Postoperative pain treatment after total knee arthroplasty: A systematic review

Introduction The aim of this systematic review was to document efficacy, safety and quality of evidence of analgesic interventions after total knee arthroplasty (TKA). Methods This PRISMA-compliant and PROSPERO-registered review includes all-language randomized controlled trials of medication-based analgesic interventions after TKA. Bias was evaluated according to Cochrane methodology. Outcomes were opioid consumption (primary), pain scores at rest and during mobilization, adverse events, and length of stay. Interventions investigated in three or more trials were meta-analysed. Outcomes were evaluated using forest plots, Grading of Recommendations Assessment, Development and Evaluation (GRADE), L’Abbe Plots and trial sequential analysis. Results The included 113 trials, investigating 37 different analgesic interventions, were characterized by unclear/high risk of bias, low assay sensitivity and considerable differences in pain assessment tools, basic analgesic regimens, and reporting of adverse events. In meta-analyses single and continuous femoral nerve block (FNB), intrathecal morphine, local infiltration analgesia, intraarticular injection of local anaesthetics, non-steroidal anti-inflammatory drugs, and gabapentinoids demonstrated significant analgesic effects. The 24-hour morphine-sparing effects ranged from 4.2 mg (CI: 1.3, 7.2; intraarticular local anaesthetics), to 16.6 mg (CI: 11.2, 22; single FNB). Pain relieving effects at rest at 6 hours ranged from 4 mm (CI: -10, 2; gabapentinoids), to 19 mm (CI: 8, 31; single FNB), and at 24 hours from 3 mm (CI: -2, 8; gabapentinoids), to 16 mm (CI: 8, 23; continuous FNB). GRADE-rated quality of evidence was generally low. Conclusion A low quality of evidence, small sample sizes and heterogeneity of trial designs prohibit designation of an optimal procedure-specific analgesic regimen after TKA.


Introduction
It is well documented that treatment of postoperative pain often remains insufficient, and there is no international consensus about the optimal analgesic intervention after most surgical procedures, including total hip arthroplasty (THA). 4,5,15,27,60he primary goal of contemporary postoperative pain treatment is to reduce pain during rest and mobilization, and, if relevant, to reduce opioid consumption and thereby opioid-related adverse effects.It is further conceived that optimal pain management may facilitate early mobilization, improve postoperative outcome, and reduce the length of hospital stay (LOS). 16,44,45ere is general agreement that medical interventions must be "evidence-based," that is, rely on results from well-conducted randomized controlled trials (RCTs) and meta-analyses of the highest up-to-date scientific quality.It has also been suggested that postoperative pain treatment should be "procedure-specific," because the efficacy of a particular intervention may vary depending on the type of surgical procedure. 41The aim of this systematic review was to investigate the evidence for analgesic effects of procedure-specific medication-based interventions after THA.

Methods
This review is structured according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses-statement. 64 The review was registered in the PROSPERO international prospective register of systematic reviews on April 23, 2014 (registration number: CRD42014009382), with amendments added on September 22, 2014.

Literature search
Trials were sought in PubMed, Embase, and The Cochrane Library according to Appendix 1 (available online as Supplemental Digital Content at http://links.lww.com/PAIN/A2).The last search date was August 22, 2014.Furthermore, the PROSPECT database 12 and reference lists of relevant reviews were screened for eligible clinical trials.

Missing data
Authors were contacted by e-mail in the occurrence of unclear bias domains (see bias assessment) or missing information in primary outcomes mentioned in the RCTs "Methods" section.In the case of unclear bias domains due to inadequate descriptions, corresponding authors were asked to elaborate on the parameters performed to secure low risk of bias.To prevent false confirmation of suggested measures, we used open questions such as "Please describe what measures were taken to secure allocation concealment."

Bias assessment
Bias assessment was performed using the 7-piece Cochrane bias assessment tool, 36 including random sequence allocation, allocation concealment, blinding of participants and personnel, blinding of outcome assessors, incomplete outcome data, selective outcome reporting, and other potential threats to validity (including conflict of interest).Each domain was rated as low, high, or unclear, and the summarized risk of bias was considered low when all domains were rated as low; high when at least 1 domain was rated as high; and unclear when at least 1 domain was rated as unclear.In addition, we evaluated trial sample size as an independent contributor to bias.Based on Dechartres et al., 19 we defined a cumulated trial sample size of ,50 patients as high risk of bias, 50 to 199 as moderate risk of bias, 200 to 499 as lower risk of bias, and .499as low risk of bias.

Handling of continuous data
Continuous data for pain and opioid consumption for subgroups of 3 or more trials were analyzed using Review Manager 5. 39 Any opioid consumption was converted to the equivalent dose of i.v.morphine according to Appendix 2 (http://links.lww.com/PAIN/A2).Pain scores, eg, VAS 0 to 10 and verbal numerical rating scale 0 to 10, were converted to a 0 to 100 scale.Results expressed as median, and interquartile range (IQR) or range were converted to mean and SD according to The Cochrane Handbook 7.7.3.5 35 or 37 as appropriate.When results were presented without SD or IQR, SDs were calculated from the P value according to The Cochrane Handbook 7.7.3.3 35 ; if the P value was expressed as P , 0.05, the conservative approach P 5 0.05 was used.In trials comprising more intervention groups, these were either merged if comparable, or the control group was split if the tested interventions were different, according to The Cochrane Handbook 7.7.a. 35epending on heterogeneity, a random or fixed-effects model forest plot was performed for continuous data in the presence of 3 or more trials considering comparable interventions, with a 95% confidence interval (CI) mean difference.
Risk ratio (RR) was calculated for dichotomous data in the presence of interventions of 3 or more trials, with a 95% CI.In both dichotomous and continuous data, P values of less than 0.05 were considered statistically significant.

Heterogeneity
Heterogeneity for morphine consumption and pain score at rest was illustrated in L'Abb é plots for each group of intervention. 78urthermore, heterogeneity was tested using x 2 test with significance level set at a P value of 0.10, and the heterogeneity was assessed by I 2 , which quantifies inconsistencies, and D 2 for information size adjustments.Fixed-effect or random-effect was used according to the calculated I 2 .

Strength of evidence
The risk of false-positive (type I error) and false-negative results (type II error) can derive from low information size (sum of included patients) and repeated significance testing in metaanalyses.However, conducting a trial sequential analysis (TSA) can reduce this risk. 86Trial sequential analysis integrates the heterogeneity and the number of theoretically conducted interim analyses in the calculation of threshold for statistical significance and of a priori estimated information size (APIS).For standard significance testing, the Z-score is stationary at 1.96 for P 5 0.05.In TSA, a low information size is penalized by increasing the Z-score that is, more prominent results are required to qualify as significant.A priori estimated information size is the information size that is required to consider a Z-score of 1.96 statistical significant.The difference between APIS and the current information size is the number of patients who need to be included in further trials before definitive confirmation or disproval of the intervention effect can be made.
We performed TSA for opioid consumption and pain score for each intervention that demonstrated a statistically significant difference and consisted of at least 3 RCTs.We used TSA Viewer 0.9 Beta The Copenhagen Trial Unit and followed guidelines made by The Copenhagen Trial Unit (an alpha value of 0.05 and a beta value of 0.9). 83Sensitivity to detect a mean difference was set to 10 mg morphine i.v.equivalents per 24 hours for opioid consumption and 15 mm on a VAS 0 to 100 mm scale for pain scores. 46,84

Summary of findings
To assess the quality of evidence, we used The Grading of Recommendations Assessment, Development, and Evaluation (GRADE).For each outcome, 5 separate factors were rated for quality: study limitations, inconsistent results, indirectness of evidence, imprecision, and publication bias. 34Outcome results and quality of evidence were summarized for comparison according to GRADE using GRADEpro 3.6.

Retrieved trials
Search on PubMed, Embase, and The Cochrane Library identified 3886, 4497, and 2012 citations, respectively.An initial screening by the first author removed 2066 duplicates, and further 4768 citations that clearly had no relation to the present review (eg, thrombosis prophylaxis, cemented vs cementless procedures, and ossification risk).The remaining 3561 citations were assessed on title and abstract by 2 authors individually.Two hundred forty-two trials were included for fulltext download, with 24 of these being in a language other than English.Two hundred twenty-two trials were successfully acquired, and 164 were subsequently excluded (Appendix 3, http://links.lww.com/PAIN/A2).
Thus, 58 RCTs concerning medication-based postoperative pain management after THA were included for final analysis (Fig. 1).We managed to comprise the included trials into 19 subgroups of different treatment interventions.The most frequently investigated interventions were various non-steroidal anti-inflammatory drugs (including COX2 inhibitors) (10 trials), local infiltration analgesia (LIA) (11 trials), intrathecal administration of various opioids (7 trials), and lumbar plexus block (4 trials).Eighteen trials had 2 separate interventions and 2 trials had 3 separate interventions.Baseline variables from the included trials are summarized in Table 1.

Risk of bias in included trials
Bias assessment demonstrated that 45 trials contained at least 1 unclear domain (a total of 116 unclear domains), and we attempted to contact the corresponding authors by e-mail.In 3 trials, e-mail addresses were not retrievable and in 8, e-mail addresses were out of use or full.Fifteen authors answered regarding 29 unclear domains, of which 17 were resolved (2 high and 15 low).
The summarized risk of bias was low in 10 trials, unclear in 11, and high in 37 (Fig. 2).Furthermore, the trial sample size implicated a high risk of bias in 17 trials, moderate in 40, and lower in 1.

Supplemental and basic analgesics
The majority (37 trials) administered patient-controlled analgesia with i.v.morphine to supplement the active intervention and placebo under investigation, and reported a cumulated 24-hour consumption, whereas the remaining 22 trials used i.v.fentanyl, oxycodone, hydromorphone, meperidine, piritramide, ketobemidone, or epidural fentanyl, ropivacaine, or mepivacaine.Fifty-seven trials reported a cumulated opioid consumption over a time interval from 4 to 48        In 33 trials, supplemental opioid on request was the only analgesic administered in conjunction with the intervention/ control.In 5 trials, an NSAID was administered as basic analgesic regimen in addition to supplemental opioid.Likewise, 5 trials administered paracetamol, 7 trials a combination of NSAID and paracetamol, and 3 trials a combination of NSAID, paracetamol, and gabapentin as basic analgesia.Five trials administered other basic analgesic regimens (Table 1).

Pain scores
Twenty-four trials reported pain score as VAS 0 to 100; 20 trials as VAS 0 to 10; and 11 trials as numerical rating scale 0 to 10 (NRS 0-10), NRS 0 to 100, verbal numerical rating scale 0-10, boxscale 0 to 10, categorical scale 0 to 3, or visual scale 0 to 6 (Appendix 4, http://links.lww.com/PAIN/A2).When converted to VAS 0 to 100 mm equivalents (not categorical scale 0-3 or visual scale 0-6), values at rest and during mobilization ranged from 4 to 67 mm and 3 to 74 mm in the control groups, respectively.The mean postoperative pain scores in control groups at 6 and 24 hours rest were 31 mm and 23 mm, respectively.Pain at rest was reported in 42 trials at 6 hours and in 47 trials at 24 hours postoperatively (Appendix 4, http://links.lww.com/PAIN/A2).
Only 12 and 16 trials reported pain during movement at 6 hours and 24 hours, respectively.Ten trials reported pain during movement at both 6 and 24 hours; thus, 18 trials reported pain during movement at any time (Appendix 4, http://links.lww.com/PAIN/A2).Because of infrequent registration, meta-analyses were not feasible for pain during movement.

Other endpoints
Forty-two trials reported vomiting, nausea, or both, 14 reported sedation, 5 reported dizziness, and 11 reported pruritus (Appendix 4, http://links.lww.com/PAIN/A2).Eleven trials reported LOS, and of these, 7 had clearly predefined discharge criteria.No trial before 2007 reported LOS, and no intervention demonstrated a statistically significant reduction in LOS.
Twenty-one trials had low assay sensitivity for pain score (ie, no scores above 30 mm in control groups).Four trials had low assay sensitivity for morphine consumption (ie, morphine consumption below 15 mg i.v.morphine equivalents/24 h in control groups).

Non-steroidal anti-inflammatory drugs
Ten trials tested an NSAID (including COX2 inhibitors) as an intervention. 1,8,18,28,40,49,53,57,74,75One trial tested the intervention in conjunction with a basic analgesic regimen. 40he risk of bias was low in 1 trial, unclear in 1, and high in 8 (Fig. 2), and the trial sample size implicated a high risk of bias in 4 trials and a moderate risk in 6. L'Abb é plots demonstrated a high degree of heterogeneity for morphine consumption and a lower degree for pain scores (Appendix 5-7, http://links.lww.com/PAIN/A2).
Meta-analyses demonstrated a statistically significant morphinesparing effect of 14.1 mg/24 h (95% CI: 8.0-20.2mg) and a statistically significant reduction in pain scores at rest of 14 (10-17) and 9 (2-16) mm at 6 and 24 hours postoperatively, respectively (Figs. 3-5).In TSA, reductions in both morphine consumption, and 6-and 24-hour pain scores at rest, were above the threshold for significance.A priori estimated information size was reached for 6-hour pain score indicating that enough patients have been included to draw conclusions about the early analgesic effect (Appendix 8, http://links.lww.com/PAIN/A2).Meta-analyses demonstrated no significant effect on opioid-related adverse events, or adverse events related to the interventions per se (Appendix 9, http://links.lww.com/PAIN/A2).
Quality of evidence (GRADE) was low for the opioid-sparing effect and reduction in 6-hour pain score, and it was very low for 24-hour pain score.Results are summarized in Table 2.
Trials were too heterogeneous (time of administration, specific drugs, oral/i.v.administration) to provide information about optimal drug and dose regimens.
Two, 6, and 3 trials were of low, unclear, or high risk of bias, respectively (Fig. 2), and trial sample sizes implicated a high risk of bias in 1 trial and a moderate risk in 10.L'Abb é plots demonstrated a high degree of heterogeneity for both morphine consumption and pain scores (Appendix 5-7, http://links.lww.com/PAIN/A2).
Meta-analyses demonstrated a statistically significant morphine-sparing effect of 7.5 (3.7-11.3)mg/24 h, and insignificant reductions in pain scores at all time points (Figs.3-5).In TSA, reduction in morphine consumption was above the threshold for significance, and pain scores were below the threshold.A priori estimated information size was reached for morphine consumption, indicating that enough patients have been included to draw conclusions about the opioid-sparing effect.A priori estimated information size was reached for the absent effect on 24-hour pain score indicating that further testing of this endpoint is futile (Appendix 10, http://links.lww.com/PAIN/A2).Reduced postoperative blood loss was demonstrated in one trial (P , 0.05) 70 ; however, no differences in opioid-related adverse effects were demonstrated (Appendix 9, http://links.lww.com/PAIN/A2).
Quality of evidence (GRADE) was low for the opioid-sparing and pain-relieving effects.Results are summarized in Table 2.
It should be noted that different combinations of drugs were administered for LIA: in addition to local anesthetics, 2 trials included ketorolac, 2,77 2 trials morphine and corticosteroids, 50,70 and 1 trial morphine and ketorolac. 7None of these trials were controlled for the systemic analgesic effect that may have occurred after administration in the LIA combinations.
Trials were too heterogeneous to provide information about optimal drug and dose regimens.
One and 6 trials were of unclear and high risk of bias, respectively (Fig. 2), and the trial sample size implicated a high risk of bias in 3 trials and a moderate risk in 4. L'Abb é plots demonstrated a lower degree of heterogeneity for both morphine   Meta-analysis demonstrated a statistically significant morphine-sparing effect of 19.8 (14.9-24.7)mg/24 h and insignificant reductions in pain scores at all time points (Figs.3-5).
In TSA, reduction in morphine consumption was above the threshold for significance, and pain scores were below the threshold for significance (Appendix 11, http://links.lww.com/PAIN/A2).A priori estimated information size was not reached for morphine consumption.An increase in RR of 6.6 (2.7-16.4) was demonstrated for pruritus (Appendix 9).
Quality of evidence (GRADE) was very low for the opioidsparing effect and reduction in pain scores and was low for increases in PONV and pruritus.Results are summarized in Table 2. Trials were too heterogeneous to provide information about optimal drug and dose regimens.
All 4 trials were of high risk of bias (Fig. 2), and the trial sample size implicated a high risk of bias in 1 trial, a moderate risk in 2, and a lower risk in one.L'Abb é plots demonstrated a high degree of heterogeneity for both morphine consumption and pain scores (Appendix 5-7, http://links.lww.com/PAIN/A2).Meta-analyses demonstrated a statistically significant morphine-sparing effect of 11.9 (6.4-17.3)mg/24 h and a statistically significant 11 (3-19)  mm reduction in pain score at rest at 24 hours postoperatively (Figs.3-5).
In TSA, reductions in morphine consumption and pain score at 24 hours were above the threshold for significance; however, APIS were not reached, indicating that inclusion of more patients is necessary to draw conclusions about the extent of the analgesic effect (Appendix 12, http://links.lww.com/PAIN/A2).A reduction in RR for PONV of 0.19 (0.1-0.35) and a reduction in pruritus of (RR 0.13 [0.02-0.98])were demonstrated (Appendix 9, http://links.lww.com/PAIN/A2).One trial demonstrated a partial or moderate motor block in the majority of patients. 55Quality of evidence (GRADE) was very low for both the opioid-sparing and pain-relieving effects, and was low for PONV and pruritus.Results are summarized in Table 2. Too few trials were available to provide information about optimal drug and dose regimens.
All interventions except intrathecal clonidine, femoral nerve block, i.v.lidocaine, and gabapentin demonstrated some effect on supplemental analgesic consumption and/or pain scores.Three trials demonstrated an effect on opioid-related adverse events (reduction in nausea in patients treated with dexamethasone 43 and reduced pruritus for both femoral nerve block and epidural ketamine [P , 0.05]. 25,55) Qualitative data are reported in Table 3.

Discussion
We identified 4 major groups of procedure-specific analgesic interventions for pain treatment after THA: non-steroidal antiinflammatory drugs including COX2 inhibitors, LIA, intrathecal opioids, and lumbar plexus block.Furthermore, trials reporting 15 other interventions were retrieved.
These 4 groups of most reported interventions, except LIA, all demonstrated a morphine-sparing effect of at least 10 mg in forest plots, and all were confirmed in TSA.To reach APIS, however, TSA suggest a need for further inclusion of 652, 209, and 194 patients in the NSAID, intrathecal opioid, and lumbar plexus block-intervention groups, respectively.LIA demonstrated a 7.5 mg reduction in morphine consumption that was confirmed in TSA, and APIS was reached.This reduction may however be clinically questionable.Furthermore, significant reductions in pain scores were demonstrated in TSA with a sensitivity of 15 mm for NSAID at 6 and 24 hours postoperatively.The overall evidence is, however, flawed by a number of serious limitations.Thus, the summarized risk of bias was high in 37 trials, unclear in 11, and low in only 10 trials.A substantial number of trials included small sample sizes.The observed high or unclear summarized risk of bias, together with small sample sizes, may have overestimated intervention effects.
The retrieved trials demonstrated substantial differences in tools and criteria for pain assessment, as well as pronounced heterogeneity in primary outcomes.Pain scores in control groups ranged from 3 to 74 mm on a 0 to 100 mm VAS-score, and 0-to 24-hour morphine consumption ranged from 5 to 63 mg.Even in control groups that received identical basic analgesic treatment, substantial heterogeneity was displayed for both opioid   ⊕⊕⊖⊖ LOW † § CI, confidence interval; NSAID, Non-steroidal anti-inflammatory drugs; OR, odds ratio; PONV, postoperative nausea and vomiting; RR, risk ratio; THA, total hip arthroplasty.GRADE Working Group grades of evidence: (1) High quality: we are very confident that the true effect lies close to that of the estimate of the effect.(2) Moderate quality: we are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.(3) Low quality: confidence in the effect estimate is limited; The true effect may be substantially different from the estimate of the effect.( 4) Very low quality: we have very little confidence in the effect estimate; The true effect is likely to be substantially different from the estimate of effect.* The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).† There were studies of unclear and high summarized risk of bias.‡ There was heterogeneity as noted by I 2 .§ There were fewer than 400 participants in total.

Outcomes
January 2015    consumption and pain levels.Forty trials in the present review did not assess pain during mobilization, which is important because 1 of the goals of postoperative pain management is early and optimized mobilization. 80n 27 of the retrieved trials, the active intervention was studied as an adjunct to a basic analgesic regimen that often included multiple analgesics.Applying a basic analgesic regimen may reduce the maximal attainable morphine-sparing effect of an additional intervention.In accordance, pain scores or morphine consumption in control groups (assay sensitivity) were low in a number of trials (VAS below 30 mm 14 or i.v.morphine consumption ,15 mg/24 h).In addition, these trials have an increased risk of reporting significant but irrelevant results. 23n consistence with previous reports, 76 irregular and nonsystematic reporting of opioid-and intervention-specific adverse events in the majority of included trials was demonstrated.Only 11 trials assessed LOS, and none of these demonstrated significant differences between treatments.The quality of evidence ranged from low to very low throughout the analyses, according to GRADE.
In summary, there is some evidence that a number of analgesic interventions, including NSAID or COX2 inhibitors, LIA, intrathecal opioids, and lumbar plexus block, have the capacity to reduce mean opioid requirements or mean pain intensity compared with controls after THA.The evidence is, however, flawed by considerable bias, heterogeneity in outcome parameters, and irregular and non-systematic reporting of adverse events, leading to very low to low quality of evidence, according to GRADE.
There are several formal limitations to this review.Thus, we were unable to retrieve 20 possibly relevant articles.When authors of retrieved articles were contacted, they often failed to reply to our questions regarding bias, and these trials were consequently rated as "high risk" although the actual quality may have been higher than that stated in our bias assessment.
In the original articles, different opioids were administered as rescue analgesics, and pain was registered on different pain scales.Conversion to morphine equivalents and VAS 0 to 100 mm may have caused inaccurate results.
Opioid consumption was often reported as median and IQR, because of a skewed distribution. 66To perform meta-analyses, we had to convert such values to mean and SD, which is a potential and serious, but necessary shortcoming of our analyses.
Finally, mean differences between intervention groups as an outcome have been criticized because pain relief often displays a bimodal distribution of either very good or poor. 65An alternative measure of analgesic effect is dichotomous reporting, where the reduction in pain score in individual patients must reach a (minimal) clinical relevant level before being considered effective."No worse than mild pain" has been suggested as a simple universal outcome. 67e did not, however, have access to individual patient data.
A contemporary systematic review with meta-analyses of RCTs may provide the best available evidence for effects of different analgesics in relation to a specific surgical procedure; furthermore, it may offer an impression of how well or "bad" the literature works in general.Traditional RCTs of specific interventions may demonstrate reductions in mean pain scores and opioid requirements from one level to another.These may not, however, be the most relevant clinical questions, and consequently, there are important limitations to this and other similar reviews of the current "procedure-specific" evidence.
The final goal of effective pain treatment is to ensure the lowest pain intensity during rest and ambulation and the lowest need for supplemental opioids in a majority of patients.Moreover, to ensure minimal adverse effects and optimized rehabilitation.These challenges are most probably not solved by (reviews of) the current literature.Thus, a majority of the published RCTs may have used a well-defined surgical procedure as a model to investigate the efficacy of a particular, analgesic intervention, rather than investigating for the best possible procedure-specific, multimodal treatment option.Accordingly, we doubt that future traditional RCTs will ever tell us how to provide an effective multifaceted postoperative pain treatment.We suggest that future efforts to improve analgesia after THA should include large-scaled randomized trials with factorial designs, of combinations of different, well-documented nonopioid analgesics and local anesthetic techniques. 17Outcomes should include the number of patients with predefined "acceptable" pain levels at rest and during ambulation, 67 opioid consumption, opioidrelated and intervention-specific adverse effects, and LOS adhering to specific goals within well-defined rehabilitation paradigms. 16n conclusion, this review of postoperative pain management after THA is, to the best of our knowledge, the first to include alllanguage trials, and to take both bias (including trial size), TSA and GRADE, into account.It has been documented that some analgesic interventions may have the capacity to reduce mean opioid requirements or mean pain intensity compared with controls.However, the quality of evidence is low or very low for the different regimens, and the available literature does not allow a designation of a "best proven intervention" for this surgical procedure.

10 A
.P. Højer Karlsen et al. • 156 (2015) 8-30 PAIN ® Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Figure 2 .
Figure 2. Risk of bias in included studies.Green plus is low risk, yellow question mark is unclear risk, and red minus is high risk of bias.

18 A
.P. Højer Karlsen et al. • 156 (2015) 8-30 PAIN ® Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Figure 3 .
Figure 3. Forest plot displaying mean difference in 0-to 24-hour morphine consumption for each major intervention.Green squares with horizontal lines depict mean differences and 95% confidence intervals (CIs) for each trial.Black tiles depict the mean difference of each intervention.

Figure 4 .
Figure 4. Forest plot displaying the mean difference in pain scores 6-hour postoperatively for each major intervention.Green squares with horizontal lines depict mean differences and 95% confidence intervals (CIs) for each trial.Black tiles depict the mean difference of each intervention.

Figure 5 .
Figure 5. Forest plot displaying mean difference in pain scores 24-hour postoperatively for each major intervention.Green squares with horizontal lines depict mean differences and 95% confidence intervals (CIs) for each trial.Black tiles depict the mean difference of each intervention.

Table 1 (
continued) (continued on next page) January 2015 • Volume 156 • Number 1 www.painjournalonline.com13 Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Table 1 (
continued) ® Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Table 1 (
continued) com Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Table 1 (
continued) postoperatively.In 8 trials, other types of opioid consumptions than morphine were successfully converted to equivalent i.v.morphine doses.Values ranged from 4.1 to 63 mg/24 h with a mean of 31 mg/24 h in the control groups, based on 45 studies.
® Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.hours

Table 3
Qualitative analysis of other interventions.

Table 3 (
continued) Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.

Table 3 (
continued) ® Copyright Ó 2014 by the International Association for the Study of Pain.Unauthorized reproduction of this article is prohibited.