Hype or hope? High placebo response in major depression treatment with ketamine and esketamine: a systematic review and meta-analysis

Placebo response contributes substantially to the measured effect of antidepressant treatments and has been estimated in earlier work to account for a large proportion of apparent benefit in randomised controlled trials. Previous meta-analyses of oral antidepressants suggested that only a minority of participants derive benefit beyond placebo, and commentators have called for novel trial designs that better control expectation and blinding. Ketamine and esketamine, NMDA-receptor antagonists that can produce rapid reductions in depressive symptoms, have emerged as promising treatments for major depressive disorder (MDD), but the magnitude of placebo response in trials of these agents has not been systematically quantified. Matsingos and colleagues set out to estimate the placebo response in double-blind, randomized, placebo-controlled trials of subanaesthetic ketamine or intranasal esketamine for MDD. The study aims to quantify pooled effect sizes for change from baseline in placebo and active arms, to compare those effects across time points and interventions, and to examine potential moderators and the role of adverse events as indicators of insufficient blinding. This meta-analysis therefore addresses a gap in the literature on psychoactive, non-oral antidepressants and assesses how much of reported treatment response may be attributable to placebo mechanisms rather than drug-specific effects.

The investigators conducted a systematic review and meta-analysis following PRISMA guidelines and registered the protocol on PROSPERO (CRD42022377591). Eligible trials were double-blind, randomized, placebo-controlled studies comparing an inert placebo (for example saline) with subanaesthetic doses of ketamine or esketamine in patients with MDD. Studies with crossover designs, active placebos (such as midazolam), no placebo arm, non-randomized designs or samples where depression was secondary to another psychiatric condition were excluded. Comprehensive database searches were run in Embase, MEDLINE (PubMed), PsycINFO and CENTRAL, with a final systematic search on 3 August 2022 and an additional manual search for 2023 publications on 17 March 2023. Two independent reviewers screened records and extracted data; discrepancies were resolved by a third reviewer. Data extraction followed Cochrane Handbook guidance, missing numerical data were sought from authors or extracted from figures (WebPlotDigitizer), and missing standard deviations were calculated from other reported statistics or imputed from similar studies. Risk of bias was assessed independently by two reviewers using the Cochrane RoB 2 tool. For synthesis the authors calculated within-group change scores (baseline to post-intervention) for placebo and medication arms, expressing effects as Cohen's d with 95% confidence intervals. A restricted maximum likelihood (REML) random-effects model was used to pool effect sizes; heterogeneity was quantified with I². A script using the Pandas library automated effect-size calculations. Pre-specified subgroup analyses examined multiple post-intervention time points (40 minutes, 2, 4, 24 hours and 7 days) and separate analyses for ketamine versus esketamine. Univariate meta-regression explored potential moderators (sample size, age, sex, treatment frequency, route, funding source, country, year, baseline severity, dosage and number of applications) with p < .05 considered significant. To probe blinding, adverse events (AEs) were categorised as medication-specific or unspecific, and a subset of psychomimetic AEs (dissociation, hallucinations) was analysed; ratios of specific or psychomimetic AEs to total AEs were compared between groups using unpaired t-tests. Sensitivity analyses removed statistical outliers to assess robustness of pooled estimates.

The review screened 3,887 abstracts, assessed 452 reports and included 14 trials with a combined sample of 1,100 participants (593 assigned to medication, 507 to placebo). The pooled sample had a mean age of 40.09 years (SD ±12.75) and 57.0% were female. Risk-of-bias assessment rated 11 studies as high risk and 3 studies as having some concerns. Using data from time points spanning approximately 4 to 72 hours post-intervention, the pooled within-group effect size for placebo was dpl = -1.85 (z = -3.42; p < .001; 95% CI -2.90 to -0.79) with substantial heterogeneity (I² = 86.22%), while the pooled treatment (medication) within-group effect size was dtr = -2.57 (z = -6.36; p < .001; 95% CI -3.36 to -1.78; I² = 58.51%). On this basis the authors report that approximately 71.43% of the observed treatment response is replicated in the placebo group. No significant interaction between pooled effect size and the different included post-intervention time points was found. A subgroup synthesis of eight studies with data at seven days post-intervention produced a pooled placebo effect dpl7d = -1.98 (z = -3.02; p = .003; 95% CI -3.26 to -0.69; I² = 86.07%) and pooled treatment effect dtr7d = -3.01 (z = -4.65; p < .001; 95% CI -4.28 to -1.74; I² = 83.13%), with the placebo accounting for about 65.78% of the treatment effect at that time point. Separate subgroup analyses indicated that the placebo accounted for 78% of the treatment response in ketamine studies and 64% in esketamine studies, though heterogeneity patterns differed between those subsets. Sensitivity analyses that identified and removed outliers reduced pooled estimates and heterogeneity: after sensitivity adjustment the overall placebo response accounted for 61.68% of the overall treatment response (I² reported as 54.83% in a supplementary figure). For the seven-day subgroup the sensitivity analysis yielded a pooled placebo effect d = -1.08 (z = -2.90; p = .004; 95% CI -1.80 to -0.35; I² = 0%) and a treatment effect d = -2.48 (z = -3.08; p < .001; 95% CI -4.06 to -0.90; I² = 81.41%), with the placebo explaining 43.54% of the treatment response after outlier removal. Meta-regressions testing candidate moderators (sample size, age, sex, treatment frequency, route, funding, country, year, baseline severity, dosage and number of applications) did not identify any statistically significant predictors of the placebo effect. Adverse-event reporting was available in 9 of 14 studies; across included reports 1,245 AEs were listed (286 in placebo arms, 959 in medication arms). Medication-specific AEs occurred significantly more often in medication groups (t(14) = -2.67; p = .009). Psychomimetic AEs were significantly more frequent in medication arms (t(21) = -5.95; p < .001), a pattern the authors interpret as evidence that distinct drug effects may have compromised participant blinding.

Matsingos and colleagues interpret their findings as indicating a large placebo response in double-blind RCTs of ketamine and esketamine for MDD. The pooled estimates imply that roughly 62–72% of the observed within-group treatment response in the included trials can be replicated by placebo change scores, with sensitivity analyses yielding somewhat lower proportions but preserving a substantial placebo contribution. The authors note that these results are in line with prior meta-analyses of oral antidepressants and emphasise the novelty of quantifying placebo response for fast-acting NMDA-antagonist interventions and non-oral routes. Heterogeneity across studies and the overall risk-of-bias profile tempered interpretation. The review highlighted substantial residual heterogeneity in many pooled estimates and identified 11 of 14 trials as high risk of bias, which the authors argue may lead to underestimation of the true placebo response. Insufficient participant blinding was flagged as a principal concern: psychomimetic adverse events occurred significantly more often in medication arms and could have unblinded participants and assessors, inflating perceived drug benefit. Differences in intervention schedules (number and frequency of applications) and small, varying sample sizes were cited as additional contributors to heterogeneity. To address these challenges the authors recommend optimising trial designs for psychoactive compounds. Proposed measures include using 2x2 factorial designs that manipulate both drug and induced expectation, systematic efforts to neutralise or control participant expectations, routine measurement of de-blinding and expectancy, independent raters for efficacy and safety, and consideration of active placebos. The investigators acknowledge limitations of their meta-analysis: heterogeneity arising from differing protocols and the pooling of ketamine with esketamine, the overall high risk of bias among included trials, and allowance of concomitant antidepressant medication in several studies. They suggest that future, more rigorous trials and separate meta-analyses for each compound could clarify pharmacological effects beyond placebo. The authors conclude that the placebo response plays a significant role in trials of ketamine and esketamine for MDD and that improved study designs are urgently needed to obtain more reliable estimates of drug-specific benefit.