Ketamine Infusions for Chronic Pain

A Systematic Review and Meta-analysis of Randomized Controlled Trials

Vwaire Orhurhu, MD, MPH; Mariam Salisu Orhurhu, MD, MPH; Anuj Bhatia, MD, FRCPC; Steven P. Cohen, MD


Anesth Analg. 2019;129(1):241-254. 

In This Article


This systematic review and meta-analysis was conducted according to the recommendations of the Cochrane Collaboration[7] and is reported per Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines.[8] It was registered with an international prospective register of systematic reviews (PROSPERO) on October 17, 2017 with the identifying code CRD42017075521.

Search Strategy and Study Selection

We systematically searched and screened titles and abstracts from MEDLINE, Embase, and Google Scholar, as well as a clinical trials website ( from inception to December 16, 2017. For Embase and MEDLINE, both controlled vocabulary terms (EMBASE-Emtree; MEDLINE-MeSH) and text word searches were conducted for the following search segments: "ketamine," "N-methylketamine," "Norketamine," "S-ketamine," "N-Methyl-D-Aspartate or NMDA antagonist," "CRPS," "complex regional pain syndrome," "fibromyalgia," "pain syndrome," "neuropathic pain," "neuropathy," "pain," "chronic pain," "cancer," "neuralgia," and "analgesia," without language restrictions (Supplemental Digital Content, Table 1, We then reviewed the full articles and applied our selection criteria. We complemented our search by reviewing the bibliographies of selected articles to identify additional articles that were missed in our initial search (Supplemental Digital Content, Table 1, Two authors (V.O. and M.S.O.) independently evaluated titles, abstracts, and full texts. All instances of discordance were discussed between the investigators and the senior author (S.P.C.) to reach consensus.

Criteria for Considering Trials

Type of Study. The search was limited to randomized controlled trials comparing the effect of IV ketamine infusions to a placebo for relieving chronic pain. Animal studies, case series, use of non-IV formulations of ketamine, uncontrolled trials, and review articles were excluded.

Follow-up for Outcomes. Only trials with a follow-up period of ≥48 hours after the cessation of the ketamine infusion were included.

Participants. Only trials performed on human subjects ≥18 years of age who had chronic pain[9–12] for ≥3 months were included in this review. The intensity of pain had to be moderate

or severe (≥4/10 on a numerical rating scale or ≥40/100 on a visual analog scale).

Interventions and Comparators. The intervention was defined as the IV administration of a bolus and/or infusion of ketamine. The comparators were placebo with or without conventional medical management which could include pharmacological, physical, psychological, or interventional therapies. We included studies that allowed for preinfusion pain management regimens to be continued and those that included the same cointerventions for both groups.

Outcomes. The primary outcome was the lowest intensity pain score recorded ≥48 hours after completion of treatment expressed on a 0–10 scale. Intensity scores reported on a visual analog scale were transformed to a 0–10 numerical rating scale because these scales have high correlation.[13,14] Given the paucity of well-designed trials, we chose to include studies that assessed the primary outcome at a wide range of time intervals (48 hours to 12 weeks after administration of ketamine).[15] The following were considered as secondary outcomes: positive response, defined as reduction in pain scores by ≥30% or ≥50% from baseline 48 hours or longer after the intervention, and the incidence of adverse effects. Although ≥50% pain relief is the most common threshold for designating a treatment response as positive,[16] studies have found that ≥30% relief constitutes clinically meaningful improvement.[17] The 48-hour cutoff was chosen because of the need to choose a time point remote from the infusion that would simultaneously be inclusive, and allow us to control for short-term anesthetic effects.

Risk of Bias Assessment

Two authors (V.O. and M.S.O.) independently assessed risk of bias using the Cochrane Collaboration's instrument, which assesses the following domains: generation of the allocation sequence, allocation concealment, blinding of investigators and participants, blinding of outcome assessors, incomplete outcome data, selective reporting, and other possible sources that contain less empirical evidence but collectively may be important (eg, unequal distribution of prognostic factors, industry participation, trial termination).[7] Each item was classified as having low, unclear, or high risk of bias. A decision to classify "overall bias" as low, unclear, or high was made using the following criteria: "High," any trial with a high risk of bias on ≥1 key domains; "Unclear," any trial with an unclear risk of bias on ≥1 key domains; and "Low," any trial without a high risk of bias on any key domain and without significant methodological concerns. We used visual inspection of the funnel plot to assess for publication bias and small study effects, with Begg rank correlation and Egger regression asymmetry tests used to confirm these results if asymmetry was detected provided a sufficient number of studies (≥10) was available.[18]

Data Extraction

Reference data, populations, and outcomes were extracted from articles into prespecified tables using a standardized data extraction form. The data-collection form was pilot-tested before use. We extracted information on studies' general characteristics (publication year, design, number of arms, and outcomes), participants (demographic characteristics and sample size), clinical information (diagnosis, duration, and pain intensity), and experimental intervention (doses and administration regimen). For continuous data, means and SDs were extracted from tables or graphs. If not reported, SDs were obtained from CIs relating to the difference between means. Median values and interquartile ranges were converted to mean and SD if data were normally distributed.[7] We contacted authors when information required about their analysis or results was not reported. In situations where numeric pain scores were not reported, we obtained their estimated values from graphs and figures.

Data Analysis

DerSimonian and Laird random-effects meta-analysis method was used due to expected heterogeneity from diverse populations, diagnoses, and ketamine dosages.[7] Heterogeneity was assessed by inspection of the forest plot and the Q test, and Higgins I2 statistic was used to quantify it.[7] We used meta-regression to evaluate potential sources of heterogeneity. Investigation of sources of heterogeneity was based on analysis of prespecified subgroups for the primary outcome including types of pain syndromes (complex regional pain syndrome versus non–complex regional pain syndrome), mixed or neuropathic versus nonneuropathic (nociceptive or nociplastic) pain, high- versus low-dose ketamine, inpatient versus outpatient setting, studies with >60% females versus those with less, and studies with a median age >46 years of age versus those with a median age under 46 years of age. The rationale for stratifying studies based on age and sex is the effect of these variables on pain processing and neuroplasticity.[19,20] Due to the small number of studies in each subgroup, a meta-regression analysis was used to determine the level of significance between the aforementioned subgroups.

Studies were categorized as either "high-dose" or "low-dose" based on the cumulative amount of ketamine administered in each study. Assuming a 70-kg patient, we calculated the cumulative amount of ketamine from each infusion regimen by multiplying the amount of the assumed weight by the infusion rate (mg/kg/min) and duration of infusion. Cumulative ketamine dose ≥400 mg was defined as "high-dose group," while cumulative dose <400 mg was defined as "low-dose group."

The weighted mean difference in pain scores was calculated with its 95% CI at 48 hours or later after the infusion. A P value of <.05 was considered significant for the primary outcome. The DerSimonian and Laird method was used for calculating the pooled relative risk with corresponding 95% CI for the rate of positive analgesic response and incidence of adverse effects. For adverse effects, proportions of patients with adverse outcomes were pooled. To avoid inducing bias toward a higher event rate from using a fixed continuity correction for studies with a zero-event rate, proportions were transformed via the Freeman–Tukey Double ArcSine method without continuity correction. Pooled estimates were then calculated based on the transformed values using a random-effects model.[7] CIs for pooled estimates were based on the Wald method. Finally, back-transformed pooled proportions were calculated by using Miller inverse transformation with the harmonic mean of the sample size.[21]

A sample size calculation was performed to estimate the number of patients required to detect a difference of 1 or 1.5 point(s) between the ketamine and comparator groups with an SD of 2.0. Sample sizes of 40 or 86 subjects per group, respectively, were required to obtain results with a probability of type I error of 5% and type II error of 10%. We performed a sensitivity analysis by excluding studies one-by-one in a stepwise fashion and reassessing how the new estimate differed. Analyses were performed using STATA version 13 (StataCorp, College Station, TX).

Quality of Evidence

The quality of evidence was classified using Guidelines of Recommendations, Assessment, Development, and Evaluation (GRADE) methodology as high, moderate, low, or very low for each outcome based on the risk of bias, inconsistency, indirectness, imprecision, and publication bias. A summary table was constructed with the GRADEpro guideline development tool (; Evidence Prime Inc, Hamilton, ON, Canada).[22]