Since the late 19th century, and especially since the post-World War II era, placebos have been instrumental in advancing evidence-based medicine, notably through their role as controls in randomised controlled trials (RCTs). Despite their prevalent use in clinical research and practice, the depth of scientific inquiry into placebos has often remained superficial, failing to extend much beyond their role as comparative controls or inert therapeutic interventions. Reference Enck, Klosterhalfen, Weimer, Horing and Zipfel1,Reference Kaptchuk2 RCTs, a cornerstone of modern biomedical research, are used to establish clinical efficacy and our interpretation of them often focuses only on the active interventions and not on the outcomes in placebo controls. This focus tends to overshadow the importance of thoroughly investigating placebo treatment dynamics, which is rarely part of academic critical appraisal. Reference Correll, Solmi, Cortese, Fava, Højlund and Kraemer3 The knowledge gap around the nature, correlates and mechanisms of placebo responses is particularly impactful in psychiatry, where diagnosis and treatment heavily rely on patients’ subjective experiences. Reference Enck, Klosterhalfen, Weimer, Horing and Zipfel1,Reference Colagiuri, Schenk, Kessler, Dorsey and Colloca4
Placebo responses across psychiatric conditions
Depressive and anxiety disorders, along with obsessive–compulsive disorder (OCD), and trauma-related disorders, represent the bulk of mental health conditions, affecting over 600 million of the 970 million people suffering from mental illnesses, accounting for approximately 61% of disability-adjusted life years (DALYs) caused by mental disorders worldwide. Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor5 Considering the shared and distinct psychopathological features of these conditions, and their significant comorbidity, understanding the characteristics of placebo treatment responses in these conditions, individually and collectively, would eventually have considerable health benefits for people suffering from them.
Differential placebo responses across psychiatric conditions
Meta-analyses have revealed a substantial placebo response in depressive and anxiety disorders, Reference Kirsch6 with a less pronounced response observed in OCD, Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor7 schizophrenia Reference McCutcheon, Pillinger, Mizuno, Montgomery, Pandian and Vano8 and major neurocognitive disorders. Reference Fernández‐López, Riquelme‐Gallego, Bueno‐Cavanillas and Khan9 Two recent publications in 2024, a meta-analysis and an umbrella systematic review, demonstrated differential placebo responses across psychiatric conditions, ranging from a large effect size in depressive and anxiety disorders to small-to-medium effects in OCD and schizophrenia. Reference Huneke, Amin, Baldwin, Bellato, Brandt and Chamberlain10,Reference Bschor, Nagel, Unger, Schwarzer and Baethge11 Alongside significant within-group heterogeneity, studies have noted several potential factors influencing placebo responses in mental health conditions.
However, comprehensive analyses exploring the extent and nuances of placebo treatment responses across a broad spectrum of psychiatric conditions remain sparse, highlighting a crucial area for further investigation. Reference Cao, Liu, Selvitella, Librenza-Garcia, Passos and Sawalha13 The influence of demographic and epidemiological factors on placebo responses, including study setting, participant characteristics and outcome assessment methodology, is still poorly understood. Reference Ahmadzad-Asl, Davoudi, Mohamadi, Hadi, Nejadghaderi and Mirbehbahani14 The variability in placebo responses based on different outcome measures and the psychometric characteristics of assessment tools underscores the complexity of the placebo phenomenon and its interaction with psychopathological changes across mental health conditions. Reference Enck, Klosterhalfen, Weimer, Horing and Zipfel1
The present study
This systematic review and meta-analysis seeks to enhance our understanding of factors impacting placebo responses across mental health disorders, offering insights for future research design and clinical guideline development. To our knowledge, no previous studies have controlled for the impact of intervention type, for example, different medications on the placebo effect estimates. For example, the recent meta-analysis on placebo treatment responses across psychiatric disorders included the highest-quality RCTs for each disorder, Reference Bschor, Nagel, Unger, Schwarzer and Baethge11 inevitably encompassing various intervention arms. Different interventions may have different associated expectations and thus are a notable limitation of previous work.
In the present meta-analysis, we use a unique approach to control for the potential impact of intervention type on placebo responses, and only include RCTs that used two commonly used first-line medications, citalopram and escitalopram, across various non-psychotic psychiatric conditions. This approach allows for comparability across conditions with a common treatment and thus allows us to reduce the impact of potential factors that may influence placebo response associated with different therapeutic interventions. We also utilise all available outcome data reported in RCTs to investigate the differential placebo responses on multiple domains of psychiatric conditions measured by various assessment tools. These features allow for a more contextual analysis of potential correlates of the placebo response.
It is important to note that the terminology in the placebo research literature can be complex, with the terms ‘placebo effects’ and ‘placebo response’ often used interchangeably in some publications. Recognising this complexity, we define ‘placebo effects’ as the presumed specific effects attributable to placebos, whereas ‘placebo (treatment) response’ refers to the total placebo effect size observed in the placebo arm of trials, which also includes various non-specific factors such as regression to the mean, spontaneous remission, natural course of the illness and the Hawthorne effect. Reference Enck, Klosterhalfen, Weimer, Horing and Zipfel1
Method
This systematic review, registered in PROSPERO (#CRD42022371957), adheres to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Reference Kaptchuk2 To facilitate the management of citations and the systematic review process, we employed Covidence (Veritas Health Innovation, Melbourne, Australia; https://www.covidence.org/) and EndNote 20.0 (Clarivate, Philadelphia, PA, USA; https://endnote.com/downloads/?gad_source=1&gad_campaignid=20987077387&gbraid=0AAAAACv9_uPXYCmGOpRfneD6hYd2ejsxt&gclid=Cj0KCQjw0JnRBhDJARIsALobnXZEBIQb3oY0qWTXjGpaJw5m0FlevfAmFBosBuDN9ENd5zXlWHR8PcMaAs3nEALw_wcB).
Search strategy and inclusion criteria
We (R.J.) conducted a comprehensive search across major databases (PubMed, Embase, Scopus, Web of Science, Ovid, PsycINFO and the Cochrane Library) from their inception until 31 August 2023, using predefined queries (see Supplementary Table 1). The aim was to identify randomised controlled trials (RCTs) comparing at least one placebo arm to a medication arm involving two common selective serotonin reuptake inhibitor (SSRI) medications (citalopram or escitalopram). Reference Correll, Solmi, Cortese, Fava, Højlund and Kraemer3 These medications, as highlighted by the STAR*D study, have been among the most commonly prescribed psychotropic drugs over the past two decades, often being the initial medication choice. Reference Colagiuri, Schenk, Kessler, Dorsey and Colloca4 Target populations were patients with major psychiatric conditions: depressive disorders, anxiety disorders, OCD and trauma and stress-related disorders, with diagnoses based on criteria from DSM III, III-R, IV, IV-TR, 5, and 5-TR and other specified criteria, (Supplementary Table 2).
To be included, studies had to be RCTs in English, including at least one placebo arm, and reporting continuous outcomes baseline and final statistics (or change from baseline) using a clinical symptom scale. Studies lacking a placebo group or without accessible outcome data for this group were excluded.
Study selection involved an initial screening of titles and abstracts, followed by a detailed review of full texts, both steps conducted independently by two reviewers (M.A., R.J., R.M., S.M., D.S., S.S.). Conflicts were mediated by a third reviewer (M.A.-A.). To manage missing data, we sought additional information from study authors or supplementary materials, resorting to statistical imputation as necessary. For example, to impute missing statistics on final outcome values, we calculated the final mean from the pre–post value change and used baseline s.d. as an estimate of final s.d. assuming equal variances between baseline and final measurements. Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor5 Studies irreconcilable through these methods were excluded from meta-analysis.
The primary outcome was the change in symptom severity from baseline to post-treatment, assessed using the principal validated clinical scale specific to each psychiatric condition used in each study. Additional clinician-rated and patient-reported scales, functional measures and other symptom domains were pre-specified as secondary outcomes. When multiple eligible clinical outcomes were reported within a study, all outcomes meeting inclusion criteria were extracted for analysis.
Data extraction
Since 15 January 2023, two reviewers (M.A., R.J., R.M., S.M., D.S., S.S.) independently extracted data, which included study identifiers, attributes, target condition, participant demographics and continuous clinical outcome measurements. For trials with multiple intervention arms or varying dosages, the highest dosage of citalopram/escitalopram was selected for analysis. To assess the quality of the studies included in our review, we employed version 2 of the Cochrane tool for assessing risk of bias in randomised trials (RoB 2). Reference Kirsch6
Statistical analysis
For both placebo and medication groups (pre versus post treatment), the primary outcomes of interest were calculated using a modified Cohen’s d effect size, Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor7 accounting for potential correlation between baseline and final values. Reference McCutcheon, Pillinger, Mizuno, Montgomery, Pandian and Vano8 A random-effects model was employed due to significant heterogeneity between studies:
$$Cohen{\unicode{x2019}}s\;d\;ES = {{\left( {{{\overline X}_2} - {{\overline X}_1}} \right)} \over {\sqrt {{s.d.}_1^2 + s.d._2^2 - 2{\rm{*}}r{\rm{*}}{{s.d.}_1}{\rm{*}}{{s.d.}_2}} }}{\rm{*}}\sqrt {2(1 - r)}$$
where s.d. and
$\overline{X}$
are baseline and final s.d. and mean values, respectively, and r represents the correlation between baseline and final scores, which we imputed as 0.75.
Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor5
We explored factors influencing placebo response magnitude, including study design (target psychiatric condition, publication year, number of study arms, setting, sample size, attrition rate, follow up duration), participant characteristics (age group, mean age, proportion of male participants, baseline severity scores) and outcome measurement features, such as measurement tool characteristics (clinician- or patient-rated), outcome domainFootnote a (depressive, anxiety, OCD, trauma-related symptom scales and general mental health scales,Footnote b frequency of measurement, analysis type (intention-to-treat; ITT, or observed cases/per protocol)).
When studies reported multiple outcome measures assessing the same clinical domain, individual effect sizes were calculated for each instrument and combined within the study to obtain a single summary effect size prior to meta-analysis. This within-study pooling ensured that each study contributed only one independent effect size per domain, thereby avoiding double counting of participants and statistical dependency among correlated outcomes.
Differences in pooled effect sizes across study and outcome characteristics (e.g. clinician- versus patient-rated outcomes, diagnostic categories, age groups and analysis type (ITT versus per-protocol)) were examined using random-effects meta-regression and subgroup analyses. Categorical moderators were entered as indicator variables, and statistical significance was assessed using Wald tests of model coefficients. These analyses were conducted to evaluate whether effect sizes differed significantly between subgroups while accounting for between-study heterogeneity.
Sensitivity analysis
Sensitivity analyses were conducted to assess the robustness of the findings. Analyses were repeated after excluding studies requiring imputed data. Outcome measures were additionally rescaled to a standardised 0–100 metric (x Adj = x − min/max − min and s.d. Adj = s.d./max − min) to evaluate whether differences in scale ranges influenced the results, and strategies such as pooling-related subscale outcomes were explored to reduce heterogeneity. We also applied an alternative method for imputing any missing s.d. based on the observed correlation between baseline and post-treatment variance among studies reporting both values, to assess the impact of imputation assumptions on effect-size estimates. In addition, because calculation of standardised mean change required specification of the correlation between baseline and post-treatment scores (r), sensitivity analyses were repeated using alternative assumed values (r = 0.5 and r = 0.9) to evaluate the stability of the results.
All meta-analyses were performed using Stata 16.0 SE (StataCorp LLC, College Station, TX, USA; https://download.stata.com/download/), utilising its ‘meta’ and ‘metareg’ functions.
Results
Study characteristics
Our screening process evaluated 5443 papers, leading to a full-text review of 440 studies. Despite contacting authors directly when articles could not be obtained, full texts for seven papers remained inaccessible (Fig. 1).
Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram of included studies. RCT, randomised controlled trial.
Fig. 1 Long description
A flowchart illustrating the process of selecting studies for a meta-analysis on placebo treatments in clinical trials. The flowchart is divided into three main sections: Identification, Screening, and Included. In the Identification section, references are gathered from databases/ registers (10990) and other sources (12), resulting in a total of 11002 references. After removing 5567 duplicate references, 5443 studies are screened. In the Screening section, 5007 studies are excluded, leaving 440 studies sought for retrieval. Of these, 7 studies are not retrieved, and 433 studies are assessed for eligibility. 353 studies are excluded for various reasons, such as being registered as clinical trials, lacking a placebo group, or not reporting target data. Finally, 80 studies are included in the meta-analysis, reporting 312 clinical outcome measurements.
From the review, 80 studies met our inclusion criteria, examining treatments across the specified conditions using at least 1 clinical outcome (in total 312 clinical outcomes, mean ± s.d. 3.9 ± 2.9 outcomes per study; see Fig. 1). Collectively, these studies enrolled 19 776 participants (6427 of whom were in placebo groups) with an average study completion rate of 85.2%. Demographically, the studies reported an average male participation rate of 42.3% (43.0% in placebo groups), and covered an age range from 7 to 91 years (mean age for all participants was 46.0 ± 16.8 and for placebo participants was 46.2 ± 17.1, Fig. 2 and Supplementary Table 2).
Summary of included studies and their characteristics, with placebo and medication effects estimates. Random-effects meta-analysis of within-group placebo effect sizes (paired Cohen’s d). Each square represents a study-specific estimate with its 95% CI. References for the studies listed in the figure are provided in the Supplementary Material. Multi, multi-centre study; Single, single centre study; Medication, active medication group of the study; Cit, citalopram; Escit, escitalopram; N-Pl, placebo group sample size; Male (%), in placebo group; Age group, study’s target population; <18 years old, child and adolescent; 18–64 years old, adults; ≥65 years old, elderly; F/U(w), duration of follow up in weeks; #Measurements, number of outcome measurements during the study; Study condition, the study target population; Dis., disorders; PTSD, post-traumatic stress disorder; OCD, obsessive–compulsive disorder; placebo ES, Cohen’s d effects size in placebo group; Medication ES, Cohen’s d effect size in medication group; Pl-Med-Dif, difference between the medication and placebo effect sizes; QA, quality assessment: study quality for risk of bias, based on the Cochrane tool for assessing risk of bias in randomised trials (RoB 2); LR, low risk; SC, some concerns; HR, high risk; I2, Higgin’s heterogeneity; ITT, intention to treat; PP, per protocol or observed cases analysis; OP, out-patient; IP, in-patient.
Placebo response across different psychiatric conditions
The overall placebo effect size across all psychiatric conditions, based on all outcomes data, was 0.97 (95% CI: 0.89–1.05), with substantial heterogeneity found across all measurements (I 2: 99.5%). The medication effect size, based on all outcome measurements across the psychiatric conditions, was 1.44 (95% CI: 1.33–1.54, I 2: 99.8%; see Fig. 2). A significant positive correlation was observed between placebo and medication effect size (r = 0.81, P < 0.001), with the placebo effect size being approximately 67.4% of the medication effect size (Fig. 3 and Supplementary Fig. 1).
Placebo and medication effects across major psychiatric conditions. #Number of outcome measurements in each group; ES, effect size; Dis., disorders; AD, anxiety disorders; DD, depressive disorders; OCD, obsessive–compulsive disorder; TS, trauma and stress-related disorders; GAD, generalised anxiety disorder; SAD, social anxiety disorder; MDD, major depressive disorder; PTSD, post-traumatic stress disorder.
Fig. 3 Long description
A table comparing the effects of placebos and medications across various psychiatric conditions. The table is divided into several sections: Target condition class, Target condition name, Medication intervention, and Study setting. Each section contains rows with specific conditions or interventions and columns for the number of measurements, placebo effect size with 95% confidence interval, and P-value, as well as medication effect size with 95% confidence interval and P-value. The table includes data for anxiety disorders, depressive disorders, OCD, trauma and stress-related disorders, and specific conditions like GAD, panic disorder, SAD, specific phobia, MDD, other depressive disorders, OCD, and PTSD. It also compares different medications like citalopram and escitalopram, and study settings like multicentre and single-centre. Notable trends include varying effect sizes and P-values across different conditions and interventions, with most P-values being less than 0.001, indicating significant results. The overall heterogeneity and test of group differences are also provided.
Significant differences in placebo response were noted across clinical conditions, independent of the medication type. PTSD and depressive disorders exhibited the highest placebo effect size (1.14 and 1.02, respectively), whereas OCD showed the lowest (effect size: 0.62, P < 0.01). Anxiety disorders displayed a moderately large pooled placebo effect size (0.86), with significant variability among anxiety disorder subclasses, specifically noting the lowest placebo effect size in specific phobia (0.23, P < 0.01). The medication effect size across major psychiatric conditions was not significantly different (P = 0.61, Fig. 2).
There were 61 different clinical outcome measurement tools across all the RCTs included, which were further categorised into 5 main domains: anxiety scales (16 tools), depression scales, Reference Trivedi, Rush, Wisniewski, Nierenberg, Warden and Ritz18 general mental health scales, Reference Cohen22 PTSD scales Reference Correll, Solmi, Cortese, Fava, Højlund and Kraemer3 and OCD scales. Reference Kaptchuk2
The number of clinical outcome tools utilised ranged from 1 to 14 per study, and there was substantial heterogeneity within studies using multiple outcomes (I 2 ranging from 24.2 to 99.8%). Out of 61 tools, 27 were used in only one study, whereas the remaining 34 were used between 2 and 55 times, with the Hamilton Rating Scale for Depression (HRSD) being the most prevalent tool with associated placebo and medication effect size of 1.17 and 1.64, respectively (Fig. 4 and Supplementary Table 3).
Different outcome measure tools used in RCTs included in the systematic review of placebo responses across major psychiatric conditions. Numbers refer to the estimated placebo effect size based on the corresponding scale outcomes across major psychiatric conditions. The size of the pie slice indicates the number of scales assessing outcomes in that domain. OCD, obsessive–compulsive disorder; PTSD, post-traumatic stress disorder; ASI, Anxiety Sensitivity Index; BAI, Beck Anxiety Inventory; Covi-Anx, Covi Anxiety Scale; FQ, Fear Questionnaire; GAD-7, generalised anxiety disorder-7; HADS, Hospital Anxiety and Depression Scale; HRSA, Hamilton Rating Scale for Anxiety; LSAS, Liebowitz Social Anxiety Scale; MPS, Multidimensional Perfectionism Scale; PA-freq., panic attack frequency; PARS, Pediatric Anxiety Rating Scale; PAS, Panic and Agoraphobia Scale; PSWQ, Penn State Worry Questionnaire; RRS,Ruminative Responses Scale; SCL-90, Symptom Checklist-90; STAI, State–Trait Anxiety Inventory; AES, Apathy Evaluation Scale; BDI, Beck Depression Inventory; BDI-II, Beck Depression Inventory-II; CDRS-R, Children’s Depression Rating Scale-Revised; CSDD, Cornell Scale for Depression in Dementia; FSS: Fatigue Severity Scale; GDS, Geriatric Depression Scale; HRSD, Hamilton Rating Scale for Depression; IDS-C, Inventory of Depressive Symptomatology-Clinician Rated; IDS-SR, Inventory of Depressive Symptomatology-Self Report; ISI, Insomnia Severity Index; MADRS, Montgomery-Åsberg Depression Rating Scale; MES, Melancholia Scale; mOAS, Modified Overt Aggression Scale; PHQ-9, Patient Health Questionnaire-9; QIDS-SR, Quick Inventory of Depressive Symptomatology-Self Report; RDS, Retardation Rating Scale; SHAPS, Snaith-Hamilton Pleasure Scale; AMD-SEQ, Anxiety and Mood Disorders Self-Efficacy Questionnaire; CBI, Caregiver Burden Inventory; CES, Clinical Efficacy Scale; CGA-S, Comprehensive Geriatric Assessment-Short form; CGI-I, Clinical Global Impressions-Improvement; CGI-PA, Clinical Global Impressions-Patient Assessment; CGI-S, Clinical Global Impressions-Severity; GAF, Global Assessment of Functioning; LLFD, Lawton’s Instrumental Activities of Daily Living Scale; PGE, Patient Global Evaluation; PGI-I, Patient Global Impression of Improvement; PSQI, Pittsburgh Sleep Quality Index; PSSS-R, Perceived Social Support Scale-Revised; QOL, Quality Of Life inventories; RMBPC, Revised Memory and Behavior Problems Checklist; SCL-56, Symptom Checklist-56; SDS, Sheehan Disability Scale; SF-36, 36-Item Short Form Survey; SOFAS, Social and Occupational Functioning Assessment Scale; TDSS, treatment discontinuation signs and symptoms; WHODAS, World Health Organization Disability Assessment Schedule; WHOQOL, World Health Organization Quality of Life; NIMH-OCS, National Institute of Mental Health Obsessive–Compulsive Scale; Y-BOCS, Yale-Brown Obsessive–Compulsive Scale; CAPS, Clinician-Administered PTSD Scale; DTS, Davidson Trauma Scale; IES, Impact of Event Scale.
Larger sample sizes and studies with more than four arms reported significantly higher placebo effect size. Additionally, clinician-rated outcomes (as opposed to self-reported measures) (placebo effect size 1.07 and 0.76, respectively) and the use of ITT data analysis strategies (placebo effect size: 1.08 v. 0.85 for per protocol analysis) were each associated with elevated placebo effect size (all P < 0.05). In the medication arm, clinician-derived effect size scores were notably higher (effect size: 1.59 v. self-report effect size: 1.11, P < 0.01), although the method of effect analysis showed no significant impact on medication effect size (P = 0.08, Fig. 3).
Children and adolescents recorded the highest effect size (1.77 for placebo and 2.37 for medication, P < 0.001), whereas the lowest effect size was observed in studies involving older adults and mixed-age adult populations (0.65 and 0.76 for placebo, respectively, P < 0.001). Meta-regression also demonstrated a positive correlation with the number of study arms (b = 0.182, s.e. = 0.043, P < 0.01), multicentre studies (P < 0.01) and a negative correlation with mean age of study participants (b = −0.010, s.e. = 0.003, P < 0.01) Figs 5 and 6.
Participants and outcome measurement factors’ association with the placebo and medication effect estimates in major psychiatric conditions. ES, effect size; Child&Adol, children and adolescents; OCD, obsessive–compulsive disorder; PTSD, post-traumatic stress disorder. Target Specific outcome refers to outcome measurements directly related to the target condition, such depressive scales in depressive disorders, other scales, such as functional status and anxiety symptom scales in randomised controlled trials on depressive disorders, if measured, were coded as non-target specific outcomes. ITT, intention-to-treat analysis; PP, per-protocol or observed cases analysis.
Fig. 5 Long description
A table comparing the effect sizes of placebo and medication across different groups and conditions in mental health disorders. The table has 12 rows and 6 columns. The columns are labeled Groups, No. of measurements, Placebo ES with 95% CI, P-value, Medication ES with 95% CI, and P-value. The rows are grouped by Age group, Outcome measurement administration, Outcome measurement group, Target condition specific outcome, and Analysis type. Each group contains specific categories with corresponding data. The table includes effect sizes (ES) with 95% confidence intervals (CI) and P-values for both placebo and medication across various conditions and measurements. Notable trends include variations in effect sizes and P-values across different age groups, outcome measurement administrations, and specific conditions. The table also includes overall heterogeneity and test of group differences for each category.
Meta-regression analysis of potential factors influencing the placebo responses across major psychiatric conditions. ES, effect size; b, regression. Bubble sizes reflect the weight of the measurement in overall placebo effect-size estimates.
Discussion
Our systematic review and meta-analysis showed robust placebo treatment responses for major non-psychotic psychiatric conditions. To our knowledge, this study is the first meta-analysis that studies placebo responses across multiple psychiatric diagnoses while synthesising trials involving the same pharmacological agents – citalopram and escitalopram – thereby reducing variability related to differences in treatment characteristics. This methodology allows a more consistent evaluation of placebo effects across conditions. This analysis, consistent with findings from prior condition-specific meta-analyses, found a pronounced placebo response in anxiety and depressive disorders, contrasting with more limited placebo responses in OCD. Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor7,Reference Huneke, Amin, Baldwin, Bellato, Brandt and Chamberlain10 Notably, our analysis reveals that the placebo response constitutes a significant proportion (approximately 67%) of the overall response in the medication group across conditions, a finding that aligns with prior research. Reference Perivolaris, Ainsworth, Alexopoulos, Bingham, Flint and Marino24,Reference Benedetti, Enck, Frisaldi and Schedlowski25 This suggests that the majority of therapeutic outcomes in RCTs examining efficacy of these SSRIs (and likely in clinical practice) may be attributed to the placebo response, encompassing placebo effects as well as other non-specific effects such as the natural illness progression, regression to the mean and observer effects. Reference Rodrigues26
Our findings also reinforce the concept that placebo responses are not uniform within or across psychiatric conditions. Instead, they vary significantly Reference Huneke, Amin, Baldwin, Bellato, Brandt and Chamberlain10,Reference Bschor, Nagel, Unger, Schwarzer and Baethge11,Reference Parker, Ricciardi and Hadzi-Pavlovic27 which may be the product of a complex interplay between the psychopathological underpinnings of each condition, the subjective nature of the illness experience and the methodologies employed for outcome assessment – be it observer-rated or self-reported. Reference Parker, Ricciardi and Hadzi-Pavlovic27,Reference Khan and Brown28 In this study, we specifically focused on data from RCTs that employed two commonly used SSRIs (citalopram and escitalopram) to assess both placebo and treatment effects across various psychiatric conditions. This approach was chosen to minimise the potential impact of differing intervention types on the observed placebo responses and to enhance comparability of these effects across conditions. To our knowledge, this methodological emphasis on specific interventions, rather than a broader spectrum of modalities, such as a large group of antidepressants, across different conditions, represents a novel approach in meta-analyses of placebo responses in psychiatry. Typically, such analyses have focused on placebo response in either a single psychiatric disorder or across a category of disorders using various intervention modalities utilising a pool of data from various intervention types. Reference Bschor, Nagel, Unger, Schwarzer and Baethge11,Reference Jones, Razza, Weissman, Karbi, Vine and Mulsant29 Several studies highlighted the potential impact of the intervention type, from novelty of the intervention to route of administration, on the placebo effect, Reference Jones, Razza, Weissman, Karbi, Vine and Mulsant30 which may have precipitated through different pathways such as the level of patients’ and clinicians’ expectations regarding treatment potentials and outcome. Reference Hróbjartsson and Gøtzsche31–Reference Jensen, Turner and Romano33
Multifaceted nature of placebo responses in psychiatry
Our findings also reveal that outcomes assessed by clinicians report higher effect sizes for both placebo and medication compared with self-reported measures. This difference points to the influential role of both patient and clinician perception on the placebo response measures. Reference Hjorth, Frick, Gingnell, Hoppe, Faria and Hultberg34 To our knowledge, the differential placebo responses based on clinician-rated outcomes compared with self-reported scales have not been extensively studied across different psychiatric conditions. For example, most systematic reviews of placebo responses in depressive disorders have utilised the clinician-administered Hamilton Rating Scale for Depression (HRSD) and generally did not include self-rated outcomes. Reference Østergaard35 Previous meta-analyses conducted by our team have shown that higher placebo responses in panic disorder and OCD are associated with estimates based on clinician-administered measurements. Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor7,Reference Ahmadzad-Asl, Davoudi, Mohamadi, Hadi, Nejadghaderi and Mirbehbahani14 Natural history and spontaneous remission likely contribute substantially to placebo response. A meta-analysis of wait-list controls in major depressive disorder reported moderate spontaneous improvement (effect size ≈ 0.5), suggesting that part of the observed placebo response reflects the natural course of illness rather than treatment-specific effects. Reference Rutherford, Mori, Sneed, Pimontel and Roose70
The discrepancies between patient- and clinician-rated outcomes underscore an essential facet of psychiatric research. Reference Yamada, Fujii, Hattori, Hori, Matsumura and Kurashimo36 The patient–clinician interaction can influence the placebo responses via a range of underexplored factors such as interpersonal dynamics, reduced anxiety, effects of clear diagnosis and information, verbal communication cues and suggestibility. Reference Conboy, Macklin, Kelley, Kokkotou, Lembo and Kaptchuk37–Reference Brody and Waters39 Contrary to expectations of lower heterogeneity in clinician-reported outcomes. Reference Fernández‐López, Riquelme‐Gallego, Bueno‐Cavanillas and Khan9,Reference Enns, Larsen and Cox12,Reference Hróbjartsson and Gøtzsche31,Reference Cuijpers, Li, Hofmann and Andersson40 our findings showed a significant difference in placebo-medication variability between assessment types, highlighting the value of patient-reported subjective mental experiences, besides clinician-based assessments, in understanding mental health conditions and emotions.
The subjective experience of mental disorders, such as depression and anxiety, varies greatly across individuals, influenced by a myriad of personal, sociocultural and neurobiological factors. Reference LeDoux and Hofmann41 Our analysis suggests a notable discordance between subjective experiences and objective clinical or neurophysiological assessments. This discordance points further to a broader issue within psychiatric evaluation and treatment, where subjective experiences and objective measurements may not fully align, thus complicating the understanding of therapeutic responses, including placebo responses. Reference Taschereau-Dumouchel, Michel, Lau, Hofmann and LeDoux42
Reflecting on Søren Kierkegaard’s notion that life can only be understood backwards but it must be lived forwards, we recognise the challenge in reconciling the objective clinical assessments with the rich, subjective experiences of individuals with mental health conditions. From this philosophical perspective, we encourage a dual approach to mental health research and treatment, combining objective, clinical and neurobiological investigations with a deep appreciation for the subjective experiences of patients. Such an approach could foster a more holistic understanding of mental health and its treatment, including the nuanced dynamics of placebo responses. Reference Phillips, Beretta, Whitaker, Smith and Whitaker43
The methods of treatment response measurement and reporting
One of the central findings of our meta-analysis was that placebo response magnitude varied substantially depending on outcome measurement methods and study design characteristics. This variance, accentuated by the use of different outcome measurement tools like HRSD and the Montgomery-Asberg Depression Rating Scale (MADRS), points to a broader challenge within psychiatric research. Reference Ma, Yang, Yang, Kang, Wang and Zhang44 These scales, despite their widespread use and validation as markers for clinical depression, introduce considerable variability into the assessment of therapeutic improvement. Reference Khan and Brown28,Reference Østergaard35 Our analysis reflects this, with both HRSD and MADRS demonstrating extensive heterogeneity in placebo response estimates (HRSD and MADRS) I 2: 99.3 and 99.8%, respectively, Supplement Table 3). This variability highlights the need for careful consideration of outcome selection when interpreting and comparing placebo responses across psychiatric trials. In our analysis, clinician-rated outcomes were associated with higher placebo effect sizes than self-reported measures. Previous meta-analyses of antidepressant trials have reported similar observations, with some studies suggesting higher response estimates when clinician-administered scales are used compared with self-reported measures. Reference Mohamadi, Ahmadzad-Asl, Nejadghaderi, Jabbarinejad, Mirbehbahani and Sinyor7,Reference Rief, Nestoriuc, Weiss, Welzel, Barsky and Hofmann66
Moreover, methods of data analysis, such as ITT analyses or the Last Observation Carried Forward (LOCF) technique previously endorsed by the Federal Drug Administration, further inflated placebo response estimates. The ITT approach, which replicates the last known score of a participant who withdraws from a trial, can subdue the differentiation between placebo and active treatment effects, especially considering the potential early onset of response to placebo compared with later responses to antidepressants. Reference Khan and Brown28 Such methodological choices in outcome measurement and analysis can significantly contribute to the observed heterogeneity in placebo responses across psychiatric conditions. Previous studies have suggested that analytical approaches such as ITT/LOCF may influence observed treatment effects, partly because early symptom improvements are retained when outcomes from participants who discontinue trials are carried forward. Reference McMahon, Kelly, Boggs, Li, Hu, Davis and Carpenter67–Reference Rutherford, Sneed and Roose69 This interpretation is consistent with evidence that symptom improvement, including placebo response, is often greatest during earlier phases of treatment.
Beyond measurement and analysis methods, other study design factors can also influence the placebo responses. Previous research suggests that the number of study arms patients are randomly assigned to, along with sample size, can significantly impact the observed placebo response size. Reference Baines, Araklitis, Flint, Robinson and Cardozo45 In our own study, we found increased placebo responses in RCTs using a higher number of study arms and multi-centre designs; however, we did not find a direct independent correlation between sample size and placebo responses. Regression analyses suggest this might be because the influence of the sample size is mediated by the number of study arms and multi-centre design of trials. Studies with more intervention arms, including placebo and other interventions, tend to show larger placebo responses. Although the reasons behind this association are unclear, we can speculate that complex study designs with multiple arms, often conducted across multiple centres, among other factors, might create higher expectations of effectiveness in both patients and clinicians, who expect a higher chance of receiving ‘active intervention’ in trials with multiple intervention arms. Reference Sinyor, Levitt, Cheung, Schaffer, Kiss, Dowlati and Lanctôt46,Reference Salanti, Chaimani, Furukawa, Higgins, Ogawa and Cipriani47
RCTs gather extensive outcome data, yet often only a fraction of these data are published and even less is utilised in systematic reviews and meta-analyses. Reference Mayo-Wilson, Fusco, Li, Hong, Canner and Dickersin15 Our study sought to bridge this gap by incorporating a broader range of outcome measurements into our meta-analysis, aiming to paint a more comprehensive picture of participant responses across psychiatric conditions (mean ± s.d. of outcome measurements per study: 3.93 ± 2.89). Previous meta-analyses have predominantly focused on single, main outcome measurements, potentially overlooking the nuanced effects visible in the wider data-set. Although our approach aimed for inclusivity, it also introduced challenges regarding the reliability and consistency of meta-analytic statistics using different outcome data. These findings suggest that reported placebo magnitude in psychiatric trials may depend substantially on how outcomes are measured, rather than solely reflecting differences in treatment efficacy.
Shared neurobiological pathways of placebo and other treatment responses
Although the present study did not directly examine neurobiological mechanisms, existing evidence suggests that placebo and active treatment responses may involve overlapping neurobiological pathways, including changes in brain networks and neurotransmitter systems related to expectation, reward processing, learning and emotional regulation. Reference Mayberg, Silva, Brannan, Tekell, Mahurin and McGinnis48–Reference Faria, Gingnell, Hoppe, Hjorth, Alaie and Frick52 Similar patterns of neurocircuitry activity, particularly within regions implicated in depressive and anxiety disorders such as components of limbic and default mode networks, have been observed in both placebo and active treatment responses, suggesting partially shared mechanisms that may help explain the positive association between placebo and medication responses observed in our analysis (r = 0.81).
Age-related differences in placebo responsiveness may also reflect variations in neuroplasticity and adaptive learning processes, with younger individuals generally demonstrating greater susceptibility to expectancy-related effects, whereas reduced neuroplasticity and accumulated treatment experiences may attenuate such responses in older populations. Reference Weimer, Gulewitsch, Schlarb, Schwille-Kiuntke, Klosterhalfen and Enck53–Reference Turner, Deyo, Loeser, Von Korff and Fordyce57 Evidence regarding gender differences in placebo responses remains inconsistent, Reference Vambheim and Flaten58,Reference Enck and Klosterhalfen59 and although our meta-regression did not identify a significant association with sex distribution, the interaction between biological, psychological and social factors warrants further investigation. Overall, these interpretations remain indirect, and future studies integrating neurobiological and clinical data are needed to better clarify mechanisms underlying placebo responses across psychiatric conditions.
Limitations and future directions
Several limitations should be considered when interpreting our findings. First, evidence of publication bias suggests that studies reporting smaller or negative treatment effects may be underrepresented in the published literature, which could influence the estimated magnitude of both placebo and medication responses. Second, substantial heterogeneity was observed across studies, likely reflecting differences in study populations, trial designs, outcome measures and follow-up durations; therefore, pooled estimates should be interpreted with appropriate caution despite the use of random-effects models.
In addition, many randomised trials incompletely reported factors that may influence placebo responses, including participants’ socioeconomic background, baseline expectations and funding characteristics, limiting our ability to fully examine these potential moderators. Our analyses also relied on available quantitative outcome measures, and in some cases incomplete reporting required statistical imputation, which may introduce additional uncertainty into the estimates.
The analyses were restricted to continuous clinical outcomes and pill placebo designs, which may limit generalisability to other placebo modalities or outcome types. Furthermore, not all psychiatric conditions were represented, as disorders such as bipolar disorder and schizophrenia were outside the scope of this study. An additional limitation is the unequal distribution of available evidence across diagnostic categories; substantially fewer trials were available for OCD and PTSD compared with anxiety and depressive disorders, resulting in less precise subgroup estimates that should be interpreted cautiously and considered exploratory.
Finally, effect sizes were derived from within-group pre–post changes rather than direct between-group comparisons. Consequently, observed improvements may partly reflect non-specific factors such as regression to the mean, natural symptom fluctuation, or repeated assessment effects, and should not be interpreted as representing purely treatment-specific change. Smaller studies and occasional extreme effect sizes may also have influenced pooled estimates despite the sensitivity analyses performed.
Future research should further examine the interaction between placebo responses, methodological factors and individual patient characteristics using individual-participant data and more standardised outcome reporting. Improved understanding of these mechanisms may help refine trial design and interpretation and contribute to more accurate evaluation of therapeutic effects in psychiatric research.
In summary, this meta-analysis demonstrates that placebo responses represent a substantial and variable component of treatment outcomes, especially across depressive and anxiety disorders, shaped not only by disorder characteristics but also by outcome measurement, analytic strategy and trial design. These findings emphasise that placebo responses in psychiatric research should be understood as reflecting meaningful contextual, psychological and natural history processes rather than methodological noise alone. Greater recognition of these influences may help refine interpretation of randomised trials and support the development of study designs that more accurately distinguish specific treatment effects while preserving clinical relevance.
Supplementary material
The supplementary material is available online at https://doi.org/10.1192/bjp.2026.10665
Author contributions
M.A.-A.: conceptualisation, methodology, supervision, screening, data analysis and writing the original draft. R.J.: project administration, conceptualisation, protocol registration, literature search, screening, data extraction, review and editing. M.A., R.M., S.M., D.S., S.S.: screening, data extraction and review. S.H.A., M.J.B., F.D., T.J.K., K.L.L., M.S.: conceptualisation, review and editing.
Funding
This study received no specific grant from any funding agency, commercial or not-for-profit sectors.
Declaration of interest
K.L.L. is supported by the Bernick Chair in Geriatric Psychopharmacology. M.S. declares salary support from an academic scholar award from the Department of Psychiatry at Sunnybrook Health Sciences Centre. The other authors declare no conflict of interest.
eLetters
No eLetters have been published for this article.