The use of ketamine as an antidepressant: a systematic review and meta-analysis

Most standard antidepressants target monoamine systems and typically require weeks to produce clinical benefit. By contrast, ketamine, an N-methyl-D-aspartate (NMDA) receptor antagonist that modulates glutamate neurotransmission, has been reported to produce rapid antidepressant effects that outlast its plasma half-life. Early clinical reports and placebo-controlled trials suggested rapid symptom reduction following intravenous ketamine infusions, including in treatment-resistant major depressive disorder (MDD) and bipolar disorder (BD), and raised interest in ketamine’s potential to reduce suicidal ideation in acute settings. Coyle and colleagues set out to synthesise the clinical trial evidence on ketamine as an antidepressant. The meta-analysis aimed to determine whether ketamine produces an immediate reduction in depressive symptoms, whether effects are sustained over time, whether repeat infusions provide additional benefit, and whether diagnosis, study design, infusion blinding, or participant sex moderate the antidepressant response. The analysis focused on four post-infusion time points (4 hours, 24 hours, 7 days and 12–14 days) and sought to compare single versus repeated infusion studies and pre-post versus placebo-controlled designs.

Coyle and Laws conducted the review following PRISMA guidelines. A systematic search of Web of Science, ScienceDirect and PubMed was performed using the terms 'ketamine' AND 'depression', covering the period from 2000 to January 2015. The ScienceDirect search was restricted to several subject areas including Pharmacology and Psychology. Studies were eligible if they administered at least one ketamine infusion for MDD or BD, used a standardised depression rating scale (for example MADRS or HDRS), and included eight or more participants. Data were extracted for four post-infusion time points: 4 hours, 24 hours, 7 days and 12–14 days. Where studies had a control arm, effect sizes compared ketamine with placebo; where no control existed, pre-post comparisons (baseline versus post-infusion) were used. The investigators contacted study authors when reported data were incomplete; studies for which additional data could not be obtained were excluded. Meta-analyses were conducted in Comprehensive Meta-Analysis Version 2.0, using Hedges’ g (standardised mean difference corrected for small-sample bias) and a random-effects model. Cohen’s conventions were applied to describe effect size magnitude (0.20 small, 0.50 moderate, 0.80 large). Heterogeneity was quantified using the I2 statistic, with conventional thresholds described in the paper (for example 50–90% indicating substantial heterogeneity). Risk of bias was assessed categorically: low risk if a control group and adequate randomisation were present and no conflict of interest was evident; medium risk for some ambiguity in randomisation or potential conflicts and designs where blindness could be compromised; high risk for uncontrolled studies and non-random selection. Moderator analyses examined experimental design (pre-post versus placebo-controlled), diagnosis (MDD, BD, mixed/unknown), number of infusions (single versus repeated), and infusion delivery (open-label versus participant-blind). A meta-regression tested percentage of male participants as a predictor of effect size. Publication bias was assessed with Fail-Safe N, Duval and Tweedie’s trim-and-fill, Begg and Mazumdar’s rank correlation and Egger’s test.

Twenty-one studies met inclusion criteria (K = 21), of which 17 were single-infusion studies and 4 involved repeated infusions. Most studies reported outcomes at 4 hours (K = 11) and 24 hours (K = 13); fewer studies reported at 7 days (K = 6) and 12–14 days (K = 4). For single-infusion studies specifically, usable counts were 9 studies at 4 hours, 11 at 24 hours, 5 at 7 days and 2 at 12–14 days. Overall pooled Hedges’ g values were reported as large and statistically significant at each time point analysed (4 h, 24 h, 7 days and 12–14 days). Heterogeneity across studies was large and statistically significant at all time points (I2 values reported as substantial in the paper). Pairwise comparisons between time points did not show significant differences in pooled effect sizes. When comparing repeat versus single infusions, effect sizes for repeat infusions were larger at all time points and differed significantly at 4 hours (g ≈ -3.34 vs -1.11, p = 0.001), 24 hours (g ≈ -4.16 vs -1.11, p < 0.001) and at 12–14 days (g ≈ -2.95 vs -0.85, p = 0.012). The 7-day comparison did not reach significance. The authors emphasise that the number of repeat-infusion studies was small at each time point (K values of 1–2), limiting confidence in these contrasts. Single-infusion studies alone produced significant large effects at 4 hours, 24 hours and 7 days; a large 12–14-day effect did not reach significance, likely because of limited data. Moderator analyses restricted to single-infusion studies indicated diagnostic differences in timing of peak effects. For MDD, effect sizes were large at 4 hours (g ≈ -1.03) and 24 hours (g ≈ -1.35) and moderate at 7 days (g ≈ -0.53). For BD, effect sizes were large at 4 hours (g ≈ -0.80) and especially at 7 days (g ≈ -1.51) while 24-hour effects were moderate (g ≈ -0.64). The MDD versus BD difference was not significant at 4 hours (p = 0.30) but was significant at 24 hours and 7 days (p < 0.001 for both), suggesting a possible shift in peak response timing by diagnosis. Design and blinding also moderated outcomes for single infusions. Pre–post (uncontrolled) designs yielded large, significant effects at all time points. Placebo-controlled trials produced effect sizes ranging from small to large and were significant at 4 hours, 24 hours and 7 days but not at 12–14 days; however, only one placebo-controlled study contributed data at 4 hours and at 12–14 days. Open-label single-infusion studies produced significant effects at 4 hours (g = -1.03, 95% CI = -1.39 to -0.68, K = 7) and 24 hours (g = -1.57, 95% CI = -2.32 to -0.82, K = 3) but were not significant at 7 days in the single available study. Participant-blind studies produced significant effects at 4 hours (g = -0.80, K = 1), 24 hours (g = -1.00, K = 7) and 7 days (g = -0.68, K = 2) but not at 12–14 days (g = -0.21, K = 1). Differences between open-label and participant-blind effect sizes were not statistically significant at the main time points examined. A meta-regression testing sex as a moderator found no significant relationship at 4 hours (p = 0.60) or 24 hours (p = 0.08), but a significant positive association between percentage male and effect size at 7 days (p = 0.008); this result was based on only four data points. Publication-bias assessments indicated evidence of bias for single-infusion studies at 4 hours and 24 hours across several tests. The authors also summarised previously reported response rates variably: around 40% at 4 hours in some studies, approximately 60–70% between 4 and 24 hours in others, but much lower rates in some BD samples (for example 4% at 6 hours and 24% at 24 hours in one report). Side effects were generally transient (headache, dizziness, nausea) and dissociative or psychotomimetic experiences were commonly noted; no major adverse effects were documented in the included trials.

Coyle and Laws interpret the pooled evidence as consistent with a rapid antidepressant effect of ketamine, with large standardised reductions in depressive symptoms evident as early as 4 hours after infusion and persisting, on average, through 7 days and up to 12–14 days in the trials included. Stability of pooled effect sizes across time points is presented as suggestive of a sustained short-term antidepressant response, although longer-term durability beyond two weeks remains unknown given the paucity of follow-up data. The discussion highlights that repeat infusions produced larger effect sizes than single infusions at several time points, but the limited number and selective nature of repeat-infusion studies (some enrolling prior responders) constrain inference about clinical benefit of serial administration. Diagnostic subgroup analyses suggested a temporal difference in peak response—MDD peaking at 24 hours and BD peaking at 7 days—yet the authors caution these findings are tentative because of small K values in subgroup comparisons. Methodological issues are emphasised. Pre–post, open-label designs tended to show larger effects than placebo-controlled, blinded trials, and expectancy bias may contribute when infusions are unblinded. Conversely, blinding is challenging in ketamine trials because of overt psychotomimetic effects; the authors recommend more placebo-controlled trials that preserve blinding, for example by using active comparators such as midazolam. Publication bias at earlier time points and the high heterogeneity across studies are cited as additional limitations. The authors note the potential for selection bias in studies that enrolled known ketamine responders and the consequent risk of inflating effect estimates. Safety and longer-term concerns are discussed: included trials did not report major adverse events, but transient side effects and dissociative experiences were common, and preclinical and observational data raise questions about sensitisation, addiction potential and cognitive or psychotomimetic sequelae with repeated exposure. The authors therefore call for randomised controlled trials with longer follow-up and assessment of safety, relapse patterns and addiction risk. Finally, they acknowledge ketamine’s potential utility in emergency settings for rapid symptom relief but stress that not all patients respond and that response rates vary across populations, underscoring the need for larger, better-controlled studies to define which groups benefit most.

Coyle and Laws conclude that ketamine is an effective and rapid short-term treatment for depressive symptoms in MDD and BD, producing detectable benefit from 4 hours and, based on available trials, up to two weeks post infusion. Repeat infusions do not clearly extend the duration of benefit beyond that seen with a single infusion according to the limited repeat-infusion data, and single and repeated administration appear similarly effective in reducing symptoms in the short term. The authors call for more adequately controlled randomised trials, preferably with active controls that preserve blinding, and for research to clarify ketamine’s role in emergency treatment and its suitability for longer-term management of depression.