Placebo

A placebo is a substance or procedure designed to resemble a treatment while lacking the specific therapeutic component being investigated. Familiar examples include inactive tablets, saline injections, and simulated procedures. Placebos are used in randomized controlled trials to help distinguish the effects of an intervention from other influences on outcomes. Their role as research controls is distinct from the placebo effect: changes attributable to expectations, learned associations, and the circumstances in which treatment is delivered. (nccih.nih.gov)

Improvement after receiving a placebo does not, by itself, establish a placebo effect. Changes can also reflect the natural course of illness, regression to the mean, concurrent care, or differences in reporting. Research therefore distinguishes the overall placebo response from effects causally attributable to the placebo intervention and its context. Evidence varies by condition and outcome: effects on reported pain and some other symptoms do not establish a general ability to alter underlying disease. (pmc.ncbi.nlm.nih.gov)

Placebos connect experimental design, psychology, and neuroscience with questions of medical ethics. Their study encompasses treatment expectations, measurable biological responses, adverse contextual effects known as nocebo effects, and openly disclosed placebo administration. Their use in research is governed by requirements concerning scientific justification, informed consent, and protection against harm from withholding effective treatment. (nccih.nih.gov)

Definition and conceptual distinctions

A placebo is an intervention that resembles a treatment but lacks the treatment component whose specific action is being investigated. Its status is relative to a particular therapeutic question and outcome, rather than an assertion that it has no biological properties whatsoever. The US Food and Drug Administration’s May 2001 guidance on control groups describes a pharmaceutical placebo as an identical-appearing treatment without the test drug. Thus, a tablet can reproduce the experience of taking medication while omitting the molecule under investigation. “Inactive” in this setting refers to the intended treatment-specific action; it does not establish that every ingredient, sensory stimulus, or element of administration is physiologically inert. (fda.gov) For example, the placebo infusion in the Trial to Assess Chelation Therapy contained 500 mL of normal saline and 1.2 percent dextrose: it was a comparator lacking the chelation mixture, not an absence of administered material. (nccih.nih.gov)

Three concepts distinguish the intervention from its observed and causal consequences. A placebo intervention is the administered substance or simulated treatment, together with its presentation. A placebo response, in a common methodological usage, is the change observed after placebo administration—for example, a reduction in a symptom score between baseline and follow-up. A placebo effect is the change causally attributable to receiving the placebo intervention or to particular features of its treatment context. These terms have not been used consistently throughout the literature, but the distinction prevents an observed sequence—improvement following administration—from being mistaken for proof of causation. (pmc.ncbi.nlm.nih.gov)

The placebo response can include spontaneous improvement, fluctuating symptoms, regression to the mean, concurrent treatments, and changes in reporting or assessment, as well as contextual effects. Consequently, improvement within the placebo arm of a randomized controlled trial does not by itself measure a placebo effect. Comparing placebo administration with a suitable no-treatment condition addresses a different question from comparing an active drug with placebo: the former investigates consequences of the placebo intervention, whereas the latter estimates benefit beyond the matched control. Even a no-treatment group may receive clinical attention and outcome assessments, so it need not represent an entirely untouched disease course. (fda.gov)

Treatment context includes information about expected benefit, previous treatment experiences, the therapeutic ritual, and interaction with a healthcare professional. These features can influence outcomes during active treatment as well as during placebo administration; contextual effects are therefore not exclusive properties of dummy pills. The National Center for Complementary and Integrative Health identifies anticipation of benefit and practitioner–patient interaction as potential contributors. In pain research, instructions and learned associations can change expectations and subsequent experience, although contextual and drug-specific effects should not automatically be assumed to combine as a simple, fixed sum. (nccih.nih.gov)

A contextual contribution does not imply that symptoms are imaginary. Pain experience involves both physiological processing and psychological influences, including expectations. In two experiments published in 2004, Tor D. Wager and colleagues associated placebo analgesia with changes in pain-related brain activity measured by functional magnetic resonance imaging. Such findings support the distinction between lacking an active pharmaceutical ingredient and lacking consequences for symptom experience. They do not establish that every reported improvement reflects the same mechanism, or that symptom relief demonstrates reversal of the underlying disease. (pubmed.ncbi.nlm.nih.gov)

Historical development

The older medical meaning of placebo concerned the physician’s purpose in prescribing, rather than the intervention’s role in an experiment. In 1811, the term was defined as a medicine given more to please than to benefit the patient. This usage did not establish that every such preparation was chemically inert or incapable of producing physiological changes. The distinction remained visible in the mid-twentieth century: John H. Gaddum argued in 1953 that experimental comparator tablets were better called “dummies,” because their function was not necessarily to please their recipients. Henry K. Beecher quoted this observation in 1955, documenting the transition from a prescribing category to a tool for separating pharmacological action from suggestion and other influences on treatment outcomes. (jamanetwork.com)

Beecher’s “The Powerful Placebo,” published in JAMA on December 24, 1955, brought quantitative prominence to the subject. It synthesized 15 studies involving 1,082 patients, encompassing postoperative pain, angina, headache, seasickness, anxiety, and other conditions. Beecher reported an average proportion satisfactorily relieved of $35.2 \pm 2.2\%$. He used these observations both to argue that placebo-associated improvement deserved investigation and to emphasize the need for controlled, masked assessments of treatments. The paper became an influential reference, but its figure was not a universal effect size: it combined responses from different conditions, interventions, and criteria for relief, rather than estimating a single causal effect applicable across medicine. (dcscience.net)

Later reassessment sharpened the distinction between improvement after placebo administration and improvement caused by it. In 1997, Gunver S. Kienle and Helmut Kiene reexamined Beecher’s evidence and identified alternative explanations, including spontaneous improvement, symptom fluctuation, regression to the mean, and additional treatment. Their criticism concerned the attribution of observed changes to placebo, not merely the numerical size of those changes. This methodological issue helped separate two research questions: whether an active treatment outperformed its placebo comparator, and whether receiving a placebo produced better outcomes than receiving no treatment. The former could establish treatment efficacy without resolving the latter. (pubmed.ncbi.nlm.nih.gov)

By 2001, these questions were represented in distinct methodological programmes. The US Food and Drug Administration’s May 2001 E10 guidance addressed selection of control groups and what different trial designs could demonstrate about efficacy. In a separate investigation published that month, Asbjørn Hróbjartsson and Peter C. Gøtzsche systematically reviewed trials that randomly assigned participants to placebo or no treatment. They identified 130 eligible trials, of which 114 supplied relevant outcome data. Unlike Beecher’s aggregation of improvement within placebo groups, this approach used a between-group contrast to investigate the consequences of placebo administration itself. It therefore made causal inference about placebo a separate task from using placebo controls in a randomized controlled trial of an active treatment. (fda.gov)

Experimental neuroscience supplied another line of investigation. In 2001, Raúl de la Fuente-Fernández and colleagues used positron emission tomography to obtain evidence of endogenous dopamine release in the striatum of participants with Parkinson’s disease following placebo administration. In 2004, Tor D. Wager and colleagues reported two functional magnetic resonance imaging experiments in which placebo analgesia was associated with reduced activity in pain-sensitive regions, including the thalamus, insula, and anterior cingulate cortex, and increased prefrontal activity during pain anticipation. These studies made neurochemical and brain-activity changes experimentally measurable; they did not establish disease modification in Parkinson’s disease or a complete causal pathway for every form of placebo-related pain relief. (pubmed.ncbi.nlm.nih.gov)

Forms and construction of placebo controls

Pharmaceutical placebos take the form of tablets, capsules, injections, or inhalers constructed to resemble the treatment being tested while omitting the investigational drug. Matching extends beyond colour and shape to weight, taste, smell, packaging, and labelling: a recognizable difference can reveal the assignment even when the products look similar. Administration route and treatment schedule must likewise reproduce the relevant treatment experience, including the number and timing of doses. The US Food and Drug Administration’s E10 guidance specifies close physical matching, while its manufacturing guidance calls for precautions to ensure that the investigational substance is absent from the placebo preparation. (fda.gov)

An active placebo adds another level of resemblance by producing noticeable effects without the therapeutic action under investigation. Its purpose is to imitate sensations or adverse effects that might otherwise identify the experimental drug. Atropine, which produces anticholinergic effects, has been used in trials of tricyclic antidepressants for this reason. R. Thomson’s 1982 retrospective examination of such studies also considered whether atropine itself might affect depressive symptoms, illustrating the difficulty of assuming that a physiologically active control is therapeutically neutral. An active placebo is distinct from an active comparator: the latter is a treatment with established efficacy, used to compare therapeutic performance rather than merely reproduce treatment cues. (pubmed.ncbi.nlm.nih.gov)

Procedural controls replace a dummy formulation with a simulated intervention. In acupuncture, needling away from recognized acupuncture points may still stimulate tissue, so changing the location does not necessarily remove relevant physiological effects. In 1998, Klaus Streitberger and Johannes Kleinhenz introduced a nonpenetrating placebo needle. Their crossover experiment involved 60 volunteers: 47 reported a sensation of penetration with the placebo needle, compared with 54 during actual acupuncture, and none suspected that the placebo needle had not penetrated the skin. These findings established credibility in that experimental setting, rather than demonstrating that every sensory component of the simulation was inactive. (pubmed.ncbi.nlm.nih.gov)

Different placebo forms can also reproduce different treatment rituals within one randomized controlled trial. Michael E. Wechsler and colleagues’ 2011 asthma experiment used both a placebo inhaler and a validated sham acupuncture needle alongside inhaled albuterol and a no-intervention condition. The active and placebo inhalers were administered double-blind, whereas sham acupuncture was single-blind. Participants judged treatment to be genuine after 66% of placebo-inhaler administrations and 85% of sham-acupuncture administrations. Thus, credibility was measured rather than assumed, and the two simulations were not equally convincing despite both functioning as placebo interventions. (nejm.org)

In surgery, reproducing credibility can require elements that retain procedural risks. J. Bruce Moseley and colleagues’ 2002 knee-arthroscopy trial assigned 180 patients to debridement, lavage, or placebo surgery. The sham procedure included three 1-centimetre skin incisions, knee manipulation, requests for instruments, and splashing saline to imitate lavage sounds, but no arthroscope entered the joint. Sham recipients received intravenous sedation and an opioid, rather than the general anesthesia used in the operative groups. This construction preserved many observable features while omitting the intra-articular procedure; it did not eliminate exposure to drugs or skin injury. Such controls therefore require a distinction between features needed for a convincing simulation and components capable of producing physiological effects or harm. (nejm.org)

Functions in clinical trials

In a randomized controlled trial, placebo controls, random allocation, and masking address different sources of bias. Random allocation tends to balance known and unknown prognostic factors between groups, rather than allowing treatment selection to depend on disease severity or clinician preference. Allocation concealment protects this process before assignment: recruiting staff cannot know the next allocation when deciding whether to enroll a participant. Centralized assignment is one method of preventing such foreknowledge. Masking, or blinding, operates after assignment by keeping participants, treating staff, or outcome assessors unaware of the assigned intervention. Thus, a trial can conceal allocation even when subsequent treatment cannot be masked. (hhs.gov)

A credible placebo makes masking possible by reproducing the treatment’s observable features without supplying the component under investigation. Keeping treatment identity unknown reduces expectation-related differences in symptom reporting, decisions about additional care, and observer judgments. In a placebo-controlled comparison, the between-group difference estimates the benefit of the test treatment relative to its matched control under the trial’s conditions; improvement within the placebo group alone does not establish a causal placebo effect. A no-treatment-controlled trial instead compares receiving the intervention with not receiving it, but participants generally know whether they are being treated. Differences in expectations, retention, and clinical management can therefore contribute to the observed contrast. Neither comparison automatically establishes superiority over an existing effective treatment. These distinctions are set out in the FDA’s May 2001 adoption of the ICH E10 guidance on control-group selection. (fda.gov)

An active-controlled trial directly compares the investigational treatment with another treatment, commonly an established therapy. A superiority design asks whether the new treatment produces better outcomes; a non-inferiority design asks whether it is worse by less than a prespecified, clinically justified margin. Failure to find a statistically significant difference does not, by itself, establish equivalence or non-inferiority. The FDA’s November 2016 guidance emphasizes assay sensitivity—the ability of a trial to distinguish an effective treatment from a less effective or ineffective one. If an active comparator fails to exert its expected effect in the study, similar outcomes could reflect two ineffective treatments rather than two effective ones. Non-inferiority conclusions consequently require evidence supporting the comparator’s effectiveness under sufficiently comparable trial conditions. (fda.gov)

In an add-on design, both groups receive background treatment, with the investigational agent added in one group and placebo added in the other. The comparison measures the incremental benefit of adding the agent, not its efficacy as a replacement for background care. For example, a trial published online in 2016 randomized 333 participants with type 2 diabetes inadequately controlled on metformin and linagliptin to empagliflozin 10 mg, empagliflozin 25 mg, or placebo for 24 weeks; all continued metformin and linagliptin. Placebo allocation therefore meant withholding the additional study drug, not withdrawing all glucose-lowering treatment. (fda.gov)

Double-dummy designs preserve masking when treatments have different formulations or delivery methods. In an illustrative tablet-versus-injection comparison, one group receives active tablets and placebo injections, while the other receives placebo tablets and active injections. Both groups undergo both administration schedules. This remains an active-controlled trial if every participant receives an active comparator or the investigational treatment. The European Medicines Agency’s 2017 semaglutide assessment documents double-dummy masking in SUSTAIN 2, which compared injectable semaglutide with oral sitagliptin. Matching administration does not prevent every form of unmasking: recognizable adverse effects or laboratory changes may reveal treatment identity. Independent masked outcome assessment can reduce resulting observer bias, although it cannot restore participants’ uncertainty once they have inferred their assignment. (fda.gov)

Identifying and measuring placebo effects

Improvement after placebo administration is not, by itself, evidence that the placebo caused the improvement. A before-and-after comparison combines treatment-context effects with spontaneous recovery, fluctuating symptoms, regression to the mean, and effects of concurrent care. Regression to the mean is particularly relevant when enrollment requires an unusually high symptom score: a subsequent measurement may be less extreme even without an effective intervention. This statistical tendency differs from biological recovery, although both can produce apparent improvement. In a 2009 analysis, Lasse Theis Krogsbøll, Asbjørn Hróbjartsson, and Peter C. Gøtzsche examined 37 three-arm trials involving 2,900 participants. Standardized changes from baseline were −0.24 in no-treatment groups, −0.44 in placebo groups, and −1.01 in active-treatment groups, with negative values indicating improvement. These results illustrate why the entire change in a placebo group cannot be attributed to placebo administration; they are not universal proportions applicable to every treatment or condition. (pmc.ncbi.nlm.nih.gov)

A three-arm randomized controlled trial assigns participants to active treatment, placebo, or no study treatment. The active-versus-placebo contrast estimates the incremental effect of the treatment component absent from the placebo, provided that administration, accompanying care, and expectations are sufficiently comparable. The placebo-versus-no-treatment contrast estimates the effect of assignment to the placebo intervention and its associated context. Active treatment versus no treatment estimates the total effect of the active-treatment package relative to that comparator. These are distinct questions: an active-versus-placebo comparison alone does not measure the placebo effect. Moreover, the contrasts do not establish that pharmacological and contextual effects are independent, separable biological quantities. FDA guidance identifies no-treatment controls as particularly vulnerable to expectation-related differences because participants and providers know whether treatment has been assigned. (fda.gov)

“No treatment” generally means absence of the designated study intervention, not absence of all influences on the outcome. Participants may still interact with study personnel, undergo assessments, and receive other care. Such a group therefore does not necessarily represent untouched natural history. Knowledge of allocation can also change symptom reporting, use of additional treatment, or withdrawal from the study. Consequently, a placebo-versus-no-treatment difference may combine changes in experienced symptoms with changes in how symptoms are reported. A no-treatment group controls for several alternative explanations of improvement, but does not automatically isolate a purely physiological placebo mechanism. (pmc.ncbi.nlm.nih.gov)

Measurement must distinguish patient-reported symptoms, observer ratings, and physiological outcomes. A patient’s pain rating measures experienced pain; a clinician’s rating adds observer judgment; a laboratory measurement or biomarker measures a specified physiological variable rather than the whole experience of illness. These categories are not interchangeable. Observer ratings may also depend on patient cooperation or disclosure, while awareness of allocation can influence both patient reports and staff assessments. Hróbjartsson and Gøtzsche’s 2010 review accordingly distinguished patient-reported from observer-reported outcomes and examined masking of outcome assessors. Such distinctions help identify susceptibility to reporting and assessment bias without treating every subjective improvement as merely a reporting artifact. (pmc.ncbi.nlm.nih.gov)

Interpretation requires an effect estimate and its confidence interval, not simply a declaration of statistical significance. The interval indicates statistical imprecision but does not account for every source of uncertainty, such as biased measurement or inconsistent findings across studies. Differences in conditions, placebo procedures, follow-up periods, and outcome scales can make a pooled estimate an inadequate description of any particular setting. A small $P$ value does not establish clinical importance: a precisely estimated change may be too small to matter to patients, whereas a wide interval may leave substantial benefit and negligible benefit both compatible with the data. Clinical importance is therefore assessed using the magnitude of change on an interpretable scale and an appropriate patient-important threshold, rather than the significance threshold alone. (cochrane.org)

Expectations, learning, and interpersonal context

Treatment expectations concern anticipated outcomes, such as whether an intervention will relieve pain and how much relief it will provide. They can be generated by verbal descriptions and strengthened by previous experiences that appear to confirm those descriptions. Tor D. Wager and colleagues demonstrated this experimentally in 2004 using identical skin creams presented as different treatments. In an experiment involving 24 participants, one application was described as an analgesic that would reduce the pain of electric shocks, while another was described as an ineffective control. Participants reported less pain under the analgesic description, although the cream was unchanged. A second experiment reinforced the treatment description by secretly reducing heat intensity during an initial demonstration; subsequent testing used equal stimulus intensities at placebo-treated and control-treated sites. This design combined verbal information with an apparent experience of treatment success rather than treating expectation as a belief formed independently of experience. (columbia.edu)

Classical conditioning provides another account of how treatment context acquires effects. A treatment cue—such as a capsule’s appearance or the circumstances of its administration—is repeatedly paired with an active drug or symptom relief. Later presentation of that cue without the original active component can elicit a learned response. Conditioning and consciously articulated expectations are not mutually exclusive: learning that a treatment reliably precedes relief can itself create an expectation of relief. Conversely, verbal information may modify the significance of an established association. The distinction is therefore between experimentally separable processes, not necessarily between two independent mechanisms operating in every recipient. (pmc.ncbi.nlm.nih.gov)

Fabrizio Benedetti and colleagues investigated this distinction in 2003 by combining prior drug exposure with opposing verbal suggestions. In experimental arm pain, suggestions that an intervention would increase or decrease pain could override the effects of the conditioning procedure. Hormonal responses behaved differently. After conditioning with sumatriptan, which increased growth hormone and decreased cortisol secretion, placebo administration reproduced changes in those directions even when participants received contrary verbal suggestions. Suggestions alone did not produce corresponding hormonal changes. These findings showed that, under the conditions studied, a learned physiological response did not require a matching consciously expressed expectation. They do not establish that all pain responses depend exclusively on expectation or that all hormonal responses are conditioned. (pmc.ncbi.nlm.nih.gov)

The clinical encounter also supplies information through the practitioner’s manner and the treatment ritual. Ted J. Kaptchuk and colleagues examined these components in a 2008 randomized controlled trial involving 262 adults with irritable bowel syndrome. Participants were assigned to a waiting list, sham acupuncture with limited practitioner interaction, or the same sham procedure accompanied by an interaction deliberately augmented with warmth, attention, and expressed confidence. After three weeks, adequate symptom relief was reported by 28%, 44%, and 62% of the respective groups. Average global-improvement scores on a seven-point scale were 3.8, 4.3, and 5.0. The comparison between the two sham-treatment groups isolated the added interpersonal package more closely than a simple comparison of treatment with a waiting list. (pubmed.ncbi.nlm.nih.gov)

That experiment did not isolate attention alone: the augmented encounter simultaneously changed warmth, communication, and confidence. Nor did it establish that practitioner interaction explains every placebo effect or that the same differences would occur in other conditions. Its symptom-based outcomes supported an effect of the specified encounter on reported improvement, without demonstrating a single psychological pathway. The waiting-list comparison likewise included differences in treatment ritual and interpersonal contact, rather than measuring an isolated effect of the sham needle itself. (pubmed.ncbi.nlm.nih.gov)

Neurobiological findings

Evidence for endogenous opioid involvement in placebo analgesia comes partly from experiments using naloxone, an opioid-receptor antagonist, to interrupt pain relief. In 1978, Jon D. Levine, Newton C. Gordon, and Howard L. Fields studied postoperative pain after extraction of impacted mandibular third molars. Participants received naloxone or placebo under randomized, double-blind conditions three and four hours after surgery. Among patients initially given placebo, subsequent naloxone increased pain in those classified as placebo responders but produced no additional increase in nonresponders. This pattern supported the hypothesis that endogenous opioid activity contributed to the observed analgesia. The experiment manipulated a candidate mechanism rather than merely recording a neural correlate, although its findings concerned a particular postoperative setting, not every form of placebo response. (pubmed.ncbi.nlm.nih.gov)

Opioid dependence also varies with how analgesia is induced. In 1999, Martina Amanzio and Fabrizio Benedetti used experimentally induced ischemic arm pain to compare responses elicited by expectations of relief, prior drug conditioning, or both. Naloxone completely blocked placebo analgesia induced by expectation cues and analgesia following conditioning with morphine, including morphine conditioning without explicit expectation cues. By contrast, placebo analgesia after conditioning with the nonopioid analgesic ketorolac was insensitive to naloxone when conditioning was used alone; adding expectation cues produced a response that naloxone blocked only partially. These differences demonstrated that even within experimental pain, superficially similar placebo outcomes can recruit different pharmacological mechanisms. Endogenous opioids therefore constitute one experimentally supported pathway, not a universal biochemical explanation. (pubmed.ncbi.nlm.nih.gov)

In 2004, Tor D. Wager and colleagues reported two functional magnetic resonance imaging experiments separating anticipation of pain from its subsequent experience. The experiments used electrical shocks and thermal stimulation, respectively, with identical inactive creams presented as either analgesic treatment or an ineffective control. Placebo-related reductions in reported pain were associated with decreased activity in pain-responsive regions, including the thalamus, insula, and anterior cingulate cortex. During anticipation, placebo treatment was associated with increased prefrontal activity. The temporal distinction was important: treatment information was associated with neural changes before the painful stimulus, as well as with altered responses during stimulation. These findings supported an influence on pain processing rather than an explanation confined to changes in verbal reporting. (researchgate.net)

The imaging results nevertheless did not establish a complete causal sequence from expectation through prefrontal activity to analgesia. Functional imaging identified regional signal changes and relationships with pain ratings; it did not selectively interrupt the implicated regions or identify the neurotransmitter responsible for each change. The anticipatory prefrontal findings were consistent with top-down modulation, but the proposed direction of influence remained an interpretation of the observed associations. In terms of causal inference, such imaging evidence and antagonist experiments answer different questions: one localizes activity associated with a response, while the other tests whether blocking a signaling system changes that response. (pubmed.ncbi.nlm.nih.gov)

A distinct neurochemical finding came from Raúl de la Fuente-Fernández and colleagues in 2001. Using positron emission tomography, they studied six patients with Parkinson’s disease under a blinded placebo–apomorphine protocol and an open condition without placebo. The tracer $[^{11}\mathrm{C}]$raclopride competes with endogenous dopamine for receptor binding, allowing reduced tracer binding to serve as an indirect indicator of dopamine release. Compared with open baseline measurements, placebo administration reduced binding potential by approximately 17% in the caudate nucleus and 19% in the putamen. The experiment thus demonstrated a measurable neurochemical response in the striatum during an anticipated treatment encounter. Its small, acute imaging design did not test preservation of dopaminergic neurons or long-term disease progression; dopamine release in this setting was not evidence of disease modification. (pubmed.ncbi.nlm.nih.gov)

Clinical evidence and limits of generalization

A 2010 Cochrane review by Asbjørn Hróbjartsson and Peter C. Gøtzsche examined randomized trials comparing placebo interventions with no treatment, rather than treating improvement within placebo groups as evidence of efficacy. Its database searches extended through March 2008, although the CENTRAL search covered Issue 4 of 2007. Of 234 eligible trials, 202 supplied outcome data for the main analyses, covering 60 clinical conditions. Only 16 trials—8% of those analysed—were judged to have a low risk of bias. The authors found no evidence that placebo interventions generally produced important clinical benefits; this conclusion nevertheless allowed for effects on particular outcomes in particular settings. (pmc.ncbi.nlm.nih.gov)

The results differed according to how outcomes were measured. Across 158 trials with continuous outcomes, involving 10,525 participants, the pooled standardized mean difference was −0.23, with negative values indicating benefit. It was −0.26 for patient-reported outcomes and −0.13 for observer-reported outcomes. For continuous measures of pain, the estimate was −0.28, with a 95% confidence interval of −0.36 to −0.19; for nausea, it was −0.25, with an interval of −0.46 to −0.04. These are standardized differences, not percentages of symptom relief. Analyses of binary outcomes did not establish statistically significant effects for pain or nausea, and their confidence intervals were wide. Thus, evidence of benefit depended partly on the outcome and its measurement. (cochrane.org)

Generalization was further limited by variation between studies. Pain effects ranged from negligible to clinically important even among trials assessed as having low risk of bias, whereas the pooled nausea effect was small but more consistent. Larger estimates were associated with physical placebo procedures, smaller trials, and studies explicitly designed to investigate placebo effects. Differences between small and large studies, together with variation in conditions and interventions, made the overall pooled estimate unsuitable as a universal placebo effect size. Because participants could distinguish receiving an intervention from receiving none, patient-reported differences could also include altered reporting, not solely altered symptoms; the review could not reliably separate these contributions. (cochrane.org)

A 2011 experiment by Michael E. Wechsler and colleagues demonstrated this outcome dependence in asthma. Forty-six patients were randomized in a crossover pilot study comparing inhaled albuterol, a placebo inhaler, sham acupuncture, and no intervention; 39 completed the study. Each intervention was administered once per block over three blocks, giving 12 visits per participant. During each visit, repeated lung-function measurements over two hours assessed forced expiratory volume in one second, $\mathrm{FEV}_1$, alongside reported improvement. Mean increases in $\mathrm{FEV}_1$ were 20.1% after albuterol, 7.5% after placebo inhalation, 7.3% after sham acupuncture, and 7.1% without intervention. Albuterol therefore improved measured airflow substantially more than either placebo procedure. (nejm.org)

Reported improvement followed a different pattern: 50% after albuterol, 45% after placebo inhalation, and 46% after sham acupuncture, compared with 21% without intervention. The placebo procedures increased perceived improvement without increasing measured airflow beyond the no-intervention condition. Similar symptom ratings therefore did not establish equivalent therapeutic efficacy: the active drug had a physiological effect that the placebo procedures did not reproduce. The small, short-term experiment also did not establish effects on subsequent exacerbations or long-term asthma control. Its findings specifically concerned the divergence between immediate perceived improvement and measured bronchodilation. (pmc.ncbi.nlm.nih.gov)

Nocebo effects and adverse outcomes

A nocebo effect is an adverse change attributable to negative expectations or other features of the treatment context, rather than to the specific action of the treatment under investigation. It can involve increased symptoms or anticipated side effects and is not restricted to people receiving placebos. Information about possible adverse effects, previous treatment experiences, and the circumstances of administration can contribute to such responses. A symptom occurring after placebo administration, however, is not automatically a nocebo effect: its timing alone does not establish causation. Disease progression, ordinary symptom fluctuations, unrelated illness, and concurrent treatments can also produce adverse events in a placebo group. (pmc.ncbi.nlm.nih.gov)

This distinction affects the interpretation of randomized controlled trials. The total frequency of adverse events among placebo recipients describes what occurred under that trial condition, not how much harm negative expectations caused. Experiments can isolate contextual influences more directly by varying information while holding the administered substance and symptom-provoking stimulus constant. An untreated comparison group can also help distinguish treatment-context effects from changes that would occur without the study intervention, although differences in attention and outcome reporting remain relevant. Consequently, adverse-event rates in placebo arms cannot by themselves supply a causal estimate of the nocebo effect. (pmc.ncbi.nlm.nih.gov)

A specific experimental manifestation is nocebo hyperalgesia: increased pain following suggestions that a treatment will worsen it. In a 2006 study, Fabrizio Benedetti, Martina Amanzio, Sergio Vighetti, and Giovanni Asteggiano investigated experimentally induced ischemic arm pain in healthy volunteers. Verbal suggestions of increased pain produced hyperalgesia accompanied by elevated plasma adrenocorticotropic hormone and cortisol, indicating activation of the hypothalamic–pituitary–adrenal axis. Diazepam antagonized both the additional pain and the hormonal response. Proglumide, an antagonist of cholecystokinin (CCK) type-A and type-B receptors, completely blocked the additional pain without blocking the hormonal response. Neither drug reduced baseline pain under non-nocebo conditions. These findings implicated anxiety-related processes and CCK signaling in that experimental model, while distinguishing the mechanisms of pain amplification from the accompanying stress-hormone response; they do not establish one mechanism for every adverse contextual effect. (pubmed.ncbi.nlm.nih.gov)

Direct harms from a placebo intervention belong to a different causal category. Pharmaceutical excipients can cause allergy or intolerance even when they have no intended therapeutic action; a reaction to an ingredient is not an expectation-mediated nocebo effect. Likewise, sham surgery may omit the operative step under investigation while retaining physically consequential procedures. In J. Bruce Moseley and colleagues’ 2002 knee osteoarthritis trial, which randomized 180 patients, placebo surgery included three one-centimeter skin incisions, an intravenous tranquilizer, and an opioid, although no arthroscope entered the knee. The designation “placebo” therefore did not mean that the control involved neither tissue injury nor pharmacological exposure. (pmc.ncbi.nlm.nih.gov)

Harm can also arise from withholding effective treatment rather than from expectations or placebo ingredients. The FDA’s May 2001 E10 guidance identifies death or irreversible morbidity as particular concerns when an available treatment prevents serious harm. Such deterioration must be distinguished from contextual symptom worsening: it reflects the loss of effective treatment’s protection. Placebo-controlled add-on designs, in which both groups retain standard therapy, illustrate how placebo allocation can be separated from omission of established care. (fda.gov)

Open-label placebo research

Open-label placebo administration explicitly informs recipients that the intervention is a placebo rather than concealing its identity or presenting it as an active drug. This differs from an unmasked trial of an active treatment: disclosure concerns not merely which treatment has been assigned, but the absence of the specific therapeutic medication being investigated. Research therefore asks whether a disclosed placebo, delivered with an explanatory rationale and a treatment ritual, can improve outcomes relative to a comparison condition. Knowledge that a pill contains no medication does not necessarily exclude expectations of benefit from taking it. (journals.plos.org)

A 2010 randomized controlled trial by Ted J. Kaptchuk and colleagues examined this question in 80 adults with irritable bowel syndrome (IBS). Conducted at Beth Israel Deaconess Medical Center in Boston between August 2009 and April 2010, the study lasted three weeks and assigned 37 participants to open-label placebo and 43 to a no-treatment control. Participants met Rome III diagnostic criteria and had scores of at least 150 on the IBS Symptom Severity Scale. Both groups received supportive practitioner contact of similar duration. Existing IBS medications could continue at stable doses, so “no treatment” meant no additional experimental treatment, not withdrawal of usual care. (journals.plos.org)

Before randomization, both groups received an approximately 15-minute explanation identifying the pills as inactive and presenting four propositions: placebo effects can be powerful; responses to pill-taking can become automatic through learned associations; a positive attitude may help but is unnecessary; and consistent pill-taking matters. These statements constituted the study’s positively framed rationale, rather than mechanisms demonstrated by the experiment. The placebo group received gelatin capsules containing Avicel, a pharmaceutical excipient, in bottles labeled “placebo pills,” with instructions to take two capsules twice daily. At the midpoint visit, this group received a brief reminder of the explanation. (journals.plos.org)

Outcomes were assessed on days 11 and 21 using patient-completed questionnaires assisted by an assessor unaware of allocation. The primary measure, the seven-point IBS Global Improvement Scale, ranged from substantially worse to substantially improved, with 4 indicating no change. At day 21, mean scores were $5.0 \pm 1.5$ in the placebo group and $3.9 \pm 1.3$ in controls, with $p=0.002$. Adequate symptom relief was reported by 59% and 35%, respectively. Mean symptom-severity reductions were 92 versus 46 points; the between-group difference in quality of life did not meet the study’s conventional statistical significance threshold ($p=0.08$). These were symptom-based outcomes, not demonstrations of altered intestinal physiology. (journals.plos.org)

The design limits causal inference about individual components of the intervention. Participants knew their assignments, leaving reported outcomes susceptible to expectations and reporting influences despite assessor masking. The comparison tested disclosed pill-taking with reinforcing explanations, not the capsule independently of its presentation; it therefore did not establish whether learning, conscious expectation, or another contextual process accounted for the difference. The small sample, single setting, recruitment partly through advertisements for a mind–body study, and three-week observation period also restricted generalizability and estimates of durability. The investigators characterized the experiment as a proof-of-principle pilot and explicitly stated that replication and longer follow-up were necessary before clinical implementation; its findings did not establish routine open-label placebo prescribing or equivalence to an effective IBS medication. (journals.plos.org)

Ethics and participant protection

Paragraph 33 of the 2024 Declaration of Helsinki, adopted by the World Medical Association in October 2024, establishes comparison with the best proven intervention as the default for testing a new intervention. It permits placebo or no-intervention controls when no proven intervention exists. Where an effective intervention is available, the exception requires both compelling, scientifically sound methodological reasons and assurance that withholding the best proven intervention will not expose participants to additional risks of serious or irreversible harm. Methodological usefulness alone therefore does not justify withholding effective care; the declaration explicitly cautions against abuse of this exception. (wma.net)

Informed consent to a placebo-controlled randomized controlled trial concerns the study arrangement, not advance knowledge of the participant’s eventual assignment. Participants must be told about the different research arms, including which interventions will or will not be provided, and understand that they may receive placebo. Authorized masking can then keep the assignment unknown without concealing the possibility of placebo allocation. Consent also requires understandable information about anticipated benefits, risks, and burdens, and the freedom to refuse or withdraw without reprisal or loss of standard care. Consent does not transfer responsibility for participant protection from researchers to participants. (code-medical-ethics.ama-assn.org)

Independent ethical review and continuing risk assessment provide protections beyond consent. Under paragraph 23 of the declaration, a research ethics committee must approve the protocol before research begins and have authority to monitor the study, require changes, suspend it, or withdraw approval. Researchers must continuously assess and document risks and reconsider continuation when these outweigh potential benefits. For placebo studies, the American Medical Association additionally emphasizes interim analysis and monitoring so that participants do not remain on placebo longer than scientifically necessary. Its research guidance excludes placebo controls when an available therapy would prevent or slow illness likely to cause death or irreversible damage. (wma.net)

Study design can reduce the risks associated with placebo allocation. The FDA’s May 2001 E10 Choice of Control Group and Related Issues in Clinical Trials guidance describes add-on designs in which both groups receive established background treatment, with the investigational intervention or placebo added to it. Prespecified early-escape rules can remove participants from ineffective therapy when deterioration reaches a defined threshold—for example, inadequate blood-pressure control or an excessive seizure rate—and provide rescue treatment. Monitoring intervals must be short enough to identify poor control promptly. A limited placebo period can also precede a longer active-controlled phase, restricting exposure while establishing short-term treatment effects. Active-comparator designs avoid placebo allocation where they can answer the research question adequately; these choices affect both ethical acceptability and the conclusions the trial can support. (fda.gov)

Consent to masking in research is distinct from undisclosed placebo prescribing in ordinary clinical care. The American Medical Association’s Code of Medical Ethics, Opinion 2.1.4, states that administering placebo without the patient’s knowledge may undermine trust, compromise the physician–patient relationship, and cause harm. It permits clinical placebo use only with the patient’s cooperation and general consent, although the precise timing of administration need not be disclosed. It also rejects giving placebo merely to placate a difficult patient, because that substitutes the physician’s convenience for the patient’s welfare. These provisions concern authorization within a clinical relationship, rather than permission to conceal placebo use whenever contextual symptom improvement is possible. (code-medical-ethics.ama-assn.org)

References

1. E10 Choice of Control Group and Related Issues in Clinical Trialsfda.gov
2. Questions and Answers: The NIH Trials of EDTA Chelation Therapy for Coronary Heart Diseasenccih.nih.gov
3. A Model of Placebo Response in Antidepressant Clinical Trialspmc.ncbi.nlm.nih.gov
4. Placebo interventions for all clinical conditionspmc.ncbi.nlm.nih.gov
5. Placebo Effectnccih.nih.gov
6. How Instructions, Learning, and Expectations Shape Pain and Neurobiological Responsespmc.ncbi.nlm.nih.gov
7. Placebo-induced changes in FMRI in the anticipation and experience of painpubmed.ncbi.nlm.nih.gov
8. The Powerful Placebodcscience.net
9. The powerful placebo effect: fact or fiction?pubmed.ncbi.nlm.nih.gov
10. What are the main methodological problems in the estimation of placebo effects?pubmed.ncbi.nlm.nih.gov
11. E10 Choice of Control Group and Related Issues in Clinical Trialsfda.gov
12. Is the placebo powerless? An analysis of clinical trials comparing placebo with no treatmentpubmed.ncbi.nlm.nih.gov
13. Expectation and dopamine release: mechanism of the placebo effect in Parkinson's diseasepubmed.ncbi.nlm.nih.gov
14. Guideline for Submitting Supporting Documentation in Drug Applications for the Manufacture of Drug Productsfda.gov
15. Side effects and placebo amplificationpubmed.ncbi.nlm.nih.gov
16. Introducing a placebo needle into acupuncture researchpubmed.ncbi.nlm.nih.gov
17. E9 Statistical Principles for Clinical Trialshhs.gov
18. Non-Inferiority Clinical Trials to Establish Effectivenessfda.gov
19. Empagliflozin as Add-on Therapy in Patients With Type 2 Diabetes Inadequately Controlled With Linagliptin and Metformin: A 24-Week Randomized, Double-Blind, Parallel-Group Trialpubmed.ncbi.nlm.nih.gov
20. Ozempic, INN-semaglutide: Public Assessment Reportema.europa.eu
21. Spontaneous improvement in randomised clinical trials: meta-analysis of three-armed trials comparing no treatment, placebo and active interventionpmc.ncbi.nlm.nih.gov
22. Chapter 15: Interpreting results and drawing conclusionscochrane.org
23. Placebo-Induced Changes in fMRI in the Anticipation and Experience of Paincolumbia.edu
24. Neurobiological Mechanisms of the Placebo Effectpmc.ncbi.nlm.nih.gov
25. Conscious Expectation and Unconscious Conditioning in Analgesic, Motor, and Hormonal Placebo/Nocebo Responsespmc.ncbi.nlm.nih.gov
26. Components of placebo effect: randomised controlled trial in patients with irritable bowel syndromepubmed.ncbi.nlm.nih.gov
27. Components of placebo effect: randomised controlled trial in patients with irritable bowel syndromepmc.ncbi.nlm.nih.gov
28. The mechanism of placebo analgesiapubmed.ncbi.nlm.nih.gov
29. Neuropharmacological dissection of placebo analgesia: expectation-activated opioid systems versus conditioning-activated specific subsystemspubmed.ncbi.nlm.nih.gov
30. Placebo interventions for all clinical conditions — Cochranecochrane.org
31. Active albuterol or placebo, sham acupuncture, or no intervention in asthma — PubMedpubmed.ncbi.nlm.nih.gov
32. Active Albuterol or Placebo, Sham Acupuncture, or No Intervention in Asthma — PMCpmc.ncbi.nlm.nih.gov
33. Experimental Assessment of Nocebo Effects and Nocebo Side Effects: Definitions, Study Design, and Implications for Psychiatry and Beyondpmc.ncbi.nlm.nih.gov
34. The biochemical and neuroendocrine bases of the hyperalgesic nocebo effectpubmed.ncbi.nlm.nih.gov
35. ‘Inactive’ ingredients in oral medicationspmc.ncbi.nlm.nih.gov
36. Placebos without Deception: A Randomized Controlled Trial in Irritable Bowel Syndromejournals.plos.org
37. Placebos without Deception: A Randomized Controlled Trial in Irritable Bowel Syndrome - PMCpmc.ncbi.nlm.nih.gov
38. WMA Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participantswma.net
39. Ethical Use of Placebo Controls in Researchcode-medical-ethics.ama-assn.org
40. Use of Placebo in Clinical Practicecode-medical-ethics.ama-assn.org