Efficacy of ketamine for major depressive episodes at 2, 4, and 6-weeks post-treatment: A meta-analysis

Major depressive episodes produce severe, persistent symptoms that impair functioning across social, occupational and other domains and occur in both major depressive disorder (MDD) and bipolar I and II disorder. Standard treatments, including antidepressant medications and psychotherapies, are effective for many patients but often require weeks to months to yield clinical benefit. Because ketamine produces rapid reductions in depressive symptoms and suicidal ideation at subanaesthetic doses, it has attracted clinical and research interest as a potential short‑acting but rapid antidepressant. Conley and colleagues set out to evaluate whether ketamine's antidepressant effects extend beyond the short time windows typically considered in prior meta‑analyses. Specifically, the present meta‑analysis examined the efficacy of ketamine on depressive symptom severity at 2, 4 and 6 weeks after administration. By pooling randomized and open‑label human studies that measured depressive symptoms with the MADRS or HDRS, the study aimed to quantify effect sizes at these later time points and to examine moderators such as dose, treatment length and frequency, route and study design.

This work is a meta‑analysis. A comprehensive search of PsycINFO, MEDLINE, ClinicalTrials.gov, PubMed and Embase was performed up to December 2019 using terms combining “ketamine” with depression‑related and study‑design keywords; reference lists of identified studies were also screened. Two independent reviewers selected studies, with a third reviewer resolving disagreements. Eligible studies enrolled adults (≥18 years) with MDD or bipolar disorder who received ketamine and reported depression outcomes at 2, 4 or 6 weeks post‑treatment using the Montgomery–Åsberg Depression Rating Scale (MADRS) or the Hamilton Depression Rating Scale (HDRS). Both randomized and open‑label designs were eligible; animal studies and non‑English reports were excluded. Three studies were ultimately excluded for selective outcome reporting, producing a final set of included studies. Study characteristics were coded for sample size, mean age, gender proportion, treatment strategy (primary or adjunctive), treatment length (<1 week, 1–2 weeks, >2 weeks), route (intravenous versus other), dosage (categorised as low when <0.5 mg/kg or ≤56 mg, and high when ≥0.5 mg/kg or ≥84 mg/100 mg equivalent), treatments per week, primary outcome measure (MADRS or HDRS), sample comparison (within‑subjects versus between‑subjects) and study design (open‑label versus randomized). When means and standard deviations were missing, the authors attempted to contact study authors, used WebPlotDigitizer to extract data from figures, and applied multiple imputation to replace missing values; imputation results are reported in the supplement. For analysis, effect sizes were calculated as Hedges’ g (standardised mean difference) and interpreted using conventional thresholds (small ≈ 0.2, medium ≈ 0.5, large ≈ 0.8). Random‑effects models were used throughout. Heterogeneity was assessed with Cochran’s Q (significance set at p < .10) and quantified with I2 (25%, 50% and 75% interpreted as low, moderate and high). Moderator effects were examined with subgroup analyses and meta‑regression; Egger’s regression intercept and funnel plots were used to assess publication bias when k ≥ 10. Multiple imputation was applied for missing standard deviations and, where necessary, for missing means; analyses incorporating imputed data are provided in the supplement.

The search identified 44 studies that initially met inclusion criteria; three were excluded for selective reporting, leaving 41 studies for analysis. All included studies contributed data at the 2‑week time point (k = 41), 31 at 4 weeks and 5 at 6 weeks. When studies reported multiple treatment lengths or dosages, the authors conservatively coded the highest length and highest dosage reported. At 2 weeks post‑treatment the pooled effect of ketamine on depressive symptoms was large: Hedges’ g = -1.28 (95% CI: -1.79 to -0.77), t(39) = -5.10, p < .01. Heterogeneity was high (Q = 474.22, p < .01; I2 = 91.8%). Funnel‑plot inspection and Egger’s test did not indicate significant publication bias for this time point (Egger intercept = -0.53, t = -0.31, p = .76). Subgroup analyses (Bonferroni‑adjusted) identified treatment length, dosage, treatments per week and study design as significant moderators. Within those subgroups, high‑dosage studies showed larger effects than low‑dosage studies (g = -1.40 versus g = -0.38). Studies reporting six treatments per week had larger effects than studies with a single treatment per week (g = -1.78 versus g = -0.80); trend‑level larger effect sizes were also noted for two and three treatments per week. Open‑label studies produced larger pooled effects than randomized trials (g = -2.19 versus g = -0.96). Although treatment length was a significant moderator overall, pairwise contrasts did not survive the Bonferroni adjustment; numerically, studies with treatment length <1 week had lower effects (g = -0.80) than those with 1–2 weeks (g = -2.15) or >2 weeks (g = -1.97). At 4 weeks post‑treatment the pooled effect remained large: g = -1.28 (95% CI: -1.66 to -0.90), t(29) = -6.89, p < .01, with high heterogeneity (Q = 239.48, p < .01; I2 = 87.9%). Publication‑bias testing did not show asymmetry (Egger intercept = -1.76, t = -1.11, p = .28). The only significant moderator at 4 weeks was sample comparison: within‑subject designs had larger effects (g = -1.66) than between‑subject designs (g = -0.86). At 6 weeks post‑treatment the pooled effect was also large: g = -1.36 (95% CI: -2.69 to -0.04), t(4) = -2.85, p = .05. Heterogeneity remained high (Q = 18.4, p < .01; I2 = 78.3%). Funnel‑plot assessment was limited by the small number of studies and Egger’s test was not conducted (k < 10). Meta‑regression identified age as a significant predictor at 6 weeks (β = -5.27, R2 = 54.16%, p < .05), with younger participants associated with larger effect sizes. Primary outcome measure moderated effects at 6 weeks: studies using the HDRS showed larger pooled effects (g = -3.20) than those using the MADRS (g = -0.80), but these subgroup results are based on very small counts (HDRS k = 2; MADRS k = 3). Across time points, heterogeneity was consistently high and several moderator analyses were constrained by small subgroup sizes. Multiple imputation was used to address missing data and analyses including imputed values are reported in the supplement.

Conley and colleagues conclude that ketamine produces large reductions in depressive symptoms that persist up to 6 weeks after administration, extending prior meta‑analytic work that primarily focused on very short‑term outcomes (e.g. 24 hours to 2 weeks). The pooled effects at 2, 4 and 6 weeks (g ≈ -1.28, -1.28 and -1.36 respectively) led the authors to describe the sustained antidepressant potential of ketamine and to note clinical implications, such as the possibility of spacing administrations to reduce cumulative side‑effect burden while providing patients time to engage with longer‑term treatments. The authors interpret the moderator results as suggesting that clinical variables (dose, treatment length, number of administrations) are more relevant to outcomes at around 2 weeks, whereas study‑related variables (sample comparison, outcome instrument) emerged as moderators at 4 and 6 weeks. Specifically, higher doses and greater numbers of administrations were associated with larger effects at 2 weeks, and open‑label designs and within‑subject comparisons tended to show larger effect sizes than randomized or between‑subject designs. The age effect at 6 weeks—larger effects among younger participants—is reported as preliminary because the 6‑week analysis was based on few studies. Key limitations acknowledged by the authors include high between‑study heterogeneity, small numbers of studies for some subgroup comparisons and for the 6‑week time point (k = 5), and missing data that required multiple imputation (with imputed‑data analyses presented in the supplement). Another important limitation was the inability to code diagnosis (MDD versus bipolar depression) reliably across studies because many reports did not specify participant counts by diagnosis; as a result, subgroup analyses by diagnosis were not performed. The authors emphasise that subgroup findings based on small k should be interpreted cautiously. In terms of implications, the authors note that evidence for sustained benefit up to 6 weeks may inform clinical planning for treatment frequency and adjunctive care, and that further research is needed to clarify optimal dosing, administration schedules, safety and comparative effects by diagnosis and route of administration. They also highlight the need for additional high‑quality randomized trials with longer follow‑up to confirm durability of response and to reduce uncertainty arising from heterogeneity and small subgroup samples.

The authors conclude that ketamine is an encouraging pharmacologic intervention for major depressive episodes, producing rapid antidepressant effects that, according to their pooled analyses, can last up to 6 weeks post‑treatment. Such sustained effects could provide clinically meaningful windows for patients to access and engage with longer‑term treatments, such as evidence‑based psychotherapy, and may allow spacing of ketamine administrations to reduce cumulative exposure and side effects. They emphasise the need for further research to better understand side‑effect profiles, optimal dosing and administration schedules, comparative effectiveness across routes (including intranasal esketamine) and differential effects by age and diagnostic subgroup. Given regulatory changes and increasing clinical availability, the authors argue that more evidence on duration of benefit and safety is essential as ketamine treatments become more accessible.