Postoperative pain treatment after total knee arthroplasty: A systematic review

1The primary goals of postoperative analgesic treatment are to reduce pain, opioid requirements and consequently opioid-related adverse events, in order to optimize rehabilitation. Enhancing these outcomes has potential beneficial influence on patient morbidity and satisfaction, the degree of required postoperative care, as well as economic perspectives. Total knee arthroplasty (TKA) is a frequently performed orthopedic procedure followed by moderate to severe pain. Therefore, an efficient postoperative analgesic treatment based on sound evidence from the published literature is important for this procedure. Recent research on postoperative pain after total hip arthroplasty suggest, however, that it may be difficult to allow a designation of a “best proven intervention” from the available scientific evidence, and it is reasonable to believe that this applies for TKA as well.

2The hypothesis of this review was, that no globally recognized, best proven, gold standard analgesic treatment or intervention exists for TKA. The aim of this systematic review of all randomized, controlled clinical trials (RCTs) considering postoperative pain treatment after TKA is therefore to document the evidence for postoperative analgesic interventions after TKA.

3The review meets requirements of the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) statement. Registration in the PROSPERO International prospective register of systematic reviews was completed on April 23, 2014, prior to initiation of the study (registration number: CRD42014014940). Updated searches were carried out on June 17, 2016, and September 19, 2016, and registered in the protocol as amendments.

4Our methods are similar to those reported in a recent review of postoperative pain treatment after total hip arthroplasty (THA) published by our research group. As the two reviews are associated the methods and results sections are reported in a similar way to secure uniformity.

5Trials were sought in Pubmed, Embase and The Cochrane Library according to S1 Appendix. The last search date was September 9, 2016. The PROSPECT database and reference lists were screened for eligible trials as well.

6Inclusion criteria were randomized controlled trials of unilateral total knee arthroplasty that compared postoperative analgesic outcomes of a perioperative analgesic intervention against placebo in a control group. Basic analgesic regimens and rescue analgesics had to be administered under equal conditions in the intervention and control groups. Trials where different rescue analgesics were administered, e.g. morphine and acetaminophen p.n., were included for qualitative analyses, but not meta-analyses. We only included trials with interventions initiated in the immediate perioperative period that reported either opioid-sparing effect, pain at rest or pain during mobilization. Trials concerning knee fractures, trials including patients less than 18 years, and data published in summary clinical trials, editorials, letters, and comments were excluded.

7The primary outcome was 0–24 hours postoperative cumulated opioid consumption.

8Secondary outcomes were pain both at rest and on mobilization at 6 and 24 hours postoperatively, opioid related and intervention associated adverse events, and length of hospital stay (LOS).

9We extracted the following data: Trial sample size; basic analgesic regimen (i.e. analgesics administered to both intervention- and control group as a fixed regimen); rescue analgesics and 24 hour cumulated dose; pain score at rest and during movement at 6 ± 2 hours and 24 ± 4 hours postoperatively; opioid-related adverse events (postoperative nausea or vomiting (PONV), sedation, dizziness, pruritus, urinary retention, constipation and respiratory depression); intervention-associated adverse events as reported; LOS; and documented and predefined discharge criteria.

10Assay sensitivity (a trials ability to detect an absolute difference between groups if there is one) was deemed low if a control group demonstrated a pain score on a visual analogue scale (VAS 0–100 mm) below 30 mm and/or a 0–24 hour cumulated i.v. morphine consumption below 15 mg.

11Data extraction and bias evaluation was carried out by two authors independently. Disagreements were solved during meetings with all authors.

12For trials with unclear bias domains or missing information regarding primary outcomes, the corresponding author was contacted by email and if unresponsive, another inquiry was sent two weeks later. We used open questions as "Please describe all measures taken to secure random sequence allocation" to avoid false confirmation on suggested measures.

13We used the Cochrane bias assessment tool to evaluate the following domains: 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). Domains were rated as low, high, or unclear risk of bias. If all domains were low the summarized risk of bias was rated low; if one or more domains were high the summarized risk was rated high; and if one or more domains were unclear with no high risk domains, the summarized risk was rated unclear.

14In addition, we evaluated trial sample size as a contributor to bias. A cumulated trial sample size of < 50 patients was rated as high risk of bias, 50–199 as moderate risk of bias, 200–499 as low risk of bias, and > 499 as very low risk of bias based on Dechartres et al..

15Meta-analyses were carried out in Review Manager 5® whenever three or more trials regarding a specific intervention reported a 0–24 hour opioid consumption. Opioids were converted to i.v. morphine equivalents according to S2 Appendix. Pain scores, side effects and LOS were analyzed when reported in three ore more trials. Visual analogue scale (VAS 0–10) and Numerical Rating Scale 0–10 (NRS 0–10), were converted to VAS 0–100. Median and interquartile range (IQR)/range was converted to mean and standard deviation according to The Cochrane Handbook 7.7.3.5, or Hozo et al, as appropriate. For results presented only as mean, a standard deviation was calculated from the p-value according to The Cochrane Handbook 7.7.3.3, and we used the conservative approach p = 0.05 if the p-value was expressed as p < 0.05. Some trials had more than one intervention group. In these cases we either merged intervention groups or split the control group, according to The Cochrane Handbook 7.7.a.

16Forest plots were calculated with a 95% confidence interval (CI) mean difference for continuous data and risk ratio (RR) with a 95% CI for dichotomous data. Random effects model was used whenever I^2 was above 30%. For I^2 between 0 and 30% fixed and random effects models were compared and the most conservative approach (the model with the widest 95% CI) was used to take into account the heterogeneity of included trials. P-values of less than 0.05 were considered statistically significant.

17L’Abbé plots were conducted for each meta-analysed intervention to describe the degree of heterogeneity for morphine consumption and pain scores.

18In meta-analyses, low information size (number of patients included) and repeated significance testing increase the risk of type I and II errors (false positive and false negative results, respectively). This risk can be reduced by performing trial sequential analysis (TSA). A forest plot describes whether the tested intervention reaches significance through the classic p<0.05, whereas TSA accounts for interim analyses and the heterogeneity of the trials as well. In TSA, the normal stationary threshold for significance with a Z-score at 1.96 for p = 0.05 is penalized if the included trials demonstrates a high degree of heterogeneity. An intervention with a high degree of heterogeneity requires a higher information size to reach the threshold for significance compared to a forest plot analysis. This is calculated as the a priori estimated information size (APIS).

19TSA was performed for morphine consumption and pain scores, for all interventions that were included in meta-analyses.

20We used Trial Sequential Analysis Viewer 0.9 Beta (The Copenhagen Trial Unit (CTU)) and followed the CTU guidelines (an alpha-value of 0.05 and a beta-value of 0.9). The sensitivity to detect a mean difference was set to 10 mg i.v. morphine equivalents/24 hours and 15 mm on a VAS 0–100 mm scale.

21Quality of evidence was assessed with The Grading of Recommendations Assessment, Development and Evaluation (GRADE). Five factors were evaluated for each outcome: Study limitations; publication bias; indirectness of evidence; inconsistent results; and imprecision (evaluation based on results in TSA).

22Outcome effects and quality of evidence were summarized according to GRADE using GRADEpro 3.6.

23Search on Pubmed, EMBASE and The Cochrane Library identified 5126, 5806 and 2646 citations, respectively. The first author removed 4952 duplicates. Two authors assessed the remaining 8626 citations individually, compared results and consequently 287 trials were downloaded in full-text, of which 22 were written in a non-English language. We managed to acquire 285 trials of which 172 met one or more exclusion criteria (S3 Appendix).

24Thus, 113 randomized placebo-controlled trials concerning postoperative analgesic interventions after TKA were included for review (Fig 1 PRISMA flowchart). The total number of patients was 8407.

25The included trials comprised 37 different treatment interventions. Interventions that qualified for meta-analyses, were single injection femoral nerve block (FNB), continuous FNB, intrathecal morphine, local infiltration analgesia (LIA), intraarticular injection with local anaesthetics, non-steroidal anti-inflammatory drugs (NSAIDs)/COX-2-inhibitors, and gabapentinoids.

26Of all trials 36, 10, 3, and 1 had two, three, four and five separate intervention groups, respectively.

27The follow-up period in the included trials was: 1 day in 20 trials, 2 days in 36 trials, 3 days in 16 trials, 4–7 days in 8 trials, ≥2 weeks in 22 trials, and unclear in 11 trials.

28Detailed study information from the included trials is summarized in Table 1.

29105 trials contained at least one unclear domain (a total of 350 unclear domains). We contacted the corresponding authors by email. Email addresses were either irretrievable or permanently out of use in 22 trials. Corresponding authors for the remaining 83 trials were contacted. Forty authors replied regarding 119 unclear domains and 74 were resolved (5 high and 69 low). Forty-four domains remained unclear.

30The summarized risk of bias was low in 18 trials, unclear in 65 and high in 30 (Fig 2). Further, the trial sample size bias was high in 41 trials, moderate in 69, and low or lower in three.

31Sixty trials administered i.v. morphine patient-controlled analgesia (PCA) as rescue medication, and reported a 0–24 hours cumulated consumption, while the remaining 53 trials administered i.v./i.m. fentanyl, oxycodone, hydromorphone, meperidine, papaveretum (a mixture of morphine, papaverine, and codeine), sufentanil or NSAIDs; patient-controlled continuous FNB; or epidural local anaesthetics/opioids. Eighty-nine trials reported cumulated opioid consumption over 20–72 hours postoperatively, seven of these also administered a second non-opioid rescue analgesic. In five trials included in meta-analyses, other types of opioids were converted to i.v. morphine equivalents. Postoperative 0–24 hours morphine consumption in the control groups for trials included in the meta-analyses ranged from 5.5–116 mg with a corrected mean of 33.1 mg per patient.

32A supplemental opioid with no underlying basic analgesic regimen, was administered in 37 trials. Sixty-three trials administered a basic analgesic regimen in addition to supplemental rescue analgesics; seven trials administered acetaminophen, 13 trials NSAIDs, 12 trials acetaminophen + NSAID, seven trials local injection + other analgesics, 15 trials nerve blocks + other analgesics, and 11 trials administered different combinations of analgesics (Table 1).

33Pain score was reported as VAS 0–100 in 42 trials; as VAS 0–10 in 52 trials; and as either numerical rating scale 0–10 (NRS 0–10), WOMAC pain scale 0–10, or verbal pain scale (VPS) 0–3 in 18 trials (S4 Appendix). After conversion to VAS 0–100 mm equivalents values in control groups ranged from 0–80 mm and 0–82 mm at rest and during mobilization, respectively. Mean pain scores in control groups for trials included in the meta-analyses were 38 mm at 6 hours rest, 33 mm at 24 hours rest, 50 mm at 6 hour movement, and 53 mm at 24 hours movement.

34Pain scores at rest at 6 hours postoperatively were reported in 84 trials, and at 24 hours postoperatively in 89 trials. Pain during mobilization was reported in 33 trials at 6 hours postoperatively, and in 42 trials at 24 hours postoperatively (S4 Appendix).

35Ninety trials reported PONV, 24 sedation, 16 dizziness, and 43 pruritus (S4 Appendix).

36LOS was reported in 36 trials of which 15 described clearly predefined discharge criteria. No trials before 2001 reported LOS. Of the 36 trials six demonstrated a statistically significant reduction in LOS.

37Nineteen trials demonstrated low assay sensitivity for pain score (i.e. pain scores below 30 mm in control groups at 6 or 24h postoperatively). Thirteen trials demonstrated low assay sensitivity for morphine consumption (i.e. no morphine consumption above 15 mg i.v. morphine equivalents 0–24 hours postoperatively in control groups).

38Seven meta-analyses were carried out. Forest plots for primary and secondary outcomes are presented in Figs 3–5 and S5–S11 Appendices, L’Abbé plots and TSA are presented in S12–S25 Appendices.

39Fig 6 presents a summary of all the meta-analysed subgroups regarding outcomes, GRADE-rated quality of evidence and the estimated risk of bias of the included trials.

40Fifteen trials tested single FNB as an intervention. Four of these trials tested the intervention in addition to a basic analgesic regimen.

41The summarized risk of bias was low in zero, unclear in 10, and high in five trials (Fig 2), and the trial sample size implicated a high risk of bias in seven trials, and a moderate risk in eight trials. L’Abbé plots demonstrated a lower degree of heterogeneity for pain score at rest and moderate degrees for morphine consumption and pain during movement (S12 Appendix).

42Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 16.6 mg (95% CI: 11 to 22; p<0.00001) (Fig 3), and a reduction in postoperative pain scores at 6 hours at rest of 19 mm (8 to 31; P = 0.0007), at 24 hours at rest of 12 mm (5 to 19; P = 0.001), and at 24 hours during movement of 9 mm (-2 to 20; P = 0.09) (Figs 4 and 5, S6 Appendix).

43In TSA, reductions in both morphine consumption and pain scores at rest at 6 and 24 hours were above the threshold for significance. Morphine consumption and 24 hours pain score at rest reached APIS concluding that single FNB has a positive effect on these outcomes (S13 Appendix).

44In meta-analyses, RR for nausea and vomiting was 0.66 (0.51 to 0.85; P = 0.002), for dizziness 0.38 (0.12 to 1.19; P = 0.1) and for pruritus 0.94 (0.51 to 1.75; P = 0.85) (S7, S9 and S10 Appendices).

45Urinary retention was registered in four trials, deep venous thrombosis (DVT) in two, soreness/pain in the back in two, hypotension in one, numbness around the knee in one, and infection around the site of injection in one. No significant differences between active and control groups were reported.

46Quality of evidence (GRADE) was moderate for PONV; low for the opioid sparing effect and pain scores and pruritus; and very low for dizziness. Results are summarized in Table 2.

47Various local anaesthetics +/- epinephrine were administered in the trials (Table 1). The evidence did not provide information about optimal drug-, and dose-regimens.

48Ten trials tested continuous FNB as an intervention. Six of these trials tested the intervention in addition to a basic analgesic regimen.

49The summarized risk of bias was low in zero, unclear in two, and high in eight trials (Fig 2), and the trial sample size implicated a high risk of bias in four trials and a moderate risk in six. L’Abbé plots demonstrated homogeneity for morphine consumption and pain scores (S14 Appendix).

50Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 12.3 mg (95% CI: 9.7 to 14.8; P<0.00001) (Fig 3), and a reduction in pain scores at rest at 6 hours postoperatively of 10 mm (2 to 19; P = 0.01), at 24 hours at rest of 16 mm (8 to 23; P<0.00001) and at 24 hours during movement of 10 mm (4 to 15; P = 0.0005) (Figs 4 and 5, S6 Appendix).

51In TSA, reductions in both morphine consumption and pain scores at rest at 6 hours and 24 hours, and pain scores during movement at 24 hours, were above the threshold for significance and reached APIS, concluding that continuous femoral nerve block has a positive effect on these outcomes (S15 Appendix).

52In meta-analyses, RR for nausea and vomiting was 0.74 (0.54 to 1.03, P = 0.07), for sedation 1.33 (0.13 to 13.66; P = 0.81) and for pruritus 1.13 (0.9 to 1.41; P = 0.31) (S7, S8 and S10 Appendices). One study demonstrated a significant increase in obturator motor blockade at 6 hours postoperatively. Urinary retention was registered in one trial, cardiac events in one and hypotension in two. No significant differences between active and control groups were reported.

53Quality of evidence (GRADE) was moderate for reduction in morphine consumption and pain score at 24 hours during movement; low for 24 hours pain score at rest, PONV and pruritus; and very low for 6 hours pain score at rest and sedation. Results are summarized in Table 3.

54Various local anaesthetics +/- epinephrine were administered in all trials (Table 1). The evidence did not allow designation of optimal drug-, and dose-regimens.

55Nine trials tested intrathecal morphine as an intervention. Four of these trials tested the intervention in addition to a basic analgesic regimen.

56The summarized risk of bias was low in zero, unclear in six, and high in three trials (Fig 2), and the trial sample size implicated a high risk of bias in two trials and a moderate risk in seven. L’Abbé plots demonstrated moderate degrees of heterogeneity for morphine consumption and pain scores (S16 Appendix).

57Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 9.8 mg (95% CI: 3.6 to 16.1, P = 0.002) (Fig 3), a reduction in pain scores at rest at 6 hours postoperatively of 15 mm (1 to 28, P = 0.04), and at 24 hours at rest of 8 mm (0 to 17; P = 0.05) (Figs 4 and 5).

58In TSA, morphine consumption reached the threshold for significance but not APIS. Pain score at 24 hours rest reached the boundary for futility and APIS concluding that there is no reason for further investigation of this outcome (S17 Appendix).

59In meta-analyses, RR for nausea and vomiting was 1.8 (1.28 to 2.54; P = 0.0008), for sedation 1.93 (0.18 to 20.24; P = 0.58) and for pruritus 5.74 (2.44 to 13.47; P<0.0001) (S7, S8 and S10 Appendices). Hypoxemia was registered in one trial, respiratory depression in two, urinary retention in three and anxiety in one. No significant differences between active and control groups were reported.

60Quality of evidence (GRADE) was high for the increase in pruritus; moderate for increase in PONV; low for opioid sparing effect and reduction in pain score at 6 and 24 hours at rest; and very low for the increase in sedation. Results are summarized in Table 4.

61Diamorphine was administered in one trial and morphine in the others. Due to heterogeneity amongst trials there were no dose-response relationship and the evidence did not provide information regarding optimal dosages.

62Eighteen trials tested LIA as an intervention. Thirteen of these trials tested the intervention in addition to a basic analgesic regimen.

63The summarized risk of bias was low in one, unclear in 12, and high in five trials (Fig 2), and the trial sample size implicated a high risk of bias in six trials, a moderate risk in 11 and a low risk in one. L’Abbé plots demonstrated low degrees of heterogeneity for morphine consumption and pain scores at rest. Moderate degrees were present for pain scores during movement (S18 Appendix).

64Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 13.4 mg (95% CI: 8.5 to 18.2; P<0.00001) (Fig 3), and a reduction in pain scores at rest at 6 hours postoperatively of 14 mm (9 to 20; P<0.00001), at 24 hours rest of 10 mm (6 to 13; P<0.00001), at 6 hours during movement of 16 mm (9 to 23; P<0.00001) and at 24 hours during movement of 14 mm (8 to 20; P<0.00001) (Figs 4 and 5, S5 and S6 Appendices).

65In TSA, threshold for significance and APIS were reached for all outcomes, concluding that LIA has a positive effect on these outcomes (S19 Appendix).

66In meta-analyses RR for nausea and vomiting was 0.68 (0.54 to 0.86; P = 0.0009) and for pruritus 0.78 (0.45 to 1.33; P = 0.36) (S7 and S10 Appendices). One study demonstrated a significant reduction in blood loss and one demonstrated a significant increase in skin blisters due to cannula. Hypotension was registered in one trial, respiratory distress/depression in two, headache in one, positive cultures from the catheter tips in one, rash in one, urinary retention in seven, DVT in four, incision complications in three, cardiac or CNS events in two, slight numbness in one and constipation in one. No significant differences between active and control groups were reported.

67Quality of evidence (GRADE) was moderate for PONV; low for the opioid sparing effect and reduction in pain scores, and very low for pruritus. Results are summarized in Table 5.

68Trials were too heterogeneous (administration of different combinations of local anaesthetics, morphine, NSAIDs, steroids and epinephrine, Table 1) to provide information about optimal drug-, and dose-regimens.

69Seven trials tested intraarticular injection of local anaesthetics as an intervention. Two of these trials tested the intervention in addition to a basic analgesic regimen.

70The summarized risk of bias was low in one, unclear in five, and high in one trial (Fig 2), and the trial sample size implicated a high risk of bias in two trials and a moderate risk in five. L’Abbé plots demonstrated homogeneity for morphine consumption and pain scores at rest at 6 hours and higher degrees of heterogeneity at 24 hours (S20 Appendix).

71Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 4.2 mg (95% CI: 1.3 to 7.2; P = 0.004) (Fig 3), and a reduction in pain scores at rest at 6 hours postoperatively of 10 mm (4 to 17; 0.001) and at 24 hours at rest of 3 mm (-8 to 14; P = 0.57) (Figs 4 and 5).

72In TSA, reductions in both morphine consumption and pain scores at rest at 6 hours were above the threshold for significance and reached APIS concluding that intraarticular injection has a positive effect on these outcomes (S21 Appendix).

73In meta-analyses RR for nausea and vomiting was 1.18 (0.51 to 2.74; P = 0.70) (S7 Appendix). Respiratory depression was registered in two trials, sinus tachycardia in one, DVT in one, wound healing complications in one. No significant differences between active and control groups were reported.

74Quality of evidence (GRADE) was moderate for opioid sparing effect and 6 hours pain score; low for increase in PONV; and very low for 24 hours pain score. Results are summarized in Table 6.

75Trials were too heterogeneous (administration of different combinations of local anaesthetics, morphine, steroids and epinephrine, Table 1) to provide information about optimal drug-, and dose-regimens.

76Ten trials tested NSAIDs/COX-2-inhibitors as an intervention. Two of these trials tested the intervention in addition to a basic analgesic regimen.

77The summarized risk of bias was low in zero, unclear in six, and high in four trials (Fig 2) and the trial sample size implicated a high risk of bias in two trials, a moderate risk in seven and a very low risk in one. L’Abbé plots demonstrated low degrees of heterogeneity for morphine consumption and moderate degrees for pain scores (S22 Appendix).

78Meta-analyses demonstrated a statistically significant 0–24 hour postoperative morphine sparing effect of 6 mg (95% CI: 3.2 to 8.7; P<0.0001) (Fig 3), and a reduction in pain scores at rest at 6 hours postoperatively of 7 mm (1 to 14; P = 0.02), at 24 hours at rest of 5 mm (3 to 8; P<0.0001), and at 24 hours during movement of 3 mm (-4 to 10; P = 0.41) (Figs 4 and 5, S6 Appendix).

79In TSA, threshold for significance and APIS were reached for morphine consumption and pain at 6 and 24 hours rest concluding that NSAIDs and COX-2-inhibitors have a positive effect on these outcomes. The reduction in pain scores during movement at 24 hours reached the threshold for futility and APIS (S23 Appendix).

80In meta-analyses RR for nausea and vomiting was 0.91 (0.61 to 1.35 P = 0.63) and for pruritus 0.91 (0.39 to 2.14; P = 0.83) (S7 and S10 Appendices).

81Bleeding was registered in one trial, hypo/hypertension in one, anemia in two, urinary retention in three, dry mouth in one, gastric pain in one and constipation, hyperhidrosis, pyrexia, headache and confusion in one. No significant differences between active and control groups were reported.

82Quality of evidence (GRADE) was low for the opioid sparing effect and reduction in pain score at 6 and 24 hours at rest, and very low for remaining outcomes. Results are summarized in Table 7.

83Trials were too heterogeneous (time of administration, specific drugs, oral/i.v. administration) to provide information about optimal drug-, and dose-regimens.

84Seven trials tested gabapentinoids as an intervention. Six of these trials tested the intervention in addition to a basic analgesic regimen.

85The summarized risk of bias was low in three, unclear in two, and high in two trials (Fig 2) and the trial sample size implicated a high risk of bias in two trials, a moderate risk in four and a low risk in one. L’Abbé plots demonstrated moderate degrees heterogeneity for morphine consumption and pain scores (S24 Appendix).

86Meta-analyses demonstrated non-significant reductions for 0–24 hour postoperative morphine sparing effect of 8.5 mg (95% CI: -3.3 to 20.3; P = 0.16) (Fig 3), and pain scores at rest at 6 hours postoperatively of 4 mm (-1 to 10; P = 0.15) and at 24 hours at rest of 3 mm (-2 to 8; P = 0.19). Significant reductions in pain scores at 6 and 24 hours during movement of 8 mm (2 to 14; P = 0.01) and 4 mm (0 to 8; P = 0.04), respectively, were demonstrated (Figs 4 and 5, S5 and S6 Appendices).

87In TSA, threshold for significance and APIS were reached for pain at 6 and 24 hours during movement concluding that gabapentinoids have a positive effect on these outcomes. Threshold for futility and APIS were reached for pain at rest at 6 and 24 hours concluding that further testing of these outcomes is futile (S25 Appendix).

88In meta-analyses RR for nausea and vomiting was 0.83 (0.65 to 1.07, P = 0.15), for sedation 1.17 (0.83 to 1.63, P = 0.37), for dizziness 0.68 (0.3 to 1.53, P = 0.35) and for pruritus 0.3 (0.15 to 0.59; P = 0.0006) (S7–S10 Appendices).

89One study reported an accumulation of undesirable reactions due to the study drug in the intervention groups; lapse of memory function, impaired balance, hypotension, diplopia, sedation, dizziness and fatigue.

90Quality of evidence (GRADE) was moderate for pain score at 6 hours at movement; low for pain scores at 24 hours at rest and during movement, PONV, sedation and pruritus; and very low for opioid sparing effect, reduction in pain score at 6 hours at rest and dizziness. Results are summarized in Table 8.

91Trials were too heterogeneous (time of administration and specific drugs) to provide information about optimal drug-, and dose-regimens.

92Forty-one trials investigated other interventions: Adductor canal block; clonidine, tramadol, fentanyl, magnesium or buprenorphine added to FNB; single and continuous sciatic plexus nerve block; obturator nerve block; continuous lumbar plexus block; morphine, ketamine or epinephrine added to intraarticular injections of local anaesthetics; continuous intraarticular injection of local anaesthetics; ketorolac added to periarticular injection of local anaesthetic; steroids added to LIA; epidural analgesia with ropivacaine and morphine; magnesium, midazolam or morphine added to epidural bupivacaine/ropivacaine; fentanyl patch; i.v. tramadol; i.v. nefopam; i.v. ketamine; p.o. nimodipine; i.v. dexmedetomidine; i.v. duloxetine; i.v. methyl-prednisolone; and i.v. magnesium.

93Twenty-four interventions were administered together with a basic analgesic regimen (S4 Appendix).

94The risk of bias was low in 12 trials, unclear in 25, and high in four.

95Nine trials did not demonstrate a significant effect on opioid consumption and/or pain scores: Clonidine, tramadol or buprenorphine added to FNB; continuous lumbar plexus block; morphine, ketamine or epinephrine added to intraarticular local anaesthetics; betamethasone added to periarticular local anaesthetics; and i.v. magnesium. The remaining trials demonstrated statistically significant analgesic effects. Four trials demonstrated a statistically significant effect on opioid-related adverse events: Sciatic nerve block a reduction in PONV, dexmedetomidine a reduction in PONV and pruritus, and epidural analgesia with ropivacaine and morphine an increase in pruritus (Table 9).

96We have reviewed randomized controlled trials regarding postoperative analgesia after TKA, and have demonstrated analgesic effects in meta-analyses for single injection- and continuous FNB, intrathecal morphine, LIA, intraarticular injection of local anaesthetics, NSAIDs/COX-2 inhibitors, and gabapentinoids; and furthermore in stand-alone trials for a number of different interventions, according to the PRISMA checklist (S26 Appendix). By conducting meta-analyses we have enhanced the evidence to the highest level possible with the present trials. While this sounds promising, the quality of evidence throughout the included data is discouragingly low due to uncertain or high risk of bias, low sample size in trials and meta-analysis interventions, heterogeneous results, and low assay sensitivity. These findings are similar to the results in our recent systematic review on pain management after THA. Consequently, we have demonstrated that no optimal strategy for postoperative pain treatment after TKA exist in the literature.

97The accepted level of pain varies in the analyzed material. In some trials no basic analgesic regimens were provided and high pain scores were accepted, whereas in other trials acetaminophen, NSAIDs, gabapentin, and even FNB were administered as adjuncts to the intervention. In these trials both intervention- and control groups tended to have lower pain scores. These differences lead to a considerable variance in assay sensitivity amongst trials. The wide variation in trial-setup may be accounted for by cultural or tradition based differences in the approach to analgesic treatment, e.g. the propensity to apply invasive procedures or the general pain threshold.

98For oral treatments we analyzed two subgroups: NSAIDs/COX-2-inhibitors and gabapentinoids.

99The included trials in the NSAID subgroup were characterized by low assay sensitivity, which contributes to a low absolute effect. However, the intervention provided a small but statistically significant effect on pain scores and morphine consumption. The adverse event profile of NSAIDs is controversial and short follow up periods in randomized pain trials in general may be problematic for the detection hereof (17). However, the meta-analyses did not demonstrate an increased risk of adverse effects which is supported by similar results in the review regarding THA.

100The evidence regarding gabapentinoids was even less convincing, with insignificant results partially due to a low number of included trials.

101Four meta-analysed interventions investigated procedure specific local anaesthetic interventions: single FNB, continuous FNB, LIA, and intraarticular injection of local anaesthetics. When reviewing the outcomes, intraarticular injection tended to be inferior compared to the other interventions.

102Single FNB performed slightly better in two out of three primary outcomes compared to continuous FNB. The strength of evidence in TSA was generally high for both interventions. Continuous FNB is a more invasive, time consuming and for the patient cumbersome procedure due to the postoperative catheterization.

103Single FNB and LIA provided equally satisfying analgesia. Both procedures demonstrated a relevant reduction in morphine consumption, pain scores and PONV. A recent systematic review of trials comparing LIA to FNB after TKA reported a small insignificant difference in analgesic effect favoring LIA. The current evidence does not allow designation of a superior intervention amongst the two, but the well-known risk of motor blockade with FNB may render this method less attractive. It should be noted that different combinations of drugs and dosages were administered for both FNB and LIA. Pinpointing optimal analgesics regimens for FNB and LIA are imperative for designation of a superior intervention.

104The meta-analysis of intrathecal morphine demonstrated some analgesic effect on morphine consumption and pain scores, but a rather large increase in morphine related adverse events.

105The large amounts of data in this review were manually typed with more than 12.000 separate boxes in Excel®, creating a major potential for typing errors. To minimize this risk data were analyzed and registered by two independent authors with prior data extraction experience, and subsequently compared.

106In a considerable number of trials data were presented as medians and range/IQR, likely because of a skewed distribution. Treating data as normally distributed by converting to mean/SD was necessary, but nonetheless a limitation.

107About half of the corresponding authors replied our emails regarding bias. This resolved 74 unclear domains and altered the total number of trials with low summarized risk of bias from six to 15. This is still only 13% of all trials. For trials included in meta-analyses this proportion was 7%, which is problematic as the quality of the meta-analyses is partially limited by the quality of the trials. The majority of trials had an unclear summarized risk of bias. We believe that in most of these trials, relatively few and easily attainable measures would be required to improve this risk from unclear to low, especially if authors had access to a standardized postoperative pain trial protocol that took into account the pitfalls leading to high or unclear risk of bias.

108Opioid consumption and pain intensity are associated, hence both outcomes require assessment. Whether opioid consumption and pain should be calculated as absolute or relative differences between treatment and control groups, or as the number of patients with a predefined level of pain, is controversial. In this review we chose to report effects as absolute (mean) differences, which may be arguable.

109The majority of included interventions each provided acceptable levels of analgesia, however it is probably reasonable to keep postoperative analgesic treatment to a limited number of interventions. Each additional intervention added to the standard postoperative analgesic regimen may increase the risk of adverse effects or events. Regarding invasive procedures we must consider the risk of inducing severe adverse effects and the time consumption by qualified personnel such as doctors. Furthermore, we know little about the effect of combining different analgesic interventions after TKA. Thus, the absolute analgesic effect may decline for each additional analgesic, because different interventions may affect the same analgesic pathways, and because the analgesic potential is probably lower when pain levels are already reduced by other analgesics.

110In conclusion, no gold treatment for pain treatment after TKA exists in the literature. The GRADE rated recommendations varied from very low to moderate (except for one high) for the different interventions. High or unclear risk of bias, heterogeneity of trial designs, and the small trial sample sizes, are challenges in designation of a best proven optimal postoperative analgesic regimen for TKA. A way to overcome these challenges may be to establish standard research guidelines regarding postoperative pain management, and focus on conducting high quality upscale trials.