Patient-controlled analgesia with opioids is commonly used for treatment and prevention of pain in the perioperative setting. Nevertheless, analgesic success is often limited by opioid-related adverse events, including postoperative nausea and vomiting (PONV), sedation, respiratory depression, ileus, urinary retention, and pruritus. Ketamine, an antagonist of the N-methyl-D-aspartate (NMDA) receptor and an inexpensive and potentially opioid-sparing drug, is of increasing interest in pain management, especially in subanesthetic doses. Concurrent treatment with ketamine has been purported to produce comparable or synergistic analgesia while potentially reducing the risk of opioid-related adverse effects. In addition, given its mechanism of action within the pathophysiology of pain as a NMDA antagonist centrally and peripherally, it has been suggested that ketamine might provide advantages for reducing the risk of progressing to chronic pain, though this claim awaits definitive clinical trials.1 Nevertheless, ketamine also brings dose-related disadvantages, including neuropsychiatric effects (e.g., hallucinations, vivid dreams, and nightmares), cardiovascular adverse effects (e.g., hypertension, tachycardia), and other adverse events (e.g., nausea, dizziness, and blurred vision). Furthermore, the dose required for adequate analgesia remains unclear.

Previously published systematic reviews2-6 of ketamine for acute postoperative pain did not include the most recent randomized trials, did not limit studies to a combination of ketamine and morphine or hydromorphone for patients receiving patient-controlled analgesia (PCA),2,4 neglected to provide transparent effect sizes or clinically relevant measures of analgesia, and/or generally failed to address important subgroups of interest to our clinical practice.2,3,6 For the above reasons, we performed a de novo comprehensive systematic review and meta-analysis of randomized trials to address the following question adequately: Does ketamine added to morphine or hydromorphone PCA provide clinically relevant reductions in postoperative pain, opioid requirements, and opioid-related adverse events without undue risk of neuropsychiatric effects when compared with morphine or hydromorphone PCA in adults undergoing surgery?


The systematic review was conducted according to a protocol that predefined the inclusion criteria, relevant outcomes, and analysis plan and was reported in accordance with the PRISMA Statement.7 Post hoc amendments to the protocol included the addition of hydromorphone PCA to our originally planned meta-analysis of morphine only PCA. We introduced this change to increase the generalizability of the results. Submission to an ethics review board was not required for approval of this meta-analysis.

Data sources and searching

In collaboration with medical librarians, we performed systematic searches—from inception to June 2, 2015—of six databases, including PubMed, EMBASE™, Cochrane Central Register of Controlled Trials (CENTRAL), and three Chinese databases, including Chinese Biomedical Literature Database (CBM), China National Knowledge Infrastructure (CNKI), and Wanfang Data. Search terms included both MeSH headings and free text for “ketamine”, “patient controlled analgesia”, “postoperative”, “surgery”, “pain”, and “randomized controlled trials” (Search strategies in the Appendix, available as Electronic Supplementary Material). No limits were placed on language, type of surgery, and mode of ketamine administration. Bibliographies of relevant systematic reviews and included studies were manually checked to identify potentially relevant studies.

Study selection

To be eligible for inclusion, the studies had to be randomized controlled trials (RCTs) with parallel group designs comparing the combination of ketamine plus morphine or hydromorphone in PCA vs morphine/hydromorphone PCA for acute postoperative pain in adults. Only studies using subanesthetic doses of ketamine were included, defined as a bolus dose of ≤ 2 mg·kg−1 when given intramuscularly, or ≤ 1 mg·kg−1 when administered via intravenous or epidural route, or an intravenous infusion rate of ≤ 20 μg·kg−1·min−1.8 The primary outcome measures of interest were acute pain scores at rest and during mobilization. Secondary outcome measures included cumulative morphine consumption (after converting hydromorphone and/or other supplemental opioids to the morphine equivalent dose), patient satisfaction, total rescue narcotics, PONV, and other adverse events (e.g., respiratory depression, drowsiness, pruritus, dizziness, hallucinations, vivid dreams or nightmares, and cardiovascular adverse effects such as hypertension and tachycardia).

Two reviewers screened citations independently and retrieved the full text of any article deemed potentially eligible. Subsequently, two pairs of reviewers independently assessed the eligibility of each full-text article (L.W. & A.K., or L.W. & F.Z.). Reviewers resolved discrepancies by discussion and, when necessary, through arbitration by a third reviewer (J.M. or D.C.).

Data extraction

Using standardized piloted forms, two reviewers independently extracted data on patient characteristics (mean age, sex, American Society of Anesthesiologists physical status, types and mean duration of surgery), interventions (ratio of ketamine-to-morphine, route, timing, dose, duration of ketamine and morphine, total dose of ketamine, maximum length of follow-up, type of anesthesia, postoperative analgesia, rescue analgesia, and nitrous oxide), and all relevant clinical outcomes.

We extracted pain intensity scores and cumulative morphine consumption for the following time points: four to six hours, ten to 12 hr, 24 hr, 48 hr, and 72 hr postoperatively. All other outcomes were extracted for the last reported time point. When the article did not report whether the pain score was measured at rest or during mobilization, we assumed it was at rest. When authors reported measuring pain at rest and during movement but neglected to differentiate the two in their study results, we assumed the pain score was collected during mobilization. Subsequently, we conducted a sensitivity analysis to address the robustness of these assumptions by using only the data in which the designation of pain was unequivocally either at rest or during movement.

Risk of bias assessment

Two reviewers independently assessed the risk of bias of the included trials using the methods recommended by the Cochrane Collaboration,9 including random sequence generation, allocation concealment, missing or incomplete outcome data, and blinding of patients, study personnel, and outcome assessors.

Data analysis

When standard deviations (SDs) were not reported directly, we estimated the SD from standard errors, confidence intervals, Student’s t values, and P values using the methods recommended in the recent Cochrane Handbook.9 When the articles reported the median with interquartile range, range, or P value, we assumed that the distribution of data was normal and estimated the mean and SD. If the article reported the data using a frequency table, we calculated the mean and SD. If authors presented continuous data in figures, we measured the mean and standard deviation from the figures. Sensitivity analysis was performed by excluding the estimated mean and SD to explore if these methods of imputing mean and SD had an impact on the overall effect size. If a study reported both nausea and vomiting separately, we chose the largest number of events to estimate the PONV. Sensitivity analysis was applied by excluding the estimated PONV.

Weighted mean difference (WMD) and 95% confidence interval (CI) were calculated for continuous data, and relative risk (RR) and 95% CI were calculated for binary data using the random effects model. Risk differences (RD) were also presented for binary data, including all adverse events. When scores for pain and patient satisfaction were reported on a 100-mm visual analogue scale (VAS), a five-point verbal rating scale, or other scales, they were converted to a 10-cm VAS.10

Both the Chi square test and I2 were used to estimate the heterogeneity across trials. If P < 0.10 or I2 > 50%, we used subgroup analyses or meta-regression (for variables with more than two categories) to detect potential sources of heterogeneity. Post hoc subgroup analyses included different types of opioid (morphine PCA vs hydromorphone PCA, with postulated larger effects with hydromorphone PCA) for the outcomes of pain on VAS, cumulative opioid consumption in morphine equivalents, patient satisfaction, and PONV. When heterogeneity was found for the primary outcome (i.e., pain) and was not explained by type of opioid PCA, predefined subgroup analyses were performed for type of surgery (postulated larger effects for cardiac or thoracic surgeries), mode (postulated larger effect with ketamine in PCA vs infusion), duration of ketamine administered (postulated longer duration would lead to larger effects), use of nitrous oxide (larger effects when nitrous oxide is coadministered) and other postoperative analgesics (larger effects with other analgesic coadministration), risk of bias (larger effects for high risk of bias studies), and language of publication (larger effects in non-English studies). Meta-regression was performed to evaluate the ketamine dose-response relationship and the association of the ratio of ketamine-to-morphine with the effect estimates of postoperative pain. For subgroup analysis, the test for interaction was used to evaluate whether effect sizes differed significantly across subgroups. This test is recommended9 rather than relying on the blunt test of significance where each subgroup is considered separately to detect whether each reached P < 0.05. For meta-regression, the P value for the slope was calculated. All reported P values are two sided.

Both visual inspection of funnel plots and Egger’s test were used to explore publication bias for pain and morphine consumption at 24 hr and 48 hr, PONV, and hallucinations.

We used GRADE to summarize confidence in estimates of effect (quality of evidence) in the critically important outcomes for decision-making, including pain scores at 24 hr and 48 hr, PONV, and the risk of other adverse events (e.g., hallucination and vivid dreams).11-17


The search identified 801 articles, of which we included 36 RCTs, 31 published in English18-48 and five in Chinese49-53 (Fig. 1).

Fig. 1
figure 1

Combination of ketamine and morphine/hydromorphone patient-controlled analgesia (PCA) vs morphine/hydromorphone PCA: flow diagram of study selection

Characteristics of included studies

Supplementary Table 1 (available as Electronic Supplementary Material) illustrates the characteristics of the included studies. Thirty-three trials18-45,49-53 comprising 2,374 patients compared ketamine plus morphine PCA in 37 treatment arms with morphine PCA in 33 control arms. Three trials involving 128 patients46-48 compared ketamine plus hydromorphone PCA with hydromorphone PCA. Patients underwent a variety of surgeries: abdominal surgery (17 trials),18,20-22,25-27,29,30,32,34,36,40,41,44,45,50 cardiac or thoracic surgery (six trials),19,23,35,37,38,53 orthopedic surgery (ten trials),24,28,31,33,39,43,47-49,51 and other surgery (three trials).42,46,52 General anesthesia was used in 34 trials; one trial used PCA during and after a uterine artery embolization procedure,29 and one used either general or regional anesthesia.43 The target patients in one study were male opium abusers undergoing orthopedic surgery with morphine PCA for postoperative pain control,24 and in another three studies, patients with chronic pain were managed with opioids preoperatively and with hydromorphone PCA for postoperative pain control.46-48

Table 1 Summary of results for combination of ketamine plus morphine/hydromorphone PCA vs morphine/hydromorphone PCA

Ketamine was administered via PCA in 26 trials, with a ketamine:morphine ratio of 1:1 (13 trials),19,27-30,32,35,39-41,43,49,51 5:1 (four trials),23,31,37,38 2.5:1 (one trial),44 2:1 (four trials),22,24,51,53 0.75:1 (one trial),36 0.5:1 (three trials),20,42,52 and 0.04 or 0.07:1 (one trial).50 Ketamine was administered by infusion in 11 trials.18,22,25,26,30,33,34,38,39,45-48,51,52 Ketamine was given perioperatively in 11 trials20,21,27,33,34,39,42,45,47,48,52 and postoperatively in 25 trials.18,19,22-26,28-32,35-38,40,41,43,44,46,49-51,53 Ten trials stated that acetaminophen,19,25,27,29,35,41,43,45 midazolam,30 and epidural bupivacaine47 were used for postoperative analgesia besides morphine or hydromorphone PCA. Also, intraoperative nitrous oxide was used in 18 trials.18,26-28,32,35,38-45,48,50-52 The duration of follow-up ranged from four to 100 hr postoperatively (Supplementary Table 1).

Risk of bias

Among 36 trials, adequate sequence generation was reported in 23 trials,20-25,29,31-35,37-39,41-43,45-48,50 allocation concealment in 21 trials,20-23,25,27,29-35,38,39,42,43,45-48 blinding in 31 trials,18,20-23,25-48,50,53 and incomplete outcome data in 18 trials.26,27,31,35-37,39,41-51 Overall, 21 out of 36 included trials were rated as low risk of bias,20-23,25,27,29-35,37,38,41-43,45,46,48 and 15 trials were rated as high risk of bias18,19,24,26,28,36,39,40,44,47,49-53 (Supplementary Table 2).

Primary outcome measure: pain intensity

As shown in Tables 1 and 2, the reduction in VAS score in the ketamine plus morphine or hydromorphone PCA group was low but statistically significant compared with the morphine or hydromorphone PCA group at every time point examined (i.e., six, 12, 24, 48, and 72 hr) postoperatively. The reduction of pain at rest ranged from 0.6 cm (95% CI, 0.3 to 0.8) at 24 hr to 1.3 cm (95% CI, 0.2 to 2.4) at 72 hr; and reduction of pain during mobilization ranged from 0.4 cm (95% CI, 0.2 to 0.6) at 24 hr to 0.5 cm (95% CI, 0.2 to 0.8) at 48 hr (Figs. 2 & 3, Supplementary Figs. 1 & 2, and Table 1). The GRADE ratings of confidence in estimates varied from moderate to high (Table 2).

Table 2 GRADE evidence profile: combination of ketamine and morphine/hydromorphone PCA vs morphine/hydromorphone PCA for acute postoperative pain
Fig. 2
figure 2

Combination of ketamine and morphine/hydromorphone patient-controlled analgesia (PCA) vs morphine/hydromorphone PCA: 24-hr pain score at rest on a 0-10 scale

Fig. 3
figure 3

Combination of ketamine and morphine/hydromorphone patient-controlled analgesia (PCA) vs morphine/hydromorphone PCA: 24-hr pain score during mobilization on a 0-10 scale

Secondary outcome measures

Cumulative morphine consumption and rescue analgesia

The addition of ketamine to PCA reduced cumulative morphine consumption compared with morphine or hydromorphone PCA after converting total opioid consumption to morphine equivalents. Reductions in total morphine consumption ranged from 5.0 mg (95% CI, 2.8 to 7.2) at 24 hr to 20.2 mg (95% CI, 12.7 to 27.7) at 72 hr (Table 1, Fig. 4, and Supplementary Fig. 3).

Fig. 4
figure 4

Combination of ketamine and morphine/hydromorphone patient-controlled analgesia (PCA) vs morphine/hydromorphone PCA: 24-hr cumulative morphine consumption (mg)

Numerically fewer patients using ketamine plus morphine/hydromorphone PCA required rescue analgesia, but significant differences were not found (14 trials; 1,069 patients; RR, 0.76; 95% CI, 0.56 to 1.05) (Table 1).

Patient satisfaction

Seven trials of 383 patients reported patient satisfaction scores using different instruments (five trials used 10-cm VAS for satisfaction;22,23,27,46,47 one used a five-point verbal rating scale for satisfaction,40 and one used a numeric rating scale (NRS) of 0-10 for discomfort.44 After converting the results to the same direction (higher score indicates more satisfied) and the same scale (10-cm VAS), no significant difference was detected for patient satisfaction scores at 24 hours (six trials; 353 patients; WMD, 0.05; 95% CI, −0.5 to 0.6) and at 48 hr postoperatively (four trials; 217 patients; WMD, 0.02; 95% CI, −1.1 to 1.1) (Table 1).

Postoperative nausea and vomiting

Ketamine added to morphine/hydromorphone significantly reduced PONV in 30 trials involving 2,143 patients (RR, 0.71; 95% CI, 0.60 to 0.85; RD, −8.9%; 95% CI, −4.6 to −12.2) (Tables 1 & 2, Fig. 5) (GRADE: moderate confidence in estimates, Table 2).

Fig. 5
figure 5

Combination of ketamine and morphine/hydromorphone patient-controlled analgesia (PCA) vs morphine/hydromorphone PCA: postoperative nausea and vomiting

Other adverse events

No significant differences were found for other adverse events, including hallucinations (22 trials; 70 events; 1,488 patients; RR, 1.27; 95% CI, 0.81 to 1.98; RD, 0.9%; 95% CI, −0.6 to 3.1) (GRADE: moderate confidence in estimates, Table 2), vivid dreams (14 trials; 62 events; 734 patients; RR, 1.21; 95% CI, 0.77 to 1.90; RD, 2.3%; 95% CI, −2.5 to 9.9) (GRADE: moderate confidence in estimates, Table 2), dysphoria (15 trials; 42 events; 882 patients; RR, 1.00; 95% CI, 0.55 to 1.84), pruritus (15 trials; 185 events; 1,287 patients; RR, 0.92; 95% CI, 0.69 to 1.22), respiratory depression (12 trials; 87 events; 1,030 patients; RR, 0.59; 95% CI, 0.30 to 1.17), urinary retention (eight trials; 86 events; 549 patients; RR, 0.76; 95% CI, 0.53 to 1.09), and diplopia (three trials; 18 events; 260 patients; RR, 1.53; 95% CI, 0.59 to 3.96), and cardiovascular adverse effects, including arrhythmia, hypotension, hypertension, or bradycardia (two trials; six events; 120 patients; RR, 1.51; 95% CI, 0.14 to 16.28).

Subgroup analyses and sensitivity analyses

Meta-regression did not detect a significant dose relationship between ketamine doses and pain scores (P value ranged from 0.44 to 0.92). Also, meta-regression did not detect a significant association between the ratio of ketamine-to-morphine and pain scores (P value ranged from 0.08 to 0.35).

No significant subgroup effects were detected between morphine PCA and hydromorphone PCA for pain at rest and movement, total opioid consumption (in morphine equivalent), patient satisfaction, and PONV (interaction P value ranged from 0.10 to 0.93) (Table 3); however, only three of the 36 included studies evaluated ketamine plus hydromorphone PCA.

Table 3 Subgroup analysis of effect of adding ketamine to morphine PCA and hydromorphone PCA

In addition, other predefined subgroup analyses did not find significant interactions between pain and different subgroups, including type of surgery, mode and duration of ketamine administered, use of nitrous oxide and other postoperative analgesics, risk of bias, and language of publication (interaction P value ranged from 0.11 to 0.98) (Supplementary Table 3).

Sensitivity analyses suggested that the effects of pain intensity and cumulative morphine consumption were robust across our approaches to imputing data for missing means and SDs and across our assumptions about pain score measured at rest or mobilization. Nevertheless, the confidence interval using the clearly reported data of mean and SD or pain scores at rest and mobilization was wider due to the smaller sample size (Supplementary Table 4). Also, the result was robust after removing the imputed PONV from the largest number of events of nausea and vomiting.

Publication bias

Publication bias was not detected for any outcomes (Supplementary Figs. 4, 5, 6) with the exception of PONV (Egger’s test P = 0.001).


Main findings

Ketamine added to morphine or hydromorphone PCA resulted in small reductions (< 1 cm pain reduction on a 10-cm VAS) in postoperative pain compared with morphine or hydromorphone PCA. These reductions were achieved despite lower morphine requirements for those receiving ketamine (reductions in cumulative morphine consumption during postoperative day 1 or day 3 of 5-20 mg).

Relationship with prior reviews

This meta-analysis adds significantly to previous systematic reviews of ketamine for acute postoperative pain since we found a number of new studies not previously incorporated. In addition, this meta-analysis quantifies the reduction of postoperative pain, cumulative morphine consumption, and PONV, which advances knowledge compared with the qualitative systematic review of 11 trials (n = 887) on the same topic.3 While our findings are consistent with two previous systematic reviews regarding a reduction of total opioid consumption and PONV, it is important to point out that both prior reviews compared ketamine given at any point (preemptively, intraoperatively, postoperatively) and administered by any routes (intravenous, intramuscular, or epidural administration) or by any method of intravenous administration (bolus, infusion, PCA), and they did not specifically address the addition of ketamine to morphine or hydromorphone PCA.2,5 There are two older systematic reviews that did address the addition of ketamine to morphine PCA, but they included only a small proportion of the studies available today and did not evaluate all clinically relevant outcomes explored in this current meta-analysis.4,6

Besides the additional pain reduction and morphine sparing effect, adding ketamine to morphine PCA also reduced PONV by an absolute risk reduction of 8.9% (95% CI, 4.6 to 12.2), which equates to a NNT of 11 (95% CI, 8 to 20) (GRADE: moderate confidence). No significant differences were found for other adverse events, including neuropsychiatric adverse events that have been attributed to ketamine at higher doses (GRADE: moderate confidence); however, very few trials reported on adverse events, and there remains a significant risk of underreporting of adverse events. Whether the reduction of morphine-related adverse events, e.g., PONV, is due to the decreased morphine consumption after adding ketamine remains uncertain. Nevertheless, this meta-analysis lends support for the relationship between reduced opioid leading to reduced PONV41,45 but without sufficient reports from the included studies regarding other opioid-related adverse events.

Since the complexity of drug administration increases when drugs are used in combination, these additional tradeoffs should be considered when deciding if this combination of drugs is worthy of the potentially small benefits, and the remaining unexplored risks, compared with morphine or hydromorphone PCA alone.

Strengths and limitations

This review is strengthened by the inclusion of comprehensive searches of six databases without limits by language or publication status. This resulted in identifying 22 RCTs not included in previous systematic reviews.2-6 We used all data available, made reasonable assumptions about unclear or missing data, and conducted sensitivity analyses that suggested our results are robust to these assumptions. We analyzed the effect of additional ketamine on pain both at rest and during mobilization and standardized the pain score measurements. We applied GRADE, a transparent method for rating confidence in estimates (quality of evidence) widely endorsed by the international systematic review and practice guideline community,11,12 to provide context for interpreting the findings.

There are some limitations to our systematic review. Substantial statistical heterogeneity was observed among pain outcomes, likely as a result of clinical heterogeneity among the patients (e.g., age, type of illness, type of surgery), and interventions (dose, route, duration of ketamine, ratio of ketamine vs morphine, anesthesia, postoperative analgesia, and other co-interventions), though subgroup analysis examining some of these variables failed to show clear relationships. The variability in results across studies was the primary reason for rating confidence in effects on pain as moderate rather than high. Also, we did not perform sensitivity analysis to address the impact of loss to follow-up, although the proportion of loss to follow-up ranged from 3-9%. Meta-regression failed to detect an association between total dose of ketamine or ratio of ketamine-to-morphine and effect sizes; thus, the optimal dose or ratio of ketamine-to-opioid still remains unclear. Finally, this meta-analysis could not address whether adding ketamine to morphine or hydromorphone reduces chronic pain since the included studies did not evaluate longer-term outcomes.

In this meta-analysis, most patients represented a low-risk population as indicated by low VAS scores in the control group, suggesting they were well controlled on morphine or hydromorphone PCA. Potentially, patients with higher VAS scores at baseline would be more likely to benefit from adjunctive ketamine; however, this meta-analysis did not provide sufficient data for us to test this hypothesis.

Significant differences were not detected for most of the opioid-related or ketamine-related adverse events, with the exception of PONV. The confidence intervals remain wide for some of these adverse events (e.g., respiratory depression, pruritus, urinary retention, diplopia, cardiovascular adverse effects), since only a few trials reported on these incidents. Therefore, the existing evidence base remains underpowered to rule out important differences that may exist for these underreported adverse events.

Implications for practice and summary

This meta-analysis of randomized trials provides objective evidence that adding ketamine to morphine or hydromorphone PCA provides a small improvement in postoperative analgesia while reducing morphine requirements in patients receiving morphine or hydromorphone PCA. Adjunctive ketamine also reduces PONV without a detected increase in the risk of neuropsychiatric effects. Nevertheless, the risk of adverse events was difficult to quantify since studies rarely reported on adverse events. Future research should explore the optimal ratio of ketamine-to-morphine and whether higher risk patients would reap more benefit (i.e., opioid-tolerant patients or patients with a high baseline pain score).