The Effect of a Single Dose of Intravenous Ketamine on Suicidal Ideation: A Systematic Review and Individual Participant Data Meta-Analysis

Two reviewers (S.T.W., E.D.B.) searched MEDLINE using the terms “ketamine,” “NMDA receptor antagonist,” “ketamine-like,” or “rapid antidepressant” and “suicide,” “suicidality,” or “suicidal ideation” (filter: clinical trials) for articles published between Jan. 1, 2000, and Nov. 15, 2016. The identified systematic reviews and meta-analyses were searched for relevant published and unpublished research. Additional studies were identified through cross-reference or communication with investigators.

This meta-analysis investigated studies of single-dose intravenous ketamine for the treatment of any psychiatric disorder; only comparison intervention trials (using saline placebo or midazolam as a control) were included. Studies in which multiple ketamine doses were administered were excluded. The corresponding authors of identified publications were contacted and asked to provide individual subject data regarding how suicidal ideation was assessed in each potentially eligible trial. Authors also provided individual suicide assessment scores, individual depression severity scores, and baseline demographic and clinical information.

Patient-level data were collected for several distinct variables, including 1) suicidal ideation, assessed via two clinician-administered and two self-report rating scales (item 10 on the Montgomery-Åsberg Depression Rating Scale [MADRS]; item 3 on the 17-item Hamilton Depression Rating Scale [HAM-D]; item 12 on the Quick Inventory of Depressive Symptomatology–Self Report [QIDS-SR]; and/or item 9 on the Beck Depression Inventory [BDI]); 2) overall severity of depressive symptoms, assessed via total score on the MADRS, HAM-D, QIDS-SR, and/or BDI; 3) treatment assignment (ketamine or control treatment); and 4) potential moderators of treatment effect (age, gender, race, inpatient versus outpatient status, use of concomitant medications). Whenever available, all data were collected from each investigator for baseline and postinfusion days 1, 2, 3, and 7. Our method for handling missing data is described in the data supplement that accompanies the online edition of this article. Table S1 in the data supplement lists which rating scales were used in each study.

Because this study sought to determine the effects of ketamine on suicidal ideation, patients who had no suicidal ideation at baseline were excluded from the analysis. We included active or passive suicidal ideation, which was operationalized a priori as a score ≥2 on MADRS item 10 (“weary of life/fleeting suicidal ideation”) or a score ≥1 on the HAM-D suicidal ideation item (“feels life is not worth living”) for the clinician-administered scales; for the self-report scales, suicidal ideation was defined as a score ≥1 on QIDS-SR item 12 (“I feel that life is empty or wonder if it’s worth living”) or a score ≥1 on BDI item 9 (“I have thoughts of killing myself, but I would not carry them out”).

We used a standard one-stage hierarchical modeling approach (19), with participants nested within studies. A general linear mixed model was used; the dependent variable was suicidal ideation. The specific hypothesis tested using this model was that ketamine would resolve suicidal ideation more rapidly than the control treatment. The following independent variables were used in the first model: baseline suicidal ideation (held constant over time), group, time, and a group-by-time interaction. The following covariates were included in an additional model: age, gender, race, treatment setting (inpatient versus outpatient), diagnosis, and whether the patient was taking concomitant psychotropic medications. In a final model, we dropped terms that had no main effect and adjusted for changes in severity of depressive symptoms over time in an attempt to assess whether ketamine’s effects on suicidal ideation were independent of its effects on other depressive symptoms; in these analyses, the suicide items were removed from the composite depression score.

Dichotomous outcomes (being free of suicidal ideation) among those with some level of baseline suicidal ideation were also analyzed in two separate models (using self-report and clinician-administered rating scales; see the online data supplement).

Effect sizes (Cohen’s d) were calculated using mean differences between baseline and each time point (postinfusion days 1, 2, 3, and 7). For all analyses, the significance threshold was set at 0.05, two-tailed. Because several trials showed evidence of carryover effects associated with ketamine treatment, we analyzed only data from the first treatment (ketamine or control treatment) in crossover trials.

The online data supplement describes how data from different scales were combined.

Correlations between change in suicidal ideation and overall severity of depressive symptoms were calculated using Pearson correlation coefficients. Changes were calculated between baseline and postinfusion days 1, 2, 3, and 7; for this analysis, suicide items were removed from the overall composite depression score.

Eleven eligible trials were identified from 153 citations (Figure 1). Corresponding authors provided individual subject data for 10 of these 11 potentially eligible citations (10–14, 17, 20–23). Correspondence with authors of the remaining study (24) indicated that none of the participants had baseline suicidal ideation and would thus not have been eligible for inclusion in the present analysis. Three of the studies were conducted at the National Institutes of Health (11–13) under a single protocol and were included in a review of ketamine’s ability to reduce suicidal ideation (15). Because these three studies were done under a single protocol, for the purposes of statistical modeling we considered this a single study (hence, k=8). In addition to participants from these published trials, 36 additional participants analyzed in the earlier review (15) were included in this analysis.

Individual patient-level data were obtained for 298 patients who participated in the 10 included ketamine trials; 167 patients met criteria for baseline suicidal ideation. Study characteristics are summarized in Table 1, and patient demographic characteristics in Table S2 in the online data supplement. Patients who were included for analysis did not differ significantly from those who were excluded in mean age (t=0.105, p=0.917) or in gender distribution (χ²=0.27, p=0.604). Overall, included patients had more severe depressive symptoms at baseline than excluded patients as measured by the MADRS (mean score, 33.4 compared with 25.9; t=9.08, p<0.001), the HAM-D (mean score, 20.5 compared with 16.3; t=4.8, p<0.001), the QIDS-SR (mean score, 17.7 compared with 14.2; t=4.81, p≤0.001), and the BDI (mean score, 29.2 compared with 21.1; t=5.62, p<0.001). Included patients were also more likely than excluded patients to be receiving concomitant psychotropic medications (47.9% compared with 27.5%; χ²=12.9, p≤0.001).

Among the included patients, those who received ketamine did not differ significantly from those who received a control treatment in mean age (t=−0.27, p=0.788), gender distribution (χ²=1.01, p=0.314), diagnoses (χ²=0.40, p=0.527), inpatient versus outpatient status at time of drug exposure (χ²=0.52, p=0.470), proportion receiving concomitant psychotropic medications (χ²=1.23, p=0.268), or baseline MADRS score (t=0.867, p=0.388).

As assessed via clinician-administered rating scales, ketamine reduced suicidal ideation more rapidly than was observed with the control treatments on the MADRS and the HAM-D, with significant benefits appearing as early as day 1 after treatment and extending up to day 7 (total N=167, k=8; group-by-time interaction, χ²=50.6, p<0.001) (Figure 2A). The benefits of ketamine compared with control treatments remained significant after baseline covariates (age, gender, race, use of concomitant psychotropic medications, outpatient versus inpatient status, diagnosis) were adjusted for in the model (group-by-time interaction, χ²=50.5, p<0.001). None of the baseline variables significantly moderated ketamine’s effects on suicidal ideation.

The mean MADRS scores for the group exposed to ketamine were 33.8 (SD=6.8) at baseline, 19.5 (SD=12.3) at day 1, 19.4 (SD=12.1) at day 2, 20.1 (SD=12.2) at day 3, and 22.0 (SD=11.5) at day 7. The mean MADRS scores for the group exposed to control treatments were 32.9 (SD=6.5) at baseline, 29.2 (SD=9.3) at day 1, 29.1 (SD=10.5) at day 2, 27.3 (SD=9.8) at day 3, and 27.5 (SD=10.0) at day 7.

When each clinician-administered rating scale outcome was analyzed separately, ketamine continued to reduce suicidal ideation significantly more rapidly than control treatments on both the MADRS (group-by-time interaction, χ²=35.0, p<0.001) (Figure 2B) and the HAM-D (group-by-time interaction, χ²=19.4; p<0.001) (Figure 2C). Meta-analysis demonstrated little heterogeneity between studies (F=0.027, df=1, 791, p=0.87). Effect sizes (group difference divided by pooled standard deviation) for ketamine on change in suicidal ideation were moderate to large for the clinician-administered rating scales at all time points (at day 1, Cohen’s d=0.85, 95% CI=0.53–1.17; at day 2, d=0.85, 95% CI=0.52–1.17; at day 3, d=0.67, 95% CI=0.35–0.99; at day 7, d=0.61, 95% CI=0.27–0.94). In addition, ketamine was associated with a significantly greater proportion of patients being free from suicidal ideation compared with control treatments, as assessed by clinician-administered ratings, at postinfusion days 1, 2, 3, and 7; over half of the participants reported no suicidal ideation across all postinfusion time points (all t values <−2.95, all p values <0.005) (Figure 3A). The number needed to treat for ketamine (compared with control treatment) for being free of suicidal ideation was in the range of 3.1–4.0 for all time points 1 to 7 days after ketamine infusion.

Using self-report outcome measures, ketamine similarly reduced suicidal ideation significantly more rapidly than was observed with the control treatments, with significant benefits noted at each individual time point (N=144, k=8; group-by-time interaction, χ²=45.5, p<0.001) (Figure 4A). The benefits of ketamine compared with control treatments remained significant after baseline covariates were adjusted for in the model. None of the baseline variables significantly moderated ketamine’s effects on suicidal ideation. When each self-report measure was analyzed separately, ketamine reduced suicidal ideation significantly more rapidly than was observed with the control treatments on the QIDS-SR (group-by-time interaction, χ²=32.5, p<0.001) (Figure 4B) but not on the BDI (group-by-time interaction, χ²=8.34, p=0.08) (Figure 4C). Notably, these analyses had smaller sample sizes (N=77 and N=67, respectively) because not all studies used each measure. Meta-analysis demonstrated little heterogeneity between studies (F=0.097, df=1, 671, p=0.76).

Effect sizes (group difference divided by pooled standard deviation) of ketamine on change in suicidal ideation were moderate to large on the self-report outcomes (at day 1, Cohen’s d=0.73, 95% CI=0.38–1.07; at day 2, d=0.84, 95% CI=0.49–1.19; at day 3, d=0.63, 95% CI=0.28–0.98; at day 7, d=0.48, 95% CI=0.12–0.83). Ketamine was again associated with a significantly greater proportion of patients being free from suicidal ideation compared with the control treatments at postinfusion days 1, 2, and 3 (all t values <−2.30; all p values <0.05) but not at day 7 (t=−1.18, p=0.238) (Figure 3B). The number needed to treat for ketamine (compared with control treatments) for being free of suicidal ideation was in the range of 3.2–5.0 for days 1–3 after ketamine infusion, and 9.6 at day 7.

Changes in suicidal ideation and overall severity of depressive symptoms were strongly correlated at all time points. As measured using clinician-administered rating scales, change in suicidal ideation and change in severity of depressive symptoms were significantly correlated at day 1 (adjusted r²=0.411, t=10.73, p<0.001), day 2 (adjusted r²=0.460, t=11.66, p<0.001), day 3 (adjusted r²=0.370, t=9.60, p<0.001), and day 7 (adjusted r²=0.414, t=10.08, p<0.001). On the self-report measures, change in suicidal ideation and change in severity of depressive symptoms were significantly correlated at day 1 (adjusted r²=0.405, t=9.75, p<0.001), day 2 (adjusted r²=0.212, t=5.46, p<0.001), day 3 (adjusted r²=0.208, t=5.95, p<0.001), and day 7 (adjusted r²=0.103, t=3.83, p<0.001).

After adjusting for change in severity of depressive symptoms over time, the time-by-treatment interaction remained significant regardless of whether clinician-administered (χ²=10.84, p=0.028) or self-report (χ²=13.19, p=0.010) outcome measures were used. Significant differences were observed between groups at all postinfusion time points on clinician-administered outcome measures (all t values <−2.30, all p values <0.05). On the self-report outcome measures, significant differences were noted at day 1 (t=−2.80, p=0.005), day 2 (t=−2.62, p=0.009), and day 3 (t=−1.99, p=0.047) but not day 7 (t=−0.36, p=0.720).

Among the group of patients who achieved a resolution of suicidal ideation (as measured by clinician-administered measures) by 24 hours after infusion, the effect of ketamine on reduced suicidal ideation persisted for up to 1 week in 86.0% of patients, compared with 52.9% of patients who received control treatments (χ²=7.98, p=0.005). Using self-report outcomes, the effect of ketamine on reduced suicidal ideation persisted for up to 1 week in 89.2% of patients in the ketamine group, compared with 42.9% who received control treatments (χ²=12.1, p<0.001).

To assess the differences between midazolam and saline as control treatments, we calculated effect sizes (group difference divided by pooled standard deviation) separately for saline and midazolam as control condition. The outcome measure was the change in suicidal ideation from baseline on clinician-administered outcome measures at each postinfusion time point. When only saline was used as a comparator (N=50 for the saline group, N=93 for the ketamine group), the effect sizes were as follows: at day 1, d=0.90, 95% CI=0.54–1.26; at day 2, d=0.90, 95% CI=0.53–1.26; at day 3, d=0.82, 95% CI=0.45–1.19; at day 7, d=0.69, 95% CI=0.31–1.07. When only midazolam was used as a comparator (N=24 for the midazolam group, N=93 for the ketamine group), the effect sizes were more modest: at day 1, d=0.62, 95% CI=0.16–1.08; at day 2, d=0.69, 95% CI=0.22–1.15; at day 3, d=0.34, 95% CI=−0.12–0.80; at day 7, d=0.41, 95% CI=−0.06–0.88. Comparing the effect of midazolam (N=24) versus saline (N=50) on suicidal ideation in the linear mixed model showed no group-by-time interaction nor a main effect of group.

There is considerable interest regarding whether ketamine’s antisuicidal properties occur independently of its general antidepressant effects (pseudospecificity), particularly because this may have an impact on the path toward U.S. Food and Drug Administration (FDA) approval for ketamine or related compounds in treating suicidal ideation or behavior (26). A related example of pseudospecificity is vortioxetine, which underwent the FDA New Drug Application process for potential approval for cognitive dysfunction in major depressive disorder. The application was submitted based on the consideration that cognitive dysfunction may be phenomenologically distinct from other symptoms of depression and that its course may be distinct from those of other depressive symptoms (27–29). Ultimately, the FDA did not grant approval for this pseudospecific indication for vortioxetine. However, vortioxetine did receive approval from the European Medicines Agency for a type II variation (a process similar to that of the FDA for pseudospecificity) for the treatment of cognitive dysfunction in major depressive disorder.

The present analysis provides evidence drawn from the largest sample to date that ketamine reduces suicidal ideation partially independently of mood symptoms. However, the specificity of this effect requires further exploration. While we purposefully included patients across a range of psychiatric diagnoses, most of the patients had major depressive disorder or bipolar depression and had treatment-resistant illness per inclusion criteria. For ketamine to have a future as a potential antisuicidal therapeutic in patients with a range of mood and anxiety disorders, studies specifically recruiting diverse populations must be conducted. Future studies should also consider whether ketamine may be a potential therapeutic for suicidal ideation in patients of varying levels of treatment resistance. This is particularly important because in the emergency settings where antisuicidal interventions might be most useful, clinical decisions must often be made quickly in the face of potential ambiguity of both diagnosis and treatment history. In one of the included studies, Murrough and colleagues (17) recruited a population considered at risk for suicide across diagnoses in an attempt to specifically measure ketamine’s effect on suicidal ideation; this differed from other trials included in the meta-analysis, which recruited patients with specific diagnoses and excluded participants deemed at imminent risk of suicide. Notably, patients with diagnoses such as substance use disorder and schizophrenia/schizoaffective disorder were uniformly excluded in these studies. Given ketamine’s abuse liability (30) and concern for exacerbation of psychotic symptoms (31), additional research is clearly required before ketamine can be considered a treatment option for these patient populations.

Despite the robust findings described above, several limitations require comment. First, the relatively small sample sizes of the included studies limited sensitivity analyses of the individual scales, which may account for our mixed findings in the BDI analysis; nevertheless, three of the four scales (both clinician-administered and self-report instruments) yielded consistent findings. Second, all of the studies included in this meta-analysis examined the effects of ketamine on suicidal ideation; whether ketamine’s effects on suicidal ideation translate to effects on suicidal behavior has not been studied. Third, with one exception (17), the included studies did not specifically recruit patients deemed at imminent risk for suicide. Thus, the assessment of suicidal ideation used in this analysis is limited to a single item from each scale. This limitation, however, is mitigated by our exclusion of patients who had no suicidal ideation at baseline. Fourth, while our finding that ketamine exerts significant effects on suicidal ideation even when the severity of depressive symptoms was controlled for was replicated in a similar analysis of open-label ketamine (32), the bulk of the current ketamine studies have been conducted in patients with active mood disorders. Fifth, the short follow-up of the studies precludes this analysis from providing further guidance on any sustained effects beyond 7 days after treatment. Finally, given the psychoactive properties of ketamine, a limitation of many, although not all, of the included studies is the possibility of functional unblinding of trials using saline as the comparator.

Despite robust findings here and elsewhere indicating that ketamine has significant antisuicidal effects, a number of questions must be answered before ketamine can be regularly used in clinical settings to treat patients at risk of suicide. While ketamine’s antisuicidal effects appeared to be maintained over 1 week, the possibility of rebound suicidal ideation remains, with potential negative outcomes within the weeks or months following exposure to ketamine (33). In addition, to date no study has demonstrated that ketamine specifically reduces the risk of suicidal behavior, only of suicidal ideation. Results from several ongoing clinical trials evaluating the efficacy of ketamine or esketamine (ketamine’s S-enantiomer) to stabilize outpatients who have significant suicidal ideation (NCT02094898, NCT01700829) or patients who are hospitalized because of acute suicide risk (NCT02133001, NCT02299440) should provide longer-term follow-up data as well as protocols for repeated dosing. Future studies should also explore the effects of combining ketamine with established somatic/pharmacologic (i.e., ECT, lithium [NCT01880593]) or psychotherapeutic (i.e., dialectical behavioral therapy or CBT) modalities to extend beneficial response. One preliminary study of ketamine combined with CBT reported improved longer-term outcomes in mood disorders (34), although we know of no such protocols examining similar combinations specifically for the treatment of suicidal ideation. Future clinical studies should also assess details of suicidal ideation, including frequency, intensity, duration, and past suicidal behaviors. Differentiation between acute and chronic factors in the emergence of suicidal ideation may also help establish expectations regarding treatment outcomes. While using midazolam as a control may attenuate the effect size of ketamine because of improved integrity of blinding, our study did not find a significant difference between the two control conditions, which may have been due to the relatively small number of patients who received midazolam. Also, no study has directly compared the effects of saline and midazolam on suicidal ideation.