutralize expectations of therapeutic benefit and increase the patient’s attentiveness to their own bodily experience. The research summarized in this article underscores the power of information, expectancies and cues communicated by study staff (verbal and nonverbal) and illustrates the effects of manipulating the type and amount of this information provided to patients. Experimental studies have found that the simplest way to reduce these pro-placebo influences is provide neutral information about expected treatment benefits and to communicate with subjects in an impartial manner that does not amplify the empathic clinician-patient relationship. An increasing body of evidence has also shown that individual factors influence the propensity of a subject to respond to the cues that drive the placebo response.

A key factor is the degree to which subjects’ attention is directed internally or externally—a phenomenon that has been called interoceptive accuracy⁹ or private self-consciousness¹⁰. The degree of a subjects’ interoceptive accuracy is manifested by variability in symptom reporting – patients with limited ability to introspect tend to have high variability in reported symptom intensity.^9,10 The work of Treister and colleagues²⁷ further suggests that training patients on how to accurately report their bodily sensations and subjective experiences can decrease the placebo response. Recent research has further demonstrated that training subjects to improve interoceptive accuracy results in a decrease in placebo response but not in the active response.¹⁸ Identifying subjects who will potentially provide more variable data in a study, and either excluding them or training them to become more accurate reporters, is a new and important concept at the forefront of clinical research.

This bidirectional approach to minimizing placebo responses, by reducing contextual effects and improving interoceptive perception, is in line with current evidence regarding the neurobiology of placebo effects. Experimental models of placebo analgesia have shown that placebos exert a top-down modulation of perception with activation of brain regions involved in cognition, valuation, and affective processing, such as the dorsolateral prefrontal cortex (dlPFC), ventromedial (vmPFC), and orbitofrontal cortices (OFC).^28,29 These top-down modulation seems to be responsible for the reductions in brain responses in somatosensory processing areas,^28,30,31 which have been associated with clinical improvement. Furthermore, research in this area using positron emission tomography (PET) and pharmacological approaches, has shown that the activation of these regions is associated with the release of endogenous neurotransmitters, such as opioids,³² dopamine,^33,34 endocannabinoids,^35,36 vasopressin³⁷ and oxytocin³⁸. Yet, the extent to which these experimental approaches translate into a greater ability to improve assay sensitivity in an RCT has not been investigated as most experimental manipulations using PET have been done outside of the RCT setting. Very few studies have, in fact, investigated the biological correlates of expectancies in a clinical trial setting. In one recent study, twenty-three patients with depression completed an fMRI followed by a RCT where expectancy was manipulated by instructions to participants about the probability of receiving active medication as opposed to placebo: one group was told that they were randomized to open trial anti-depressant (100% chance of receiving active treatment); the other group was told that they were randomized to placebo-controlled antidepressant (50% chance of receiving active treatment); however, no active treatment was administered to any group.⁴⁰ They found that amygdala reactivity was significantly reduced in the high-expectancy citalopram group, compared to the low-expectancy citalopram group, during the processing of sad versus neutral faces. This evidence further supports that the manipulation of expectancies in the context of experimental RCTs is associated with the activation of neurotransmitter systems and neural changes that when activated, result in real symptomatic improvements. Furthermore, this research supports minimizing expectancies in the context of RCTs can impact clinical outcome and potentially improve assay sensitivity.

One limitation of research on the placebo effect in the laboratory setting is that it is often conducted in healthy volunteers, without the administration of active treatments. Little is known about how those findings relate to the clinical trial setting where patients with diseases are provided active pharmacologic treatments. We were thus able to find only a small number of studies randomizing patients to both active treatment or placebo, as well as different placebo manipulation conditions. More studies with this design need to be performed for a more robust examination of placebo manipulation conditions across indications. Identifying the factors and circumstances that determine the magnitude of placebo responses should be a priority in order to inform the design, conduct, and interpretation of clinical trials.

Conclusions

Interventions that can reduce reported benefits in the placebo arms of RCTs while leaving the drug treatment arms unaffected improve signal detection. In turn, the time, costs, and efforts involved in bringing new treatments to market are reduced, hopefully leading to the increased drug discovery and ultimately higher availability of better treatments for patients. It is important to note, however, that outside the realm of RCTs such as in clinic, there is an opposite goal when it comes to the placebo response. Clinicians may aim to maximize a patient’s placebo response to increase symptom relief and benefit as part of a patients treatment.³⁹

Of the interventions designed to reduce placebo responses that have been tested in RCT studies, those that neutralize staff and subject expectations and improve the ability of subjects to accurately report symptom severity have shown the most promise. Research staff can minimize expectations by using neutral phrases when discussing expected treatment outcomes, providing neutral information about the study treatments, having limited time spent with study subjects, and emphasizing to subjects that the purpose of the trial is to determine whether a treatment is efficacious rather than to heal or cure. Interventions designed to educate and train subjects and research staff about what to expect in a clinical trial and how to complete study assessments can also improve the assay sensitivity and eventually, the likelihood of clinical trial successes. Training subjects to direct their attention inward when composing their symptom scores appears to reduce placebo responses by decreasing vulnerability to these external cues. Both approaches together – decreasing the cues that amplify expectancies while directing patients’ attention inward – may be one of the most effective approaches⁴⁰ although further research on this topic is needed.

Figure 2. Effect of treatment labeling on migraine pain.

Changes in self-reported migraine pain (0–10 numerical rating scale) between before (30 min after migraine onset) and after (2 h after migraine onset) treatment. P = “placebo” labeling, U = “Maxalt or placebo” labeling, “M” = “Maxalt” labeling. Error bars = 95% confidence intervals. Data extracted from Kam Hanson et al.²³

Figure 3. Effect of an Accurate Pain Reporting Training Program (APRT) on the placebo response, in a trial of pregabalin versus placebo in painful diabetic neuropathy.

The trained group completed the APRT program prior to treatment whereas the untrained group did not. Change in 24-hour average pain in the placebo group is the difference in pain score before and after treatment. NPS Numeric Pain Scale. Error bars represent standard error of the mean (SEM) Data extracted from Treister et al.¹⁸

Study Highlights.

• High Placebo responses are known to reduce the assay sensitivity of clinical trials. Various studies have been performed with the aim of increasing or decreasing the placebo response.

• Of the interventions designed to reduce placebo responses that have been tested in RCT studies, those that neutralize staff and subject expectations and improve the ability of subjects to accurately report symptom severity have shown the most promise.

• Implementing effective methods to reduce the placebo response can reduce the time, costs, and efforts involved in bringing new treatments to market, hopefully leading to the increased drug discovery and ultimately higher availability of better treatments for patients.