A Model of Placebo Response in Antidepressant Clinical Trials

In antidepressant trials for adults, the mean placebo response rate is 31%, compared with a mean medication response rate of 50%, and it has risen at a rate of 7% per decade over the past 30 years (1). In children and adolescents with major depression, rates of placebo response are even higher (a mean rate of 46%, compared with a mean medication response rate of 59%), and they have also been increasing over time (2). High placebo response reduces medication-placebo differences and leads investigators to make methodological modifications (e.g., the use of multiple study sites to increase sample size) that increase measurement error, both of which make it more difficult to demonstrate a statistically significant benefit of a putative antidepressant agent over placebo. Consequently, the average difference between medication and placebo observed in published antidepressant trials has decreased from an average of 6 points on the Hamilton Rating Scale for Depression (HAM-D) in 1982 to 3 points in 2008 (3). For most currently approved antidepressants, fewer than half of the efficacy trials filed with the U.S. Food and Drug Administration for regulatory approval found active drug superior to placebo (4, 5). While not all trials that fail to distinguish medication from placebo represent false negatives, meta-analyses of antidepressant trials suggest that high placebo response rather than low medication response explains most of the variability in drug-placebo differences (2). The increasing number of failed trials in recent years has made developing psychiatric medications progressively more time-consuming (an average of 13 years to develop a new medication) and expensive (estimates range from $800 million to $3 billion per new agent) compared with medications for non-CNS indications (6). These considerations contributed to recent decisions by several large pharmaceutical companies to reduce or discontinue research and development on medications for brain disorders, prompting warnings of “psychopharmacology in crisis” (7).

From a different perspective, enhancing the therapeutic components leading to placebo response may be a way of improving the clinical treatment of patients with depression. Major depression affects approximately 120 million people worldwide (including nearly 15 million American adults each year) and is a leading cause of illness-related disability (8). With currently available treatments, many patients will not experience sustained remission of their depression (9). As described further below, treatment with antidepressant medication involves much more than simply the dispensing of pills: an expectation of improvement is instilled in patients receiving treatment, and they are exposed to a health care environment that has many supportive and therapeutic features. These nonpharmacologic aspects of clinical management likely cause a substantial portion of observed medication and placebo response in clinical trials (10), yet they are typically not provided to the same extent in standard clinical practice. Whereas it was once believed that the therapeutic effects of expectancy and contact with health care staff were transient, more recent data show that 75% of placebo responders stay well during the continuation phase of treatment (11). Thus, optimizing the therapeutic components leading to placebo response has the potential to significantly improve treatment outcomes in clinical practice.

“Placebo response” can be defined as the change in symptoms occurring during a clinical trial in patients randomized to receive placebo. Placebo response, which is directly observed and quantified in a research study, is often conflated with the term placebo effect. A “placebo effect” can be defined as the therapeutic effect of receiving a substance or undergoing a procedure that is not caused by any inherent powers of the substance or procedure (12). For example, an inert cream has no inherent power to relieve pain, but being informed that the cream is an analgesic may result in a person feeling less pain when a painful stimulus is applied because the expectation of pain relief causes the release of endogenous opioids in the brain. While placebo effects are one contributor to the placebo response observed in a research study, many other factors may also influence placebo response.

To clarify this differentiation further, the sources of symptom change in antidepressant clinical trials can be grouped into categories (Figure 1, Table 1). Treatment factors comprise all the interventions and study procedures experienced by a patient in a clinical trial. Taking a pill that one believes to be an effective treatment for depression may generate a placebo effect (based on expectancy of improvement), while supportive contacts with study clinicians and undergoing medical procedures may also have therapeutic elements. Expectancy-based placebo effects and effects of the therapeutic setting may be moderated by the demographic and clinical characteristics of participating patients. Measurement factors represent sources of bias and error inherent in measuring depressive symptoms, and natural history factors reflect spontaneous improvement and worsening in the patient’s condition that is unrelated to the study procedures. The symptom change observed in a patient receiving placebo in a clinical trial (i.e., placebo response) is caused by the sum effects of these factors and their interactions. While large placebo effects may lead to large observed placebo response rates, it is also possible that high placebo response rates could be observed even where minimal placebo effects are operative (e.g., if there has been substantial improvement resulting from the therapeutic setting, regression to the mean, rater bias, or spontaneous fluctuation in illness severity).

In most clinical trials, investigators assess change in the severity of patients’ depressive illness on the basis of changes in measures of symptom severity that are self-reported by patients or elicited by trained raters. Measurements of depressive symptoms are subject to random error like any other measurement, but unlike more objective measurements, such as serum cholesterol level or blood pressure, depression symptom severity scores may be subject to additional sources of bias. Conceptually, one can distinguish cases in which a patient’s depression score decreased because symptoms actually improved from cases in which the score changed for other reasons.

Given the ethical difficulties associated with following the untreated course of depression, few prospective nonintervention studies using modern diagnostic criteria have been conducted that can be used to estimate the magnitude of spontaneous improvement that can be expected in acute depressive episodes (43). Alternative sources of information about the natural course of depressive disorders are psychotherapy studies in which a waiting list control group is used as a means of determining whether psychotherapy has an effect on depression beyond the passage of time. A recent meta-analysis of acute symptom change in waiting list control conditions found that individuals with major depression experience an average improvement of 4 points on the HAM-D (effect size, Cohen’s d=0.5) over a mean follow-up duration of 10 weeks (44). For comparison, meta-analyses of medication and placebo response in antidepressant clinical trials report standardized effect sizes of approximately 1.5 for medication conditions and 1.2 for placebo conditions (10). Thus, patients in waiting list control conditions experience approximately 33% of the improvement occurring with medication treatment and 40% of the improvement seen with placebo administration.

It is intriguing to speculate that natural history factors may be playing a greater role in antidepressant clinical trials over time as the population of patients enrolling in research studies changes. Most research participants in the 1960s and 1970s were recruited from inpatient psychiatric units, whereas current participants are symptomatic volunteers responding to advertisements (45). Studies are needed to compare the baseline characteristics, treatment response, and attrition rates of self-referred depressed patients with those who respond to advertisements. Natural history factors may be more important in the latter population, with the symptoms experienced by advertisement respondents being more variable and transient, resulting in greater placebo response rates compared with self-referred patients.

It may be possible to mitigate natural history factors by requiring longer durations of current illness for inclusion in antidepressant trials, but the spontaneous fluctuation of depression severity is less subject to investigator control than measurement factors. Epidemiologic studies show that patients with major depression typically experience symptoms for several months before seeking treatment (46). Patients are most likely to seek treatment during periods of increased stress or symptomatic worsening, so those who enroll in a clinical trial at such a time may experience a natural waning of symptoms or alleviation in the precipitating stressors irrespective of the treatment they are provided.

Expectancy-based placebo effects are dependent on intact cognition in order to generate placebo response (47), and a physiologic pathway must exist by which expectancy can influence the illness (e.g., minimal placebo effects are observed on tumor regression in metastatic cancer). However, among cognitively intact patients with depression, research has failed to identify consistent characteristics of patients likely to respond to placebo, giving rise to the term “the elusive placebo reactor” as early as the 1960s (48). Patient characteristics such as neuroticism, suggestibility, introversion, intelligence, and self-esteem have not been found to have significant associations with response to placebo (49). Such findings have led many to reject the hypothesis that certain people respond to placebos while others do not, and instead the consensus view arose that most individuals appear capable of being influenced by placebo effects under the appropriate conditions (50).

While evidence for patient characteristics influencing placebo response has been limited, features of the depressive illness being treated appear to influence the magnitude of placebo response observed in an antidepressant clinical trial. One of the most replicated findings in recent years has been that placebo response decreases with increasing severity of baseline depression scores. Both Kirsch et al. (51) and Fournier et al. (52) found that placebo response significantly declined as baseline HAM-D score increased (particularly as the score exceeded 25). Other studies suggest that patients suffering from major depressive disorder with psychotic features (53) and recurrent major depressive disorder (54) have lower rates of placebo response, as do elderly depressed patients with early-onset (prior to age 60) depression (55).

A fascinating and important question in pharmacologic treatments is how to conceptualize the contribution of placebo response to medication response. “Medication response” denotes the change in symptoms occurring during a clinical trial in patients randomized to receive medication. In contrast, the “medication effect” is the specific physiologic effect of the medication being studied on the target disorder (e.g., the effect of serotonin reuptake inhibition on major depressive disorder). By extension from Figure 1, medication response represents a combination of the specific medication effect with the previously described sources of placebo response (expectancy-based placebo effects, effects of the therapeutic setting, measurement factors, and natural history factors). The simplest and most common way to understand the nature of this combination is to assume that medication effects are additive with placebo response (i.e., placebo response is the same in the medication and placebo groups). In other words, if the observed placebo response rate in a clinical trial is 30% and the observed medication response rate is 50%, then the specific effects of the medication account for 50%–30%=20% response rate. If the specific effects of medication are relatively constant between similar patient samples, differences in observed medication response between different types of antidepressant studies (e.g., placebo-controlled, comparator, open) would presumably be caused by differences in expectancy-based placebo effects and other nonpharmacologic factors.

The logic of clinical trials and their application to clinical treatment are based on this “assumption of additivity” (56). Medications are approved for use after demonstration of a significant difference compared with placebo, and the “number needed to treat” is calculated relative to placebo response. It is therefore surprising that there is little or no evidence in the pharmacologic treatment of major depressive disorder to prove that medication effects and the placebo response are additive. Equally possible is that all of the observed medication response is caused by the specific medication effect without any contribution from placebo response, in which case the medication effect would be substantially larger than that calculated based on additivity. Alternatively, the presence of interactions between medication effects and sources of placebo response could result in the true contribution of medication effects to medication response being smaller than previously thought. For example, experiencing side effects known to be caused by the medication under study may “unblind” patients during a clinical trial, increasing their expectancy of improvement and leading to larger expectancy-based placebo effects.

To answer the question of how medication effects combine with sources of placebo response, it is necessary to conduct studies designed to test the assumption of additivity in antidepressant clinical trials. Kirsch (57) suggested that a modified form of the balanced placebo design may serve this function by randomizing a sample of patients with major depression to be told either that they are receiving active medication or that they are receiving placebo. Each group is then further randomized to actually receive medication or placebo, allowing researchers to estimate the true drug effect by comparing response in the “told placebo/receive medication” and “told placebo/receive placebo” groups. Utilizing an active placebo with this study design would make it possible to control for unblinding by side effects as well. While this study design represents a cogent approach to disentangling medication effects from placebo response, ethical concerns with the temporary deception of depressed patients have precluded its use to date.

Alternatively, new statistical approaches have been developed to more accurately estimate the causal effect of medication on depression using standard clinical trial data sets. Muthén and Brown (58) conceptually delimit four types of patients in a randomized controlled trial: those who would not respond to medication or placebo (“never responders”), those who would respond to both medication and placebo (“always responders”), those who would respond to medication but not placebo (“medication-only responders”), and those who would respond to placebo but not medication (“placebo-only responders”). Medication-only responders are the group of interest to pharmacologic researchers, since these subjects are assumed to be experiencing a true medication effect. Placebo-only responders are those who experience adverse effects from medication, and they are assumed to be few in number. Applying growth mixture modeling with maximum-likelihood estimation to an antidepressant clinical trial data set, Muthén and Brown report a method resulting in larger effect size estimates for medication-only responders compared with conventional analyses. While this approach appears promising, it is not clear that it accounts for the possibility of interactions between medication effects and placebo response described above.

It appears likely that the specific effect of medication is at least partially additive with placebo response, although its precise magnitude cannot be determined without knowing whether there are significant medication effect-by-placebo response interactions. The considerations discussed in this section make it clearer how rising placebo response makes it difficult to demonstrate statistically significant benefits of medications over placebo. First, rising placebo response leads to decreased medication-placebo differences (3), because the group of patients who do not respond to medication or placebo (never responders) places a ceiling on medication response in an antidepressant clinical trial. An increase in placebo response thereby causes a corresponding decrease in the medication-placebo difference. Second, investigators have attempted to compensate for the decreasing effect sizes observed for antidepressant medication by employing multiple study sites to obtain larger sample sizes. The greater measurement error associated with multicenter clinical trials leads to lower effect sizes for medication and often offsets the benefits of larger sample sizes.

Media coverage of placebo response has been used as a platform for critiques of the pharmaceutical industry and a stalking horse for questioning the efficacy of antidepressants. For example, a recent 60 Minutes report titled “Treating Depression: Is There a Placebo Effect?” (CBS, February 19, 2012) focused on a meta-analysis of 35 clinical trials of four antidepressants for major depression published in 2008 (51). Standardized mean effect sizes were calculated for the symptom change observed from baseline to end of treatment in participants randomized to antidepressant medication and placebo, resulting in an effect size of 1.24 for medication and 0.92 for placebo (p<0.001). However, when standardized mean differences between drug and placebo were examined as a function of initial depression severity, the difference in mean effect size exceeded the threshold for a “clinically significant” difference (defined as an effect size difference of 0.5) only among patients whose baseline HAM-D score was ≥28.

Criticism of this meta-analysis has focused on its relatively small and selective sample of the available studies. In addition, comparing overall mean differences between drug and placebo groups does not account for different trajectories of treatment response (59) and is relatively insensitive to large and significant changes experienced by subgroups of the sample (60). However, other analyses using different methods and samples have reported similar findings, suggesting that the mean HAM-D difference between antidepressant medication and placebo may be modest for patients with milder depression (52).

A frequent conclusion drawn from these studies by the media as well as the public is that “antidepressants do not work better than no treatment at all.” This conclusion represents an incorrect interpretation of the data and may have the dangerous public health consequence of dissuading patients with depression from accessing treatment. First, patients receiving antidepressants in clinical trials improve a great deal—the standardized mean effect size of 1.24 for antidepressant treatment is large. Second, being assigned to placebo in an antidepressant trial is far from “no treatment,” since it entails intensive contact with health care staff that greatly exceeds what is delivered in standard community treatment. More accurately, the meta-analytic findings discussed above suggest that eliciting placebo effects and providing a therapeutic setting are powerful treatments for mild to moderate depression, and the specific effects of antidepressant medication result in small additional decreases in HAM-D scores over these interventions.

In order to benefit from placebo effects and a therapeutic setting, patients must receive some type of treatment. Most psychiatrists would agree that deception and the covert prescription of placebo are unethical for patients with depression. The question facing clinicians then becomes the following: Of currently available treatments, what is the optimal way to elicit placebo effects and provide a therapeutic setting for patients with milder forms of depression? Some policy makers have advised utilizing psychotherapy as the initial treatment approach, presumably as a means of eliciting placebo response without the cost and potential side effect burden associated with antidepressants (61). While this may be reasonable for some patients, we would caution against blanket recommendations in favor of a thoughtful consideration of the risks and benefits of medication and psychotherapy for each individual. Psychotherapy is not always cheaper than antidepressant medication; analyses of their comparative cost-effectiveness have mostly shown minimal differences (62), and in some populations psychotherapy may be more expensive than medication as a treatment for depression (63). Like medication, psychotherapy may be associated with significant side effects, such as worsening symptoms and the fostering of dependence and regression (64). Finally, both psychotherapy and antidepressant medication may have therapeutic effects other than decreasing HAM-D scores that should be considered. For example, treatment with antidepressant medication has been shown to result in salutary personality change (decreased neuroticism) compared with cognitive therapy or placebo (65).

Valid evaluation of putative antidepressant agents requires that placebo response be minimized in the drug development setting, while the best care for patients with depression may involve maximizing placebo response in clinical treatments. Data presented in this review suggest that the next generation of research aimed at reducing placebo response should focus on limiting patient expectancy and the intensity of therapeutic contact in antidepressant clinical trials (Table 2). More research is needed to elucidate how antidepressant study design, the content and process of informed consent discussions, and clinician attitudes affect patient expectancy and treatment outcome. Designs in which patients have a higher probability of receiving placebo (i.e., 50%) may be preferable to designs in which patients are randomized to multiple active treatment arms and placebo. In terms of therapeutic contact intensity, future studies should examine whether clinical trials could be made more efficient and generalizable to clinical practice by reducing the amount of contact with health care staff to levels resembling those associated with treatment in the community. However, it must be determined whether acceptable attrition rates and patient safety can be maintained while reducing placebo response.

Conversely, the optimal strategy in clinical practice may be to combine active medication with a presentation and level of therapeutic contact designed to enhance placebo effects, leading to greater medication response. Updated clinical management techniques may involve educating patients about the effectiveness of the prescribed medication and utilizing a confident and enthusiastic interpersonal style. While efforts to identify the “elusive placebo reactor” have historically been unsuccessful, it is possible that demographic or clinical characteristics related to the formulation of treatment expectancies or patients’ experience of the treatment setting are associated with placebo response. Identifying patients who are likely to respond to these factors may permit them to be targeted with enhanced interventions. In developing these interventions, it may be useful to employ more “objective” measures of treatment response, such as neuroimaging, to ensure that these interventions actually improve depression rather than simply introducing response bias.

Studies of the neurobiology of placebo effects, although not within the scope of this review, may provide useful data on the mechanisms of placebo response. Functional MRI studies of expectancy-based placebo effects have provided converging evidence suggesting that the brain areas associated with generating and maintaining expectancies include prefrontal cortex subregions, the orbitofrontal cortex, and the rostral anterior cingulate cortex (66). In treatment for depression, response to placebo has been associated with region-specific alterations in brain metabolism that represent a subset of the areas associated with response to antidepressant medication (67). Research aimed at further elucidating the biological pathways that lead to placebo effects is under way, and results of these studies are likely to yield information about the pathophysiology of major depressive disorder and the mechanisms of action of antidepressant treatments.

Placebo response in antidepressant clinical trials is a fascinating and complex phenomenon worthy of scientific investigation. Investigating the mechanisms of placebo effects has the potential to illuminate the pathophysiology of depression as well as the active ingredients of clinical trials. The public health benefit of research on placebo effects is that knowledge of their mechanisms may permit the development of strategies to modulate their magnitude depending on the context, thereby improving signal detection in drug development studies and increasing antidepressant response in clinical treatment.